Title of Invention	A MULTI/HYPER-SPECTRAL DATA ANALYZING PROCESS FOR COMPLETE QUANTIFICATION, CHARACTERIZATION AND COMPRESSION OF NATURAL RESOURCE SPECIFIC INFORMATION
Abstract	In remote sensing earth features are primarily characterized through multi-spectral signatures, recorded either as per cent reflectance or gray levels in different wavebands. However, in order to make characterization quantitative and more specific some spectral indices derived from information in these spectral channels/wavebands are often used, which compress the data partially in two or more selected wavebands. Data analysis of simple gray scale, color, and color-infrared images is fairly straightforward. Current techniques for analysis of Landsat-7 band images are adequate, but there are currently no methods for analysis of hyper-spectral data that are both powerful and fast. Current methods tend to either: 1) Revert hyper-spectral images to Landsat channels; 2) Rely on information from a few selected bands; or 3) Explore the entire spectrum through complex data analysis procedures such as Partial Least Squares (PLS), whose computational requirements increase with the square of the data"s dimension (i.e. number of spectral channels). In fact all these techniques are based on a simple assumption that some wavelengths or portions of the spectrum are rich in information about a feature of interest while the others are poor. Thus all these techniques totally ignore the fact that the spectrum as a whole has another dimension of information that is lost in treating it as discrete channels. Besides this, all these techniques involve complicated class-separability and clustering analysis in n-dimensional space; where "n" is the number of spectral channels. 1 developed a novel, powerful and fast hyper-spectral data analyzing method for quantifying information contained in the whole spectrum, with any number of data/spectral channels from 2 to infinity, of any earth feature based on the basic principles of communication theory. Application of this new hyper-spectral data analyzing method to multi-/ hyper-spectral databases from various platforms, such as field, aircraft & satellite imaging spectrometers has shown that the new method can lead to: 1) Easy identification of previously unrecognized systematic noise in the RDACS/H3 push broom hyper-spectral sensor; 2) Distinct characterization of edges of linear/ non-linear natural/man-made resources such as metallic roads, railway lines, canals, rivers, drains and water- bodies (A Good Edge Detection Tool); 3) Distinct characterization of and discrimination between vegetated areas, non- vegetated areas, natural resource mining sites, railway lines, water-bodies, rivers & its tributaries and drains/ canals & their distributaries; 4) Easy discrimination between structural and natural vegetation types thereby leading to more accurate estimates of areas under these vegetation types; 5) Distinct discrimination between soil systems with different physico-chemical characteristics; 6) Distinct characterization and discrimination of different moisture levels in soils; 7) Great reduction in data storage space requirement; and 8) Simplified 1-Dimensional clustering analysis.

Title of Invention

A MULTI/HYPER-SPECTRAL DATA ANALYZING PROCESS FOR COMPLETE QUANTIFICATION, CHARACTERIZATION AND COMPRESSION OF NATURAL RESOURCE SPECIFIC INFORMATION

Abstract

In remote sensing earth features are primarily characterized through multi-spectral signatures, recorded either as per cent reflectance or gray levels in different wavebands. However, in order to make characterization quantitative and more specific some spectral indices derived from information in these spectral channels/wavebands are often used, which compress the data partially in two or more selected wavebands. Data analysis of simple gray scale, color, and color-infrared images is fairly straightforward. Current techniques for analysis of Landsat-7 band images are adequate, but there are currently no methods for analysis of hyper-spectral data that are both powerful and fast. Current methods tend to either: 1) Revert hyper-spectral images to Landsat channels; 2) Rely on information from a few selected bands; or 3) Explore the entire spectrum through complex data analysis procedures such as Partial Least Squares (PLS), whose computational requirements increase with the square of the data"s dimension (i.e. number of spectral channels). In fact all these techniques are based on a simple assumption that some wavelengths or portions of the spectrum are rich in information about a feature of interest while the others are poor. Thus all these techniques totally ignore the fact that the spectrum as a whole has another dimension of information that is lost in treating it as discrete channels. Besides this, all these techniques involve complicated class-separability and clustering analysis in n-dimensional space; where "n" is the number of spectral channels. 1 developed a novel, powerful and fast hyper-spectral data analyzing method for quantifying information contained in the whole spectrum, with any number of data/spectral channels from 2 to infinity, of any earth feature based on the basic principles of communication theory. Application of this new hyper-spectral data analyzing method to multi-/ hyper-spectral databases from various platforms, such as field, aircraft & satellite imaging spectrometers has shown that the new method can lead to: 1) Easy identification of previously unrecognized systematic noise in the RDACS/H3 push broom hyper-spectral sensor; 2) Distinct characterization of edges of linear/ non-linear natural/man-made resources such as metallic roads, railway lines, canals, rivers, drains and water- bodies (A Good Edge Detection Tool); 3) Distinct characterization of and discrimination between vegetated areas, non- vegetated areas, natural resource mining sites, railway lines, water-bodies, rivers & its tributaries and drains/ canals & their distributaries; 4) Easy discrimination between structural and natural vegetation types thereby leading to more accurate estimates of areas under these vegetation types; 5) Distinct discrimination between soil systems with different physico-chemical characteristics; 6) Distinct characterization and discrimination of different moisture levels in soils; 7) Great reduction in data storage space requirement; and 8) Simplified 1-Dimensional clustering analysis.

Full Text	Resource managers require fast and accurate methods for acquiring and interpreting data on development and management of natural/ man-made resources. Information on natural resources like soils, crops, water, forest, land cover, topography etc and man-made resources like canals, drains, railway lines etc is essential to understand the potential of land in any given area and to maintain the growth and development of that area. Besides this, a continued exploitation of these resources makes it imperative to know them in the context of their conservation, preservation and reclamation. Remote sensing technique has proved to be of immense use in the study and monitoring of these earth resources. These remote sensing based resource surveys not only provide information in the areas of agriculture, forestry and grasslands for wasteland mapping; agricultural crop acreage and yield estimation, drought monitoring and assessment, land use/land cover mapping and forest productivity estimations but also in the areas of engineering and hydrology for flood monitoring and damage assessment, water resource management, ground water targeting, marine resource survey, urban planning, mineral targeting and environmental impact assessment etc. The important thing about these resource surveys is that they are fast, repetitive and temporo-spatially synoptic (i.e. cover large areas). Besides this, they are generally less expensive and less time consuming than the prevailing conventional field/ laboratory methods. As the name suggests, remote sensing is the science and art of obtaining information about an object, area or phenomenon through the analysis of data acquired by a device that is not in contact with the object, area or phenomenon under investigation. (Lillesand and Kieffer, 1979). A remote sensing system principally comprises of an energy source; an object (here, any earth resource/ feature) of interest; a sensor mounted on a platform; and the data generated from the interactions between the energy and the surface of the object of interest. The energy source, sun, emits Electro-Magnetic Radiation (EMR). These electro-magnetic radiations, falling on the surface of the earth, interact with the objects on the earth surface. After interaction some amount of energy is absorbed, some scattered and the rest is reflected back by the object. A sensor (mounted on a platform) senses this reflected energy that provides some information on the object. The amount of this energy absorbed, scattered and reflected by an object is in fact a function of the object's physical and chemical properties. Thus by recording reflected EMR through sensors, objects on the earth surface can be differentiated and their ground distribution can be mapped. The ratio of the EMR that is reflected from the surface of an object to that of the incident radiation is known as the spectral reflectance and the graph showing the spectral reflectance of an object at different wavebands is known as the spectral reflectance curve or spectral signature. In a remote sensing system, the amount of these reflected radiations (i.e. the data) acquired by the sensors (mounted on any platform, such as a tripod stand, a stepladder, an aircraft or a space station) is recorded in a digital form. In the case of the handheld field/ laboratory sensors, this is stored as a record comprising of an array of data elements corresponding to different spectral channels/wavebands per spectrum. While in the case of the aircraft/ satellite sensors, this is stored as a digital image comprising of a two-dimensional matrix of picture elements (commonly called pixels). Every such pixel is associated with an address in the matrix f (e.g. column & row number). Further each pixel has a third dimension, known as the gray value / brightness value/ intensity value/ radiometric value. This value in fact represents the reflectance characteristics of the corresponding object space elements (i.e. the ground objects falling within the spatial resolution of a picture element or a pixel) in different spectral channels/ wavebands. Existing Spectral Data Analyzing Techniques In remote sensing earth features are thus primarily characterized through multi-spectral signatures, recorded either as per cent reflectance or gray levels in different wavebands. However, in order to make characterization quantitative and more specific some spectral indices derived from information in these spectral channels/wavebands are often used which compress the data partially in two or more selected wavebands. When sensors collected light in a single channel, both the spectral data and its analysis were simple, though inadequate of the task demanded of them. Advances in the sensor technology over last decade have outstripped the ability of the existing methods to adequately analyze the spectral data. Color and color-infrared imagery brought new demands for analysis and the LANDSAT Thematic Mapper (TM) with its 7 channels and other airborne platforms with even more than 120 channels have increased the degree of complication associated with the spectral data and its analysis significantly. Analysis of spectral data from all these sources is of several types: • Analysis of individual bands, • Simple band ratioing (e.g. Near Infrared/Red) • Related normalized differences (of which the NDVI is a widely used measure), • Linear combinations of wavebands (e.g. Near Infrared - Red, Near Infrared + Red, Greenness index of Kauth and Thomas, 1976, etc.) • Principal Component Analysis (PCA, Cooley and Lohnes, 1985), and • Partial Least Squares (PLS - Modified PCA) analysis Shortcomings of Existing Spectral Data Analyzing Techniques Of the above, the first four spectral data analysis techniques can compress data in only 1-2 discrete spectral channels. It is well known by now that an individual waveband is related to only a few properties of the object under consideration, and not to the complete set of object characteristics. Thus all these spectral data analysis techniques are capable of only partial data compression/ quantification (i.e. in only 1-2 wavebands). Other linear band combination spectral indices such as the greenness index of Kauth and Thomas (1976), though capable of compressing the information contained in four Multi-Spectral Scanner (MSS) bands fully, do not lead to a meaningful interpretation of results as the number of spectral channels is further increased (Jackson, 1983). Thus none of these techniques can be used for hyper-spectral data analysis. The last two techniques, PCA and PLS, are n-dimensional spectral data analysis techniques as they can be used to compress/quantify information contained in even more than 4 spectral channels. The Principal Component Analysis (PCA) technique uses data from all the bands by multiplying the multi-channel image by an eigen-vector, giving different weights to different bands. The transformation itself is relatively rapid, but extensive analysis and computation goes into the development of the eigen-vector. (The actual computation of the eigen-vector is a slower operation (generally requiring many hours of processing time) with the computational efforts and time increasing with the square of the number of data channels. Partial Least Squares (PLS) is another modified form of PCA technique. While PLS has some advantages over PCA, the same computational work (if not more) is involved. Thus both these techniques require many hours of processing time for hyper-spectral data analysis. Besides this even these techniques are based on a simple assumption that some wavelengths or portions of the spectrum are rich in information about a feature of interest while the others are poor. Thus all these techniques totally ignore the fact that the spectrum as a whole has another dimension of information that is lost in treating it as discrete channels. Further as the number of spectral bands increases the dimensionality of state space in which an object is to be classified also increases; thereby leading to complicated class-separability and clustering analysis (Duda and Hart, 1973). Thus, besides other shortcomings, both these (PCA and PLS) techniques also involve complicated class-separability and clustering analysis in n-dimensions; where 'n' is the number of spectral channels. New Spectral Data Analyzing Process In view of the above shortcomings associated with the existing techniques, I developed a new and novel process for quantifying information contained in the whole spectrum of any earth feature. The new spectral data analyzing process is based on the basic principles of communication theory, developed for quantifying redundancy in speech or electrical signals for ensuring fail-safe communication (Shannon, 1948). I applied this theory for the first time to the multi-spectral data (from LISS-III sensor of IRS-1C/1D Satellites) and to the hyper-spectral data (from RDACS/H3 sensor of NASA Aircrafts and from handheld field/ laboratory spectro-radiometers) to develop a novel process for quantification of information contained in a spectrum of any earth feature to its resource characteristic value. For this essentially, I treated each (point's / pixel's) spectrum as a data stream. The number of data elements (or gray levels) in one such spectrum varied with the desired radiometric resolution. These data elements (or gray levels) were taken as spectral events. First, the number of occurrences of each spectral event (at a user specified radiometric resolution) in a measured spectrum was calculated. From these outcomes per spectral event, the probability of occurrence of each possible spectral event and the total number of outcomes per spectrum were determined for obtaining total spectral information content. Hence, this way each point/ pixel was assigned a single value based upon its total spectral information content. The investigations revealed that the new spectral data analyzing process could lead to distinct characterization and discrimination of various types of earth surfaces in a 1-D data space thereby leading to simplified class-separability and clustering analysis. The new process is simple, powerful, fast and can be used to process spectra with any number of data/spectral channels greater than 2. Accordingly, this invention relates to a spectral data analyzing process comprising of the following steps: i. acquiring a spectral data stream of any earth feature; ii. determining the minima and maxima of each such spectral data stream; iii. determining the desired radiometric resolution; iv. determining the number of possible data elements in a given spectral data stream, based on the desired radiometric resolution and the maxima and minima; v. determining the probability of occurrence of each possible spectral data element per spectral data stream; vi. determining the total number of outcomes of all possible data elements per spectral data stream; vii. determining the total information content, based on the probability of occurrence and total number of outcomes of all possible data elements, per spectral data stream; viii. checking for any negative total information content per spectral data stream, at the above desired radiometric resolution; ix. re-defining the desired radiometric resolution, in case of negative total information content per spectral data stream and repeating the steps from (iv) to (ix) till positive total information content per spectral data stream, at a particular re-defined radiometric resolution, is obtained. Application Potential of New Hyper-Spectral Data Analyzing Process Application to hyper-spectral data bases from laboratory/ field soectro-radiometers To validate the application potential of the new hyper-spectral data analyzing process, I applied it to a huge independent soil spectral database. This database comprised of spectral information on 1295 soil samples: 802 soil samples from 4 Major Land Resource Areas (MLRA), 273 samples from S220 field in Walnut Creek watershed, Iowa and 120 samples from Minnesota site in U.S. The MLRA group comprised of the surface (0-3 or 1-10 cm) and the sub-surface (3-10 or 10-30 cm) soil samples collected from 448 sites in the four Major Land Resource Areas (MLRAs). These included: MLRA-9 (the Palouse and Nez Perce Prairies located in eastern Washington and western Idaho); MLRA-67 (the Central High Plains located in eastern Colorado, southeastern Wyoming, and western Nebraska); MLRA-77 (the Southern High Plains in New Mexico and the panhandle of Texas); and MLRA-105 (the Northern Mississippi Valley-Drift less area, located in northeastern Iowa, southeastern Minnesota, and southwestern Wisconsin). The Iowa group on the other hand comprised of soil cores, ranging between 38 to 108 cm, from 10 sampling sites within five soil series, namely Canisteo, Nicollet, Harps, Webster and Clarions. Five soil cores were collected at each site. The average length for the soil cores was 78.4 cm. The soil from each core was divided into 15-cm increments for a total of 277 samples. These soils in the Iowa group were very diverse in texture and other properties. The soil samples from Minnesota site comprised of three Mollisols, one Alfisol, and one Ultisol. Waukegan silt loam (fine-silty over sandy or sandy-skeletal, mixed mesic Typic Hapludoll), Weld silt loam (fine, smectitic, mesic Aridic Argiustolls), and Palouse silt loam (fine-silty, mixed, superactive, mesic Pachic Haploxerolls) were collected from sites near Rosemont, MN, Akron, CO, and Pullman, WA, respectively. Miami silt loam (fine-loamy, mixed, active, mesic Oxyaquic Hapludalfs) was collected from a site near West Lafayette, IN while Cecil sandy loam (Fine, Kaolinitic, thermic Typic Kanhapludults) was from a site in Oconee County, GA. Except for the Waukegan silt loam, both surface (A-horizon) and subsurface (B-horizon) soil samples from each soil series were investigated. Each soil was maintained at six different levels of moisture, including oven-dried soils. These included soils maintained at RH=0, 32.3, 55, 72.6, and 98%. The matric potentials for RH=32.3, 55, 72.6, and 98% are equal to -470, -81, -43, and -2.7 MPa, respectively. The gravimetric moisture content of all samples was measured immediately after the spectral readings. Soil spectral measurements on these soil samples were recorded in the visible to near-infrared region (400-2500 nm) at 2 run intervals (1050 spectral channels) with a Perstorp NIR Systems 6500 scanning monochromator (Foss NIRSystems, Silver Spring, MD) - a laboratory spectrometer. The results of this analysis (Fig. 1) clearly show that the new hyper-spectral data analyzing process has a great potential to discriminate between widely varying soil systems or Major Land Resource Areas 9/105; 67 and 77 (Fig. 1). This could be majorly attributed to a strong relationship of their total spectral information content (H) with the major color contributing soil-parameters such as sand, silt, iron (Fe) and total organic carbon (Fig. 1). Other soil properties such as P, Mn, Mg and K (which do not have any affect on soil color) did not reveal any such relationship with the total spectral information content. The analysis clearly shows (Fig. 1) that the soils with higher total organic carbon (and thus total nitrogen), iron, zinc (Zn) and smaller particle sizes have lower total spectral information content (H) as compared to the soils with lower total organic carbon, Fe, Zn and higher particle sizes. Since the soils of the MLRA 77 contained lowest Fe and total organic carbon and highest sand and silt fractions therefore they were characterized with very high total spectral information contents. This was followed by the soils of MLRA 67, 9 and 105. It could further be observed that the soils of MLRA 9 and 105, having almost comparable soil- characteristics, were characterized with similar total spectral information contents. That the new process has a potential to discriminate between widely varying soil systems could be further confirmed from the analysis on the Iowa group of soils (Fig. 1). It can be seen from fig. 1 that a majority of Iowa soils, with total carbon values (ranging between 1-2 %) comparable with those for MLRA 67 and 77 soils, were quite distinct from the soils of MLRA 67 and 77; majorly because of their lower iron contents. The analysis thus clearly brought out the strength of the new hyper-spectral data analyzing process to discriminate between different soil systems. Besides this, it could also be seen from fig. 2, that the total spectral information content (H) has a strong positive relationship (R2 ranging from 0.89-0.98) with soil moisture between 0 to -2.7 M Pas matric potentials (i.e. between air-dry to field capacity), for most soil types. It could thus be inferred that the new hyper-spectral data analyzing process could even discriminate between the same soils with widely varying soil moisture levels (i.e. it has capability to discriminate between wet and dry soil types). Application to multi/ hyper-spectral databases from airborne platforms Seeing the application potential of the new spectral data analyzing process on the above laboratory spectrometer based soil spectral database, I further applied it to the multi/ hyper-spectral digital images acquired through satellites / aircraft sensors. The structure of these databases is completely different from those obtained through laboratory/ field spectrometers. Each pixel of the digital images can contain spectral information in any number of bands (ranging from 2 to infinity). The new process computes pixel-wise total spectral information content for these digital images, at a desired radiometric resolution. Application of the new process to a hyper-spectral digital image (with 1m spatial resolution and 120 bands ranging from 471-828nm at 3nm bandwidth) collected by Stennis Space Center in September 1997 using a RDACS/H3 hyper-spectral sensor showed that the so processed hyper-spectral image (Fig. 3) of the Iowa corn (still to be harvested in the south) and soybean (already harvested in the north) fields could distinctly characterize and discriminate between various land features viz. cropped areas, wet and dry bare fields, areas with varying levels of weedy growth, pot holes, and weedy and non-weedy roads. In traditional soil survey activities, soil profile studies take about 30% of the total time spent for fieldwork, whereas traversing for plotting soil boundaries takes around 70% time. Traversing is not only time consuming but also arduous element of soil survey. In fact in traditional system of soil survey this is directly related to the skill of soil surveyor and the traversing plan. Thus the use of the new hyper-spectral data analyzing process for such purposes would not only help in overcoming errors associated with subjectivity and drudgery in locating and plotting soil boundaries but would also help to minimize traversing to only a few well planned ground observations for validating soil boundaries and in generating soil maps of even inaccessible/ arduous areas. Besides, this since the new hyper-spectral data analyzing process-based image could distinctly discriminate between wet and dry fields. Therefore it can even serve as a robust tool for planning sowing operations. It further showed that the new method based processing of the test hyper-spectral image (840 pixels high x 462 pixels wide x 120 channels x 16 bits per channel) took just 30 seconds. Besides this, it was observed that the size of the primary hyper-spectral digital data, in 120 spectral channels/ wavebands, was 90,959-KB. While the size of the new process based hyper-spectral image (i.e. Fig. 3), derived from the above primary data, was just 1,908 KB. Thus, the new process based hyper-spectral image required just 2% of the storage space required by the primary image. Hence the new hyper-spectral data analyzing process, besides being fast, could even lead to a greater reduction in data storage space requirement Of more immediate interest, is the regular speckle like pattern overlaid upon the entire (new- process based) image (Fig. 3). This speckle can be readily removed from the new hyper-spectral method based image using First Fourier Transform (FFT). However, the point here is that the above speckle like pattern in the processed image revealed the capability of the new hyper-spectral data analyzing process to detect a previously unrecognized defect in the RDACS/H3 sensor (Chengye, NASA Personnel, personal communication). This defect causes a regular oscillation between the smoother and the noisier spectra, and results in slightly lower accuracy of predictions, with possible detection of non-existent patterns. To check the new process for any errors and its further application potential to the multi/ hyper-spectral satellite/ airborne data, I used it to analyze a 4 band IRS-1C (LISS III), 8-bit, multi-spectral digital data (of 2740 by 2360 pixels size) of Sohna watershed for 9th May, 1999. The above data was purchased from NRSA, Hyderabad. The processing of this huge multi-spectral digital image (with 2740 by 2360 pixels) took just 8-10 minutes (depending on the desired radiometric resolution). As it can be observed from fig. 4, the new process based image showed no "interference/ speckle like pattern", as observed in the case of the above processed hyper-spectral image in fig. 3. This once again confirmed: a. No errors in the new spectral data analyzing process; and b. The sensor defect-detection potential of the new process. A comparison of a portion of the new process based multi-spectral image (Fig. 4) with the same portion of the conventional FCC (Fig. 5) of the Sohna watershed further revealed that the new process based multi-spectral image could distinctly characterize and discriminate between the vegetated & non-vegetated surfaces, ridges, water-bodies, stone quarrying sites, orchards/ farm houses, roads, road side trees, railway line and Gurgaon canal. It can be seen from the zoomed part (Fig. 6a) of the fig. 4 above, that the new process based multi-spectral image could even distinctly discriminate between the Gurgaon canal and the metal road and the cart track. In fact, the edges of a canal, in the new process based image in fig. 6a, were sharper than those for a metallic road. Further, the non-metallic roads or cart tracks or pack-tracks did not show any sharp edges in the new process based image. This discrimination between a canal and the metallic / non-metallic roads or cart tracks was not possible with the conventional FCC-image in Fig. 6a. Besides this, the new process based multi-spectral image could even distinctly discriminate a metal road from a railway track (Fig. 6b). As it can be seen from the zoomed part (Fig. 6b) of fig. 4, the edges of a railway track are more sharp and bold than those for a metal road in the new process based multi-spectral image. This discrimination between a railway track and a metal road was not possible with the conventional FCC image in fig. 6b. Figures (6c; 6d; 6e and 6f) further compare a canal/ its distributaries; drain/ its distributaries; a river/ its tributaries and a water body (lake) in the conventional multi-spectral images with the new process based mutli-spectral images. It can be clearly seen from these images that the new process based image could even distinctly discriminate between these features. All this clearly shows that the new process has a great potential in many change detection regulatory programs for updating information on vegetation/ terrain conditions and in land use planning also. Besides this it can be seen from figs 6(a-f), the new process based multi-spectral images could distinctly characterize the edges of the various natural/ man-made (linear/ non-linear) features such as metal roads, railway lines, canals, rivers, drains and water bodies. This was not possible with the conventional FCC based images. Thus the new hyper-spectral data analyzing process was also found to be an excellent edge detection tool. I further analyzed another 4 band IRS-ID (LISS III), 8-bit multi-spectral digital data (with 2218 by 2024 pixels) for Sohna watershed for 05th February 2001 in order to test the potential of the new process to discriminate between agricultural and forest areas. The above data was also purchased from NRSA, Hyderabad. On comparing a portion the new method based image (Fig. 7b) with the same portion of the conventional False Colour Composite (FCC; Fig. 7a), it could be observed that the new process based image could distinctly characterize and discriminate between the cropped (i.e. agricultural) and the forested/ orchard (farmhouse) areas. As it can be seen from fig. 7b (see red encircled area in Fig. 7b), the cropped areas in the new process based image appeared as bright white patches while the forest/ orchard areas appeared as non-uniform or uniform grayish patches. This discrimination between the cropped and the forest areas was not possible with the conventional FCC (Fig. 7a), as here both these areas (see red encircled area in Fig. 7a) appeared as similar blue patches. To separate these similarly appearing blue coloured areas under crops and trees in the conventional FCC, one has to do an extensive ground truthing of the area. However with the new spectral data analyzing process this ground truthing could be either completely avoided (if the area is known) or greatly minimized (to only a few selected points of confusion). Thus the new process based processing of spectral data can even lead to much better (and more accurate) estimates of areas under crop and forest covers with no/ very minimal ground truthing. It was further found that although the size of the primary IRS 1-C (LISS III) and IRS-1D (LISS III) data in 3 bands was 18,955 and 13, 165 KB respectively; yet the size of the new process based multi-spectral images was just 6,316 and 4,387 KB, respectively. Thus it was once again confirmed that the new process leads to a great reduction in the data storage space requirement Hence it can be observed that the new process is the only simple and fast hyper-spectral data analyzing means for complete quantification of information contained in a spectrum, comprising of any number of spectral channels/ wavebands from 2 to infinity, of any natural surface. In fact Indian Space Research Organization has planned to launch a narrow band spectrometer comprising of about 200 wavebands (in visible and MR region) for satellite based remote inventory of natural resources. In this context this is the only hyper-spectral data analyzing process, which can be used straight away for any natural/ man made resource inventory (i.e. its characterization/ discrimination). Besides this, since the digital image generated through the new hyper-/multi-spectral data analyzing process is a single channel-image therefore it's unsupervised/ supervised classification would be based on a simplified 1-D clustering analysis techniques as compared to the more complicated 3-D clustering analysis techniques required for unsupervised/ supervised classification of a conventional FCC. Thus the new spectral data analyzing process can even lead to a simplified 1-D clustering analysis. CAPTIONS TO FIGURES/ DRAWINGS SHEET NO. 1: Fig. 1 Relationship of total spectral information content (H) with major soil colour contributing soil parameters. SHEET NO. 2: Fig. 2 Relationship of total spectral information content (H) with soil moisture between air-dry to field capacity. SHEET NO. 3: Fig. 3 New process-based image of Iowa fields. Note speckle pattern caused by noise in the hyper-spectral sensor. See that the cropped areas, moist and dry bare fields, areas with varying levels of weedy growth, potholes, and weedy and non-weedy roads are clearly distinguishable. SHEET NO. 4: Fig. 4 New process-based image of Sohna watershed, obtained through the processing of IRS-1C (LISS III) band- 2, 3 and 4 data for 9th May, 1999. SHEET NO. 5: Fig. 5 Conventional False Colour Composite (FCC) of Sohna watershed, derived from IRS-1C (LISS III) band- 2, 3 and 4 data for 9th May 1999. Notice, all vegetated areas are in light blue colour. SHEET NO. 6: Fig. 6a Comparison of metallic / non-metallic roads and canal features in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate between and show sharp edges of the Gurgaon canal and the metal roads. These edges were sharper for a canal than a metallic road. The cart tracks did not show any sharp edges in the new process based image. This discrimination between a canal, a metallic road and a cart track was not possible with the conventional FCC image. SHEET NO. 7: Fig. 6b Comparison of metallic road and railway track features in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate a metallic road and a railway track. This discrimination between a railway track and a metallic road was not possible with the conventional FCC image. SHEET NO. 8: Fig. 6c Comparison of canal, its distributaries and drain features in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate between and show sharp edges of all these features. This was not possible with the conventional FCC image. SHEET NO. 9: Fig. 6d Comparison of Nuh drain and its distributaries in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate between and show sharp edges of these features. This was not possible with the conventional FCC image. SHEET NO. 10: Fig. 6e Comparison of river and its tributaries in the conventional FCC with the new process based image. Notice that the new process based image could show sharp edges of these features. This was not possible with the conventional FCC image. SHEET NO. 11: Fig. 6f Comparison of water body (Damdama lake) features in the new process based image with the conventional FCC. See that the edges of a water body are distinctly discriminated by the new process. SHEET NO. 12: Fig. 7 Comparison of areas under crops and trees (forest)/ orchards in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate between the cropped and forested/ orchard (farm house) areas. In the conventional FCC both cropped and forest areas appear as blue coloured (see red encircled area in fig. 7a) while in the new process based image the cropped areas appear as bright white patches while the forest areas, with trees, appear as grayish patches (see red encircled area in fig. 7b). References Cooley, W. W. and Lohnes, P. R. (1985) Multivariate data analysis (2nd ed.), Robert E. Kreiger Publ. Co., Malabar, FL. Duda, R.O. and Hart, P.E. (1973) Pattern classification and scene analysis (New York: John Wiley and Sons). Jackson, R.D. (1983) Spectral indices in n-space. Remote Sensing of Environment, 13, 409-421. Kauth, R.J. and Thomas, G.S. (1976) The tasseled cap- a graphical description of the spectral temporal development of agricultural crops as seen by J.ANDSAT. Proceedings of the third international symposium on machine processing of remotely sensed data (W. Lafayette: Purdue University), pp. 48/41-51. Lillesand, T.M and Keiffer, R.W. (1979) Remote sensing and map interpretation, 2nd Edition John Willey and Sons, New York. Chengye, M. (Spectral Visions, affiliated with Stennis Space Center), Personal Communications via. e-mail, December 6,2000. Shannon, C. E. (1948) A mathematical theory of communications, Bell System Technical Journal 27:379-423. 4. I claim: 1. A spectral data analyzing process characterized with the capability to completely quantify, characterize and compress natural resource specific information contained in any spectral data type, containing any number of multi (broad)/ hyper (narrow)-spectral bands, and comprising the following steps: i) acquiring a spectral data stream of any earth feature; ii) determining the minima and maxima of each such spectral data stream; iii) determining the desired radiometric resolution; iv) determining the number of possible data elements in a given spectral data stream, based on the desired radiometric resolution and the maxima and minima; v) determining the probability of occurrence of each possible spectral data element per spectral data stream; vi) determining the total number of outcomes of all possible data elements per spectral data stream; vii) determining the total information content, based on the probability of occurrence and total number of outcomes of all possible data elements, per spectral data stream; viii) checking for any negative total information content per spectral data stream, at the above desired radiometric resolution; ix) re-defining the desired radiometric resolution, in case of negative total information content per spectral data stream and repeating the steps from (iv) to (ix) till positive total information content per spectral data stream, at a particular re-defined radiometric resolution, is obtained. 2. A spectral data analyzing process, as claimed in claim no. 1,wherein the said spectral data type comprises of either a record of an array of data elements corresponding to different spectral channels/wavebands per spectrum or a digital image comprising of a two-dimensional matrix of picture elements, commonly called pixels, acquired through the sensors mounted on platforms located anywhere from ground to space. 3. A spectral data analyzing process substantially as herein before described with reference to the accompanying drawings.

Full Text

Resource managers require fast and accurate methods for acquiring and interpreting data on development and management of natural/ man-made resources. Information on natural resources like soils, crops, water, forest, land cover, topography etc and man-made resources like canals, drains, railway lines etc is essential to understand the potential of land in any given area and to maintain the growth and development of that area. Besides this, a continued exploitation of these resources makes it imperative to know them in the context of their conservation, preservation and reclamation.
Remote sensing technique has proved to be of immense use in the study and monitoring of these earth resources. These remote sensing based resource surveys not only provide information in the areas of agriculture, forestry and grasslands for wasteland mapping; agricultural crop acreage and yield estimation, drought monitoring and assessment, land use/land cover mapping and forest productivity estimations but also in the areas of engineering and hydrology for flood monitoring and damage assessment, water resource management, ground water targeting, marine resource survey, urban planning, mineral targeting and environmental impact assessment etc. The important thing about these resource surveys is that they are fast, repetitive and temporo-spatially synoptic (i.e. cover large areas). Besides this, they are generally less expensive and less time consuming than the prevailing conventional field/ laboratory methods.
As the name suggests, remote sensing is the science and art of obtaining information about an object, area or phenomenon through the analysis of data acquired by a device that is not in contact with the object, area or phenomenon under investigation. (Lillesand and Kieffer, 1979). A remote sensing system principally comprises of an energy source; an object (here, any earth resource/ feature) of interest; a sensor mounted on a platform; and the data generated from the interactions between the energy and the

surface of the object of interest. The energy source, sun, emits Electro-Magnetic Radiation (EMR). These electro-magnetic radiations, falling on the surface of the earth, interact with the objects on the earth surface. After interaction some amount of energy is absorbed, some scattered and the rest is reflected back by the object. A sensor (mounted on a platform) senses this reflected energy that provides some information on the object. The amount of this energy absorbed, scattered and reflected by an object is in fact a function of the object's physical and chemical properties. Thus by recording reflected EMR through sensors, objects on the earth surface can be differentiated and their ground distribution can be mapped. The ratio of the EMR that is reflected from the surface of an object to that of the incident radiation is known as the spectral reflectance and the graph showing the spectral reflectance of an object at different wavebands is known as the spectral reflectance curve or spectral signature. In a remote sensing system, the amount of these reflected radiations (i.e. the data) acquired by the sensors (mounted on any platform, such as a tripod stand, a stepladder, an aircraft or a space station) is recorded in a digital form. In the case of the handheld field/ laboratory sensors, this is stored as a record comprising of an array of data elements corresponding to different spectral channels/wavebands per spectrum. While in the case of the aircraft/ satellite sensors, this is stored as a digital image comprising of a two-dimensional matrix of picture elements (commonly called pixels). Every such pixel is associated with an address in the matrix
f
(e.g. column & row number). Further each pixel has a third dimension, known as the gray value / brightness value/ intensity value/ radiometric value. This value in fact represents the reflectance characteristics of the corresponding object space elements (i.e. the ground objects falling within the spatial resolution of a picture element or a pixel) in different spectral channels/ wavebands.

Existing Spectral Data Analyzing Techniques
In remote sensing earth features are thus primarily characterized through multi-spectral signatures, recorded either as per cent reflectance or gray levels in different wavebands. However, in order to make characterization quantitative and more specific some spectral indices derived from information in these spectral channels/wavebands are often used which compress the data partially in two or more selected wavebands.
When sensors collected light in a single channel, both the spectral data and its analysis were simple, though inadequate of the task demanded of them. Advances in the sensor technology over last decade have outstripped the ability of the existing methods to adequately analyze the spectral data. Color and color-infrared imagery brought new demands for analysis and the LANDSAT Thematic Mapper (TM) with its 7 channels and other airborne platforms with even more than 120 channels have increased the degree of complication associated with the spectral data and its analysis significantly. Analysis of spectral data from all these sources is of several types:
• Analysis of individual bands,
• Simple band ratioing (e.g. Near Infrared/Red)
• Related normalized differences (of which the NDVI is a widely used measure),
• Linear combinations of wavebands (e.g. Near Infrared - Red, Near Infrared + Red,
Greenness index of Kauth and Thomas, 1976, etc.)
• Principal Component Analysis (PCA, Cooley and Lohnes, 1985), and
• Partial Least Squares (PLS - Modified PCA) analysis
Shortcomings of Existing Spectral Data Analyzing Techniques
Of the above, the first four spectral data analysis techniques can compress data in only 1-2 discrete spectral channels. It is well known by now that an individual waveband is related to only a few properties of the object under consideration, and not to the

complete set of object characteristics. Thus all these spectral data analysis techniques are capable of only partial data compression/ quantification (i.e. in only 1-2 wavebands). Other linear band combination spectral indices such as the greenness index of Kauth and Thomas (1976), though capable of compressing the information contained in four Multi-Spectral Scanner (MSS) bands fully, do not lead to a meaningful interpretation of results as the number of spectral channels is further increased (Jackson, 1983). Thus none of these techniques can be used for hyper-spectral data analysis.
The last two techniques, PCA and PLS, are n-dimensional spectral data analysis techniques as they can be used to compress/quantify information contained in even more than 4 spectral channels. The Principal Component Analysis (PCA) technique uses data from all the bands by multiplying the multi-channel image by an eigen-vector, giving different weights to different bands. The transformation itself is relatively rapid, but extensive analysis and computation goes into the development of the eigen-vector. (The actual computation of the eigen-vector is a slower operation (generally requiring many hours of processing time) with the computational efforts and time increasing with the square of the number of data channels. Partial Least Squares (PLS) is another modified form of PCA technique. While PLS has some advantages over PCA, the same computational work (if not more) is involved. Thus both these techniques require many hours of processing time for hyper-spectral data analysis. Besides this even these techniques are based on a simple assumption that some wavelengths or portions of the spectrum are rich in information about a feature of interest while the others are poor. Thus all these techniques totally ignore the fact that the spectrum as a whole has another dimension of information that is lost in treating it as discrete channels. Further as the number of spectral bands increases the dimensionality of state space in which an object is to be classified also increases; thereby leading to complicated class-separability and

clustering analysis (Duda and Hart, 1973). Thus, besides other shortcomings, both these (PCA and PLS) techniques also involve complicated class-separability and clustering analysis in n-dimensions; where 'n' is the number of spectral channels.
New Spectral Data Analyzing Process
In view of the above shortcomings associated with the existing techniques, I developed a new and novel process for quantifying information contained in the whole spectrum of any earth feature. The new spectral data analyzing process is based on the basic principles of communication theory, developed for quantifying redundancy in speech or electrical signals for ensuring fail-safe communication (Shannon, 1948). I applied this theory for the first time to the multi-spectral data (from LISS-III sensor of IRS-1C/1D Satellites) and to the hyper-spectral data (from RDACS/H3 sensor of NASA Aircrafts and from handheld field/ laboratory spectro-radiometers) to develop a novel process for quantification of information contained in a spectrum of any earth feature to its resource characteristic value. For this essentially, I treated each (point's / pixel's) spectrum as a data stream. The number of data elements (or gray levels) in one such spectrum varied with the desired radiometric resolution. These data elements (or gray levels) were taken as spectral events. First, the number of occurrences of each spectral event (at a user specified radiometric resolution) in a measured spectrum was calculated. From these outcomes per spectral event, the probability of occurrence of each possible spectral event and the total number of outcomes per spectrum were determined for obtaining total spectral information content. Hence, this way each point/ pixel was assigned a single value based upon its total spectral information content. The investigations revealed that the new spectral data analyzing process could lead to distinct characterization and discrimination of various types of earth surfaces in a 1-D data space thereby leading to simplified class-separability and clustering

analysis. The new process is simple, powerful, fast and can be used to process spectra with any number of data/spectral channels greater than 2.
Accordingly, this invention relates to a spectral data analyzing process comprising of the following steps: i. acquiring a spectral data stream of any earth feature;
ii. determining the minima and maxima of each such spectral data stream; iii. determining the desired radiometric resolution;
iv. determining the number of possible data elements in a given spectral data stream, based on the desired radiometric resolution and the maxima and minima;
v. determining the probability of occurrence of each possible spectral data element per spectral data stream;
vi. determining the total number of outcomes of all possible data elements per spectral data stream;
vii. determining the total information content, based on the probability of occurrence and total number of outcomes of all possible data elements, per spectral data stream;
viii. checking for any negative total information content per spectral data stream, at the above desired radiometric resolution;
ix. re-defining the desired radiometric resolution, in case of negative total information content per spectral data stream and repeating the steps from (iv) to (ix) till positive total information content per spectral data stream, at a particular re-defined radiometric resolution, is obtained.

Application Potential of New Hyper-Spectral Data Analyzing Process
Application to hyper-spectral data bases from laboratory/ field soectro-radiometers
To validate the application potential of the new hyper-spectral data analyzing process, I applied it to a huge independent soil spectral database. This database comprised of spectral information on 1295 soil samples: 802 soil samples from 4 Major Land Resource Areas (MLRA), 273 samples from S220 field in Walnut Creek watershed, Iowa and 120 samples from Minnesota site in U.S.
The MLRA group comprised of the surface (0-3 or 1-10 cm) and the sub-surface (3-10 or 10-30 cm) soil samples collected from 448 sites in the four Major Land Resource Areas (MLRAs). These included: MLRA-9 (the Palouse and Nez Perce Prairies located in eastern Washington and western Idaho); MLRA-67 (the Central High Plains located in eastern Colorado, southeastern Wyoming, and western Nebraska); MLRA-77 (the Southern High Plains in New Mexico and the panhandle of Texas); and MLRA-105 (the Northern Mississippi Valley-Drift less area, located in northeastern Iowa, southeastern Minnesota, and southwestern Wisconsin).
The Iowa group on the other hand comprised of soil cores, ranging between 38 to 108 cm, from 10 sampling sites within five soil series, namely Canisteo, Nicollet, Harps, Webster and Clarions. Five soil cores were collected at each site. The average length for the soil cores was 78.4 cm. The soil from each core was divided into 15-cm increments for a total of 277 samples. These soils in the Iowa group were very diverse in texture and other properties.
The soil samples from Minnesota site comprised of three Mollisols, one Alfisol, and one Ultisol. Waukegan silt loam (fine-silty over sandy or sandy-skeletal, mixed mesic Typic Hapludoll), Weld silt loam (fine, smectitic, mesic Aridic Argiustolls), and Palouse

silt loam (fine-silty, mixed, superactive, mesic Pachic Haploxerolls) were collected from sites near Rosemont, MN, Akron, CO, and Pullman, WA, respectively. Miami silt loam (fine-loamy, mixed, active, mesic Oxyaquic Hapludalfs) was collected from a site near West Lafayette, IN while Cecil sandy loam (Fine, Kaolinitic, thermic Typic Kanhapludults) was from a site in Oconee County, GA. Except for the Waukegan silt loam, both surface (A-horizon) and subsurface (B-horizon) soil samples from each soil series were investigated. Each soil was maintained at six different levels of moisture, including oven-dried soils. These included soils maintained at RH=0, 32.3, 55, 72.6, and 98%. The matric potentials for RH=32.3, 55, 72.6, and 98% are equal to -470, -81, -43, and -2.7 MPa, respectively. The gravimetric moisture content of all samples was measured immediately after the spectral readings.
Soil spectral measurements on these soil samples were recorded in the visible to near-infrared region (400-2500 nm) at 2 run intervals (1050 spectral channels) with a Perstorp NIR Systems 6500 scanning monochromator (Foss NIRSystems, Silver Spring, MD) - a laboratory spectrometer.
The results of this analysis (Fig. 1) clearly show that the new hyper-spectral data analyzing process has a great potential to discriminate between widely varying soil systems or Major Land Resource Areas 9/105; 67 and 77 (Fig. 1). This could be majorly attributed to a strong relationship of their total spectral information content (H) with the major color contributing soil-parameters such as sand, silt, iron (Fe) and total organic carbon (Fig. 1). Other soil properties such as P, Mn, Mg and K (which do not have any affect on soil color) did not reveal any such relationship with the total spectral information content. The analysis clearly shows (Fig. 1) that the soils with higher total organic carbon (and thus total nitrogen), iron, zinc (Zn) and smaller particle sizes have lower total spectral information content (H) as compared to the soils with lower total

organic carbon, Fe, Zn and higher particle sizes. Since the soils of the MLRA 77 contained lowest Fe and total organic carbon and highest sand and silt fractions therefore they were characterized with very high total spectral information contents. This was followed by the soils of MLRA 67, 9 and 105. It could further be observed that the soils of MLRA 9 and 105, having almost comparable soil- characteristics, were characterized with similar total spectral information contents. That the new process has a potential to discriminate between widely varying soil systems could be further confirmed from the analysis on the Iowa group of soils (Fig. 1). It can be seen from fig. 1 that a majority of Iowa soils, with total carbon values (ranging between 1-2 %) comparable with those for MLRA 67 and 77 soils, were quite distinct from the soils of MLRA 67 and 77; majorly because of their lower iron contents. The analysis thus clearly brought out the strength of the new hyper-spectral data analyzing process to discriminate between different soil systems.
Besides this, it could also be seen from fig. 2, that the total spectral information content (H) has a strong positive relationship (R2 ranging from 0.89-0.98) with soil moisture between 0 to -2.7 M Pas matric potentials (i.e. between air-dry to field capacity), for most soil types. It could thus be inferred that the new hyper-spectral data analyzing process could even discriminate between the same soils with widely varying soil moisture levels (i.e. it has capability to discriminate between wet and dry soil types).
Application to multi/ hyper-spectral databases from airborne platforms
Seeing the application potential of the new spectral data analyzing process on the above laboratory spectrometer based soil spectral database, I further applied it to the multi/ hyper-spectral digital images acquired through satellites / aircraft sensors. The structure of these databases is completely different from those obtained through
laboratory/ field spectrometers. Each pixel of the digital images can contain spectral information in any number of bands (ranging from 2 to infinity). The new process computes pixel-wise total spectral information content for these digital images, at a desired radiometric resolution.
Application of the new process to a hyper-spectral digital image (with 1m spatial resolution and 120 bands ranging from 471-828nm at 3nm bandwidth) collected by Stennis Space Center in September 1997 using a RDACS/H3 hyper-spectral sensor showed that the so processed hyper-spectral image (Fig. 3) of the Iowa corn (still to be harvested in the south) and soybean (already harvested in the north) fields could distinctly characterize and discriminate between various land features viz. cropped areas, wet and dry bare fields, areas with varying levels of weedy growth, pot holes, and weedy and non-weedy roads.
In traditional soil survey activities, soil profile studies take about 30% of the total time spent for fieldwork, whereas traversing for plotting soil boundaries takes around 70% time. Traversing is not only time consuming but also arduous element of soil survey. In fact in traditional system of soil survey this is directly related to the skill of soil surveyor and the traversing plan. Thus the use of the new hyper-spectral data analyzing process for such purposes would not only help in overcoming errors associated with subjectivity and drudgery in locating and plotting soil boundaries but would also help to minimize traversing to only a few well planned ground observations for validating soil boundaries and in generating soil maps of even inaccessible/ arduous areas. Besides, this since the new hyper-spectral data analyzing process-based image could distinctly discriminate between wet and dry fields. Therefore it can even serve as a robust tool for planning sowing operations.

It further showed that the new method based processing of the test hyper-spectral image (840 pixels high x 462 pixels wide x 120 channels x 16 bits per channel) took just 30 seconds. Besides this, it was observed that the size of the primary hyper-spectral digital data, in 120 spectral channels/ wavebands, was 90,959-KB. While the size of the new process based hyper-spectral image (i.e. Fig. 3), derived from the above primary data, was just 1,908 KB. Thus, the new process based hyper-spectral image required just 2% of the storage space required by the primary image. Hence the new hyper-spectral data analyzing process, besides being fast, could even lead to a greater reduction in data storage space requirement
Of more immediate interest, is the regular speckle like pattern overlaid upon the entire (new- process based) image (Fig. 3). This speckle can be readily removed from the new hyper-spectral method based image using First Fourier Transform (FFT). However, the point here is that the above speckle like pattern in the processed image revealed the capability of the new hyper-spectral data analyzing process to detect a previously unrecognized defect in the RDACS/H3 sensor (Chengye, NASA Personnel, personal communication). This defect causes a regular oscillation between the smoother and the noisier spectra, and results in slightly lower accuracy of predictions, with possible detection of non-existent patterns.
To check the new process for any errors and its further application potential to the multi/ hyper-spectral satellite/ airborne data, I used it to analyze a 4 band IRS-1C (LISS III), 8-bit, multi-spectral digital data (of 2740 by 2360 pixels size) of Sohna watershed for 9th May, 1999. The above data was purchased from NRSA, Hyderabad. The processing of this huge multi-spectral digital image (with 2740 by 2360 pixels) took just 8-10 minutes (depending on the desired radiometric resolution). As it can be observed from fig. 4, the new process based image showed no "interference/ speckle like pattern", as

observed in the case of the above processed hyper-spectral image in fig. 3. This once again confirmed:
a. No errors in the new spectral data analyzing process; and
b. The sensor defect-detection potential of the new process.
A comparison of a portion of the new process based multi-spectral image (Fig. 4) with the same portion of the conventional FCC (Fig. 5) of the Sohna watershed further revealed that the new process based multi-spectral image could distinctly characterize and discriminate between the vegetated & non-vegetated surfaces, ridges, water-bodies, stone quarrying sites, orchards/ farm houses, roads, road side trees, railway line and Gurgaon canal. It can be seen from the zoomed part (Fig. 6a) of the fig. 4 above, that the new process based multi-spectral image could even distinctly discriminate between the Gurgaon canal and the metal road and the cart track. In fact, the edges of a canal, in the new process based image in fig. 6a, were sharper than those for a metallic road. Further, the non-metallic roads or cart tracks or pack-tracks did not show any sharp edges in the new process based image. This discrimination between a canal and the metallic / non-metallic roads or cart tracks was not possible with the conventional FCC-image in Fig. 6a. Besides this, the new process based multi-spectral image could even distinctly discriminate a metal road from a railway track (Fig. 6b). As it can be seen from the zoomed part (Fig. 6b) of fig. 4, the edges of a railway track are more sharp and bold than those for a metal road in the new process based multi-spectral image. This discrimination between a railway track and a metal road was not possible with the conventional FCC image in fig. 6b. Figures (6c; 6d; 6e and 6f) further compare a canal/ its distributaries; drain/ its distributaries; a river/ its tributaries and a water body (lake) in the conventional multi-spectral images with the new process based mutli-spectral images. It can be clearly seen from these images that the new process based image could even

distinctly discriminate between these features. All this clearly shows that the new process has a great potential in many change detection regulatory programs for updating information on vegetation/ terrain conditions and in land use planning also. Besides this it can be seen from figs 6(a-f), the new process based multi-spectral images could distinctly characterize the edges of the various natural/ man-made (linear/ non-linear) features such as metal roads, railway lines, canals, rivers, drains and water bodies. This was not possible with the conventional FCC based images. Thus the new hyper-spectral data analyzing process was also found to be an excellent edge detection tool.
I further analyzed another 4 band IRS-ID (LISS III), 8-bit multi-spectral digital data (with 2218 by 2024 pixels) for Sohna watershed for 05th February 2001 in order to test the potential of the new process to discriminate between agricultural and forest areas. The above data was also purchased from NRSA, Hyderabad. On comparing a portion the new method based image (Fig. 7b) with the same portion of the conventional False Colour Composite (FCC; Fig. 7a), it could be observed that the new process based image could distinctly characterize and discriminate between the cropped (i.e. agricultural) and the forested/ orchard (farmhouse) areas. As it can be seen from fig. 7b (see red encircled area in Fig. 7b), the cropped areas in the new process based image appeared as bright white patches while the forest/ orchard areas appeared as non-uniform or uniform grayish patches. This discrimination between the cropped and the forest areas was not possible with the conventional FCC (Fig. 7a), as here both these areas (see red encircled area in Fig. 7a) appeared as similar blue patches. To separate these similarly appearing blue coloured areas under crops and trees in the conventional FCC, one has to do an extensive ground truthing of the area. However with the new spectral data analyzing process this ground truthing could be either completely avoided (if the

area is known) or greatly minimized (to only a few selected points of confusion). Thus the new process based processing of spectral data can even lead to much better (and more accurate) estimates of areas under crop and forest covers with no/ very minimal ground truthing.
It was further found that although the size of the primary IRS 1-C (LISS III) and IRS-1D (LISS III) data in 3 bands was 18,955 and 13, 165 KB respectively; yet the size of the new process based multi-spectral images was just 6,316 and 4,387 KB, respectively. Thus it was once again confirmed that the new process leads to a great reduction in the data storage space requirement
Hence it can be observed that the new process is the only simple and fast hyper-spectral data analyzing means for complete quantification of information contained in a spectrum, comprising of any number of spectral channels/ wavebands from 2 to infinity, of any natural surface. In fact Indian Space Research Organization has planned to launch a narrow band spectrometer comprising of about 200 wavebands (in visible and MR region) for satellite based remote inventory of natural resources. In this context this is the only hyper-spectral data analyzing process, which can be used straight away for any natural/ man made resource inventory (i.e. its characterization/ discrimination).
Besides this, since the digital image generated through the new hyper-/multi-spectral data analyzing process is a single channel-image therefore it's unsupervised/ supervised classification would be based on a simplified 1-D clustering analysis techniques as compared to the more complicated 3-D clustering analysis techniques required for unsupervised/ supervised classification of a conventional FCC. Thus the new spectral data analyzing process can even lead to a simplified 1-D clustering analysis.
CAPTIONS TO FIGURES/ DRAWINGS
SHEET NO. 1: Fig. 1 Relationship of total spectral information content (H) with major soil colour contributing soil parameters.
SHEET NO. 2: Fig. 2 Relationship of total spectral information content (H) with soil moisture between air-dry to field capacity.
SHEET NO. 3: Fig. 3 New process-based image of Iowa fields. Note speckle pattern caused by noise in the hyper-spectral sensor. See that the cropped areas, moist and dry bare fields, areas with varying levels of weedy growth, potholes, and weedy and non-weedy roads are clearly distinguishable.
SHEET NO. 4: Fig. 4 New process-based image of Sohna watershed, obtained through the processing of IRS-1C (LISS III) band- 2, 3 and 4 data for 9th May, 1999.
SHEET NO. 5: Fig. 5 Conventional False Colour Composite (FCC) of Sohna watershed, derived from IRS-1C (LISS III) band- 2, 3 and 4 data for 9th May 1999. Notice, all vegetated areas are in light blue colour.
SHEET NO. 6: Fig. 6a Comparison of metallic / non-metallic roads and canal features in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate between and show sharp edges of the Gurgaon canal and the metal roads. These edges were sharper for a canal than a metallic road. The cart tracks did not show any sharp edges in the new process based image. This discrimination between a canal, a metallic road and a cart track was not possible with the conventional FCC image.
SHEET NO. 7: Fig. 6b Comparison of metallic road and railway track features in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate a metallic road and a railway track. This discrimination between a railway track and a metallic road was not possible with the conventional FCC image.
SHEET NO. 8: Fig. 6c Comparison of canal, its distributaries and drain features in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate between and show sharp edges of all these features. This was not possible with the conventional FCC image.
SHEET NO. 9: Fig. 6d Comparison of Nuh drain and its distributaries in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate between and show sharp edges of these features. This was not possible with the conventional FCC image.
SHEET NO. 10: Fig. 6e Comparison of river and its tributaries in the conventional FCC with the new process based image. Notice that the new process based image could show sharp edges of these features. This was not possible with the conventional FCC image.
SHEET NO. 11: Fig. 6f Comparison of water body (Damdama lake) features in the new process based image with the conventional FCC. See that the edges of a water body are distinctly discriminated by the new process.
SHEET NO. 12: Fig. 7 Comparison of areas under crops and trees (forest)/ orchards in the conventional FCC with the new process based image. Notice that the new process based image could distinctly discriminate between the cropped and forested/ orchard (farm house) areas. In the conventional FCC both cropped and forest areas appear as blue coloured (see red encircled area in fig. 7a) while in the new process based image the cropped areas appear as bright white patches while the forest areas, with trees, appear as grayish patches (see red encircled area in fig. 7b).
References
Cooley, W. W. and Lohnes, P. R. (1985) Multivariate data analysis (2nd ed.), Robert E. Kreiger Publ. Co., Malabar, FL.
Duda, R.O. and Hart, P.E. (1973) Pattern classification and scene analysis (New York: John Wiley and Sons).
Jackson, R.D. (1983) Spectral indices in n-space. Remote Sensing of Environment, 13, 409-421.
Kauth, R.J. and Thomas, G.S. (1976) The tasseled cap- a graphical description of the spectral temporal development of agricultural crops as seen by J.ANDSAT. Proceedings of the third international symposium on machine processing of remotely sensed data (W. Lafayette: Purdue University), pp. 48/41-51.
Lillesand, T.M and Keiffer, R.W. (1979) Remote sensing and map interpretation, 2nd Edition John Willey and Sons, New York.
Chengye, M. (Spectral Visions, affiliated with Stennis Space Center), Personal Communications via. e-mail, December 6,2000.
Shannon, C. E. (1948) A mathematical theory of communications, Bell System Technical Journal 27:379-423.

4. I claim:
1. A spectral data analyzing process characterized with the capability to completely
quantify, characterize and compress natural resource specific information contained in
any spectral data type, containing any number of multi (broad)/ hyper (narrow)-spectral
bands, and comprising the following steps:
i) acquiring a spectral data stream of any earth feature;
ii) determining the minima and maxima of each such spectral data stream;
iii) determining the desired radiometric resolution;
iv) determining the number of possible data elements in a given spectral data stream,
based on the desired radiometric resolution and the maxima and minima; v) determining the probability of occurrence of each possible spectral data element per
spectral data stream; vi) determining the total number of outcomes of all possible data elements per spectral
data stream; vii) determining the total information content, based on the probability of occurrence and
total number of outcomes of all possible data elements, per spectral data stream; viii) checking for any negative total information content per spectral data stream, at the
above desired radiometric resolution; ix) re-defining the desired radiometric resolution, in case of negative total information
content per spectral data stream and repeating the steps from (iv) to (ix) till positive
total information content per spectral data stream, at a particular re-defined
radiometric resolution, is obtained.
2. A spectral data analyzing process, as claimed in claim no. 1,wherein the said spectral
data type comprises of either a record of an array of data elements corresponding to
different spectral channels/wavebands per spectrum or a digital image comprising of a
two-dimensional matrix of picture elements, commonly called pixels, acquired through
the sensors mounted on platforms located anywhere from ground to space.
3. A spectral data analyzing process substantially as herein before described with
reference to the accompanying drawings.

Documents:

825-DEL-2001-Abstract.pdf

825-del-2001-claims.pdf

825-del-2001-correspondence-others.pdf

825-del-2001-correspondence-po.pdf

825-del-2001-description (complete).pdf

825-del-2001-drawings.pdf

825-del-2001-form-1.pdf

825-del-2001-form-19.pdf

825-del-2001-form-2.pdf

825-del-2001-form-3.pdf

« Previous Patent

Next Patent »

Patent Number

214697

Indian Patent Application Number

825/DEL/2001

PG Journal Number

09/2008

Publication Date

29-Feb-2008

Grant Date

14-Feb-2008

Date of Filing

02-Aug-2001

Name of Patentee

INDIAN COUNCIL FOR AGRICULTURAL RESEARCH

Applicant Address

KRISHI BHAWAN, DR. RAJENDRA PRASAD ROAD, NEW DELHI 110001, INDIA

Inventors:

#	Inventor's Name	Inventor's Address
1	DR(MRS) RAVINDER KAUR	DIVISION OF ENVIRONMENTAL SCIENCES, INDIAN AGRICULTURAL RESEARCH INSTITUTE, NEW DELHI 110012, INDIA

PCT International Classification Number

G01J 3/28

PCT International Application Number

N/A

PCT International Filing date

PCT Conventions:

#	PCT Application Number	Date of Convention	Priority Country
1			NA