Methods for in-scene atmospheric compensation by endmember matching
Methods are provided for automatically performing atmospheric compensation of a multi or hyper spectral image. One method comprises transforming at least two endmembers extracted from an image into at-ground reflectance. The transformation may be approximate and/or only in certain spectral bands in order to reduce processing time. A matching component is then located in a spectral library for each of the at least two extracted endmembers. Gain and offset values are then calculated using the at least two matched extracted endmember and spectral library component pairs. At least part of the image is then compensated using the calculated gain and offset values. Another method uses at least one endmember extracted from the image and a black level. Methods for atmospheric compensation using water vapor content of pixels are also provided. In addition, methods for shadow correction of hyper and multi spectral images are provided.
1. A method for automatically performing atmospheric compensation of a multi or hyper spectral image, the method comprising:
extracting at least two endmembers from the image;
transforming, the at least two extracted endmembers to an at-ground reflectance;
finding a matching reflectance component in a spectral library for each of the transformed at least two extracted endmembers, where each extracted endmember and found matching reflectance component taken together form a matched extracted endmember and spectral library component pair;
calculating for each of a plurality of spectral bands in the image a gain value and an offset value using the matched extracted endmembers and spectral library component pairs; and
compensating at least part of the image using the calculated gain and offset values.
2. The method of claim 1 wherein the transforming uses radiative transfer modeling to transform the at least two extracted endmembers into approximate at-ground reflectance.
3. The method of claim 2 , wherein the radiative transfer modeling in the transforming uses a delta-Eddington multiple scattering algorithm where an atmosphere modeled in the scattering algorithm is composed of a single uniform representative atmospheric layer.
4. The method of claim 1 , wherein the transforming uses empirical or statistical-empirical modeling to transform the at least two extracted endmembers into approximate at-ground reflectance.
5. The method of claim 1 , wherein the calculating uses a method of least squares to fit a line to a data set comprising a plurality of the at least two extracted endmembers and spectral library component pairs.
6. The method of claim 1 , further comprising:
assigning, after the finding and before the calculating, a weight value to a plurality of the at least two matched extracted endmember and spectral library component pairs, the weight value being proportional to a degree of matching between the transformed endmember and its respective matched spectral library component,
wherein the calculating uses the assigned weight values in a method of weighted least squares to fit a line to a data set comprising a plurality of the at least two matched endmember and spectral library component pairs.
7. The method of claim 6 , wherein the degree of matching between the transformed endmember and its respective matched spectral library component is based on at least one of a spectral angle between the transformed endmember and the matched spectral library component and a Euclidean distance between the transformed endmember and the matched spectral library component.
8. The method of claim 1 , wherein the finding uses the metric F=D c tan(θ s ) to find a matching component in the spectral library for the transformed endmember, where θ s is a spectral angle between the transformed endmember and a spectral library component, and D s is a Euclidean distance between the transformed endmember and a spectral library component, where the finding involves comparing the transformed endmember to a plurality of components in the spectral library, and wherein the component in the spectral library selected as a match for the transformed endmember is the component that produces the lowest F metric value.
9. The method of claim 1 , further comprising: before the transforming,
masking one or more anomalous pixels in the image; and
extracting the at least two end members from the image.
10. The method of claim 9 , wherein the one or more anomalous pixels include at least one of a cloud pixel, a corrupted pixel and a vegetation pixel.
11. The method of claim 9 , further comprising: transforming, before the masking, at least part of the image into top-of-atmosphere reflectance.
12. The method of claim 11 , wherein at least some of the portion of the image corresponds to one or more window regions of the atmosphere, a window region being a spectral band in which there is little or no absorption of light by molecules in the atmosphere.
13. The method of claim 11 , wherein the transforming into top-of-atmosphere reflectance includes using downwelling flux data obtained using a downwelling irradiance sensor.
14. The method of claim 5 , further comprising:
computing, after the calculating, a regression error for the method of least squares in at least one band of the input image;
forwarding the regression error to an exploitation algorithm.
15. The method of claim 1 , wherein the calculating an offset value is further based on a black level obtained from at least one of the image and radiative transfer modeling.