Method for creating thematic maps using segmentation of ternary diagrams

• US 6,356,646 B1
• Filed: 02/19/1999
• Issued: 03/12/2002
• Est. Priority Date: 02/19/1999
• Status: Expired due to Fees

• Current Assignee: Clyde H. Spencer
• Original Assignee: Clyde H. Spencer
• Inventors: Spencer, Clyde H.
• Primary Examiner(s): Boudreau, Leo
• Assistant Examiner(s): Tabatabai, Abolfazl
• Application Number:
  US09/252,718
• Time in Patent Office: 1,117 Days
• Field of Search: 382/103, 348/144, 348/145, 348/146, 348/147, 348/149
• US Class Current: 382/103
• CPC Class Codes: G06T 2207/10036 Multispectral image; Hyperspectral image
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/174 involving the use of two or more images
• CPC Class Codes: G06V 20/13 Satellite images

Abstract

A method and process is disclosed for computer processing and display of remote sensing, multispectral imagery for the purpose of identifying ground targets and classifying the imagery to create thematic maps. The successful application relies on the use of images from optimal bands; those bands include a near-infrared, mid-infrared, and visible band. The percentages of each of the three bands are plotted on a ternary diagram (34), which is used as a graphical device to allow target identification. Additionally, the ternary diagram feature space is segmented to define thematic classes and allow thematic classification with a computer. Vegetation (10) and mineral clusters (11) are separated easily. Further discrimination within the separate vegetation (10) and mineral (11) point clusters is possible. A spectral library (49), convolved to the bandwidth of the employed multispectral sensor bandpasses, and converted to ternary percentages, is utilized to locate specific spectral targets. The graphical ternary diagram (34) and false-color multispectral image display (46) are linked in real-time through a lookup-table (43) to allow an operator to interactively alter each of them through a change to the other. Included is a method to accomplish automatic, hierarchical classification of the images without operator intervention. The percent of vegetation ground-cover (i.e. a vegetation index) can be estimated, and atmospheric scattering and absorption effects are empirically corrected, interactively.

Claims

1. A method of identifying spectrally-distinctive reflective ground targets, represented by a plurality of pixels in multispectral imagery, comprising the following steps:

(a) selecting at least three image bands from a multitude of coregistered multispectral bands, wherein said selected bands consist of a band in the visible, a band in the near-infrared, and a band in the mid-infrared portion of the electromagnetic spectrum;

(b) transformation means for correcting said pixels'"'"' original values for sensor calibration, solar spectrum characteristics, sensor and illumination geometry, and atmospheric scattering and absorption, thereby resulting in values that closely approximate true reflectance, herein referred to as apparent reflectance values;

(c) conversion means for converting plurality of said pixels'"'"' apparent reflectance values to ternary percentages, wherein each member of a triad of percentages is the fractional part, of the sum, of one pixels'"'"' values from each of said selected bands;

(d) plotting indicia on a conventional ternary diagram or triangular plot, wherein each pixel of the plurality of said pixels has an indicium corresponding to said pixel;

(e) determining a plurality of spectral-class segment boundaries by visual inspection, from a reference ternary diagram containing plotted points or indicia obtained from a spectral library of reflectors of interest convolved to the bandwidth of the multispectral sensor used, wherein said boundaries are changeable and determined by class-resolution requirements of a classification task;

(f) transferring said changeable boundaries to a triangular-shaped, transparent template that is congruent with said ternary diagram; and

(g) superimposing said template over said ternary diagram to delineate plurality of said spectral-class boundaries;

whereby, the position of the indicia on the ternary diagram, with respect to the plurality of the enclosing boundaries of said superimposed template, constrains the identification of the corresponding image pixels to be members of the spectral class plotted on said reference ternary diagram, within said enclosing boundaries, thus identifying said pixels.

2. The method of identifying spectrally-distinctive reflective ground targets, as set forth in claim 1, step (e) wherein the template boundaries determination further includes, optionally:

(a) spectral-class boundary determination means empirically derived from actual spectral ground-truth to establish the position on a master ternary diagram suitable for use with the imagery being analyzed; and

(b) spectral-class boundary determination means empirically derived from photointerpretation-derived ground-truth of imagery, to establish the position on a master ternary diagram suitable for use with the imagery being analyzed.

3. The general method of identifying spectrally-distinctive reflective ground targets, as set forth in claim 1, comprising the following software-controlled steps:

(a) providing an input means to ingest multispectral image data;

(b) providing a storage means to store programmatic control and data;

(c) reading said image data into storage;

(d) providing computer input devices of known type, thereby enabling an operator to select requisite image bands interactively;

(e) reading a subset of multispectral image data, comprising said selected bands, into an array;

(f) reading pixels values from said array;

(g) correcting said pixel values, thereby obtaining apparent reflectance values;

(h) displaying said pixel values as a matrix, thereby forming an image;

(i) selecting a region of interest in said image, interactively;

(j) converting said apparent reflectance values, within said region of interest, to ternary percentages;

(k) generating a graphical output, on a display device, in the form of a ternary diagram containing plotted points or indicia corresponding to said pixels;

(l) displaying pre-calculated, general spectral-class segment boundaries on said ternary diagram; and

(m) comparing indicia positions to a table of pre-determined virtual boundaries to establish finer class membership discrimination;

whereby, the steps recited above result in the semiautomatic identification of said selected sub-set of pixels through programmatic control of a general-purpose computer.

4. The method of identifying spectrally-distinctive reflective ground targets, as set forth in claim 1, wherein step (c) said pixel'"'"'s initial values further include, optionally:

(a) raw digital numbers, representing a function of radiance recorded by an overhead sensor.

5. The method of identifying spectrally-distinctive reflective ground targets, as set forth in claim 1, wherein step (c) said pixel'"'"'s initial values further include, optionally:

(a) exoatmospheric reflectance values resulting from a processing means to correct for sensor calibration, solar spectrum characteristics, and sensor and illumination geometry.

6. The method of identifying spectrally-distinctive reflective ground targets, as set forth in claim 1, wherein step (d) plotting further includes:

(a) a coloration means to assign a color to an indicium where said color is determined by ternary percentages in a manner whereby a red component of a resulting hue is directly correlated with one percentage, a green component with another, and a blue component with the remaining percentage.

7. The method of identifying spectrally-distinctive reflective ground targets, as set forth in claim 1, wherein step (d) plotting further includes:

(a) a coloration means to assign a color to an indicium where said color is determined by ternary percentages in a manner whereby a red component of a resulting hue is directly correlated with a mid-infrared percentage, a green component with a near-infrared percentage, and a blue component with a visible percentage.

whereby, said indicium is assigned a hue that approximates the natural hue of the ground-cover spectral class the indicium corresponds to.

8. The method of identifying spectrally-distinctive reflective ground targets, as set forth in claim 1, comprising the following software-controlled steps:

(a) providing an input means to ingest multispectral image data;

(b) providing storage means to store programmatic control and data;

(c) reading said image data into storage;

(d) providing computer input devices of known type, thereby enabling an operator to select the requisite image bands interactively;

(e) reading a subset of multispectral image data, comprising said selected bands, into an array;

(f) reading a plurality of pixels values from said array;

(g) correcting said pixel values, thereby obtaining apparent reflectance values;

(h) converting said apparent reflectance values to ternary percentages; and

(i) writing said ternary percentages to a sequential file;

whereby, a transformed image is created, which is in machine-readable form, and can be displayed on a graphics-enabled computer;

thereby, if the near-infrared percentage surrogate is directed to the green gun of a cathode ray tube, the mid-infrared percentage surrogate is directed to the red gun of a cathode ray tube, and the visible percentage surrogate is directed to the blue gun of a cathode ray tube, a standardized, false-color image will be displayed that has the desirable properties of displaying ground cover in hues approximating their natural hues.

9. A method of locating spectrally-distinctive reflective ground targets, represented by a plurality of pixels in multispectral imagery, and deriving a thematic map, comprising the following software-controlled steps:

(a) providing an input means to ingest multispectral image data;

(b) providing a storage means to store programmatic control and data;

(c) reading said image data into storage;

(d) providing computer input devices of known type, thereby enabling an operator to select requisite image bands interactively;

(e) reading a subset of multispectral image data, comprising said selected bands, into an array;

(f) reading pixels values from said array;

(g) correcting said pixel values, thereby obtaining apparent reflectance values;

(h) displaying said pixel values as a matrix, thereby forming an image;

(i) converting said apparent reflectance values to ternary percentages;

(j) loading a lookup table, sequentially, with said ternary percentages;

(k) generating a graphical output, on a display device, in the form of a ternary diagram containing plotted points or indicia;

(l) segmenting said ternary diagram by selecting a plurality of boundaries, interactively, that defines a desired spectral class;

(m) comparing entries in said lookup table with boundaries of spectral-class segment;

(n) highlighting image pixels if said pixels correspond to said lookup table entries within said region of interest;

whereby, the steps recited above result in locating the plurality of said pixels in said image that spectrally correspond to said indicia in said selected spectral-class segment;

further, the steps (l), (m) and (n), recited above, iteratively performed on the whole of the plurality of indicia, with non-overlapping segments, result in a plurality of correspondences between said segments and the plurality of said image pixels;

whereupon, assigning unique a color to each spectral class, for which there is at least one correspondence in the image pixels, results in the generation of a thematic map, depicting the types of ground cover as represented by the plurality of said spectral classes.

10. A method of locating spectrally-distinctive reflective ground targets and deriving a thematic map as set forth in claim 9, wherein steps (h), (J), (k), (m), (n), and (o) are omitted and (l) is replaced with the further steps:

(a) creating a pre-determined plurality of virtual spectral-class boundaries, comprising a plurality of test parameters, each parameter consisting of no more than two ternary percentages, which establish a boundary between two super-classes;

(b) performing hierarchical logical tests between the relative values of said virtual, spectral-class boundaries and the corresponding ternary percentages for the plurality of said image pixels;

(c) continuing said logical comparisons, with additional parameters, on the pixel along a decision tree until an identification is made;

(d) assigning an identification label, such as a number, to the pixel;

(e) writing said label to a sequential file;

(f) iterating the procedure by selecting the next pixel in sequence;

(g) continuing until all of the plurality of said pixels have been tested;

whereby, the steps recited above result in an automatic classification of said multispectral image suitable for unattended, batch processing in a volume production environment, wherein the resulting thematic map is in a machine-readable form.

11. A method of locating spectrally-distinctive reflective ground targets as set forth in claim 9, wherein step (l) is replaced with the further steps:

(a) storing a reference spectral library convolved to appropriate bandpasses;

(b) providing an interactive selection means for extracting at least one member from said reference spectral library;

(c) converting said selected spectral library member reflectance values to ternary percentages;

(d) providing a means for establishing a tolerance of acceptance of a match;

(e) comparing said spectral library member, consecutively, with every entry in said lookup table;

(f) highlighting the corresponding image pixel if said match is obtained;

(g) repeating step (e) for each additional said member of said spectral library selected;

whereby, the steps recited above result in locating the plurality of said pixels in said image that spectrally correspond to said selected member of said reference spectral library.

12. A method of interactively, empirically correcting multispectral imagery for multiplicative and additive haze and atmospheric absorption, comprising the following software-controlled steps:

(a) providing an input means to ingest multispectral image data;

(b) providing storage means to store programmatic control and data;

(c) reading said multispectral image data into storage;

(d) providing computer input devices of known type, thereby enabling an operator to select the requisite image bands interactively;

(e) reading a subset of multispectral image data, comprising said selected bands, into an array;

(f) reading pixels values from said array;

(g) converting raw image pixel-values to ternary percentages;

(h) loading a lookup table, sequentially, with said ternary percentages;

(i) duplicating said ternary percentages in a shadow lookup table;

(j) generating a graphical output, on a display device, of a plurality of different ternary diagrams containing plotted points or indicia;

(k) displaying at least one pre-determined alignment band on each of said plurality of ternary diagrams;

(l) providing interactive input means whereby an operator adjusts visible indicators of gain and bias;

(m) providing a means to record and apply adjusted gain and bias values to said ternary percentages;

(n) providing a means to update said shadow lookup table with the adjusted percentages;

(o) regenerating said graphical output comprising adjusted said indicia, on said display device, using the updated shadow lookup table;

(p) continuing said interactive adjustments until satisfactory alignment with plurality of said alignment bands is achieved;

(q) providing a means to convert final, adjusted ternary-percentage gain and bias values to values to apply directly to the raw data to obtain apparent reflectance;

whereby, said raw image data is corrected for a multitude of multiplicative and additive effects that are essential to take into account to create apparent reflectance images.