Target Classification Using Adaptive Feature Extraction and Subspace Projection for Hyperspectral Imagery
- 855 Downloads
Hyperspectral imaging sensors have been widely studied for automatic target recognition (ATR), mainly because a wealth of spectral information can be obtained through a large number of narrow contiguous spectral channels (often over a hundred). Targets are man-made objects (e.g., vehicles) whose constituent materials and internal structures are usually substantially different from natural objects (i.e., backgrounds). The basic premise of hyperspectral target classification is that the spectral signatures of target materials are measurably different than background materials, and most approaches further assume that each relevant material, characterized by its own distinctive spectral reflectance or emission, can be identified among a group of materials based on spectral analysis of the hyperspectral data.
We propose a two-class classification algorithm for hyperspectral images in which each pixel spectrum is labeled as either target or background. The algorithm is based on a mixed spectral model in which the reflectance spectrum of each pixel is assumed to be a linear mixture of constituent spectra from different material types (target and background materials). In order to address the spectral variability and diversity of the background spectra, we estimate a background subspace. The background spectral information spreads over various terrain types and is represented by the background subspace with substantially reduced dimensionality. Each pixel spectrum is then projected onto the orthogonal background subspace to remove the background spectral portion from the corresponding pixel spectrum.
The abundance of the remaining target portion within the pixel spectrum is estimated by matching a data-driven target spectral template with the background-removed spectrum. We use independent component analysis (ICA) to generate a target spectral template. ICA is used because it is well suited to capture the structure of the small targets in the hyperspectral images. For comparison purposes a mean spectral template is also generated by simply averaging the target sample spectra. Classification performance for both of the above-mentioned target extraction techniques are compared using a set of HYDICE hyperspectral images.
KeywordsIndependent Component Analysis Independent Component Analysis Constant False Alarm Rate Hyperspectral Imagery Spectral Variability
Unable to display preview. Download preview PDF.
- Bhanu, B.: Automatic target recognition: State-of-the-art survey. IEEE Transactions on Aerospace and Elect. Syst. 22 (1986) 364–379Google Scholar
- Reed, I.S., Yu, X.: Adaptive multiple-band CFAR detection of an optical pattern with unknown spectral distribution. IEEE Transactions Acoustics, Speech and Signal Process. 38 (1990) 1760–1770Google Scholar
- Schachter, B.J.: A survey and evaluation of FLIR target detection/segmentation algorithm. In: Proc. of DARPA Image Understanding Workshop. (1982) 49–57Google Scholar
- Aviram, G., Rotman, S.R.: Evaluating human detection performance of targets and false alarms, using a statistical texture image metric. Optical Engineering 39 (2000) 2285–2295Google Scholar
- Manolakis, D., Shaw, G., Keshava, N.: Comparative analysis of hyperspectral adaptive matched filter detector. In: Proc. SPIE. Volume 4049. (2000) 2–17Google Scholar
- Joliffe, I.T.: Principal Component Analysis. Springer-Verlag, New York (1986)Google Scholar
- Bishop, C.M.: Baysian PCA. In: Neural Information Processing System. Volume 11. (1998) 382–388Google Scholar
- Rajan, J.J., Rayner, P.: Model order selection for the singular value selection and the discrete Karhunen-Loeve transform using a bayesian approach. IEE Proc. Image Signal Process 144 (1997) 116–123Google Scholar
- Hyvärinen, A., Oja, E.: Independent component analysis: Algorithms and applications. Neural Networks 13 (2000) 411–430Google Scholar
- Hyvärinen, A.: Survey on independent component analysis. Neural Computing Surveys 2 (1999) 94–128Google Scholar
- Hyvärinen, A.: Sparse code shrinkage: Denoising of nongaussian data by maximun likelyhood estimation. Neural Computation 11 (1999) 1739–1768Google Scholar
- Harsanyi, J.C., Chang, C.I.: Hyperspectral image classification and dimensionality reduction: An orthogonal subspace projection approach. IEEE Transactions Geosci. Remote Sensing 32 (1994) 779–785Google Scholar
- Ashton, E.L.: Detection of subpixel anomalies in multispectral infrared imagery using an adaptive bayesian classifier. IEEE Transactions Image Process. 36 (1998) 506–517Google Scholar
- Chang, C.I., Zhao, X.L., Althouse, M., Pan, J.J.: Least squares subspace projection approach to mixed pixel classification for hyperspectral images. IEEE Transactions Geosci. Remote Sensing 36 (1998) 898–912Google Scholar
- Bartlett, M.S.: a dissertation: Face Image Analysis by Unsupervised Learning and Redundancy Reduction. University of California, San Diego (1998)Google Scholar
- Comon, P.: Independent component analysis: A new concept? Signal Processing 36 (1997) 287–314Google Scholar
- Hyvärinen, A.: Fast and robust fixed-point algorithms for independent component analysis. IEEE Transactions Neural Networks. 10 (1999) 626–634Google Scholar
- Minka, T.P.: Automatic choice of dimensionality for PCA. M.I.T Media Laboratory Perceptual Computing Section Technical Report (2000)Google Scholar
- Strang, G.: Linear Algebra and Its Applications. Harcourt Brace & Company (1986)Google Scholar
- Kwon, H., Der, S.Z., Nasrabadi, N.M.: An adaptive unsupervised segmentation algorithm based on iterative spectral dissimilarity measure for hyperspectral imagery. In: Proc. SPIE. Volume 4310. (2001) 144–152Google Scholar