A recently developed orthogonal subspace projection (OSP) approach has been successfully applied to AVIRIS as well as HYDICE data for image classification. However, it has found that OSP performs poorly in multispectral image classification such as 3-band SPOT data. This is primarily due to the fact that the number of signatures to be classified is greater than that of spectral bands used for data acquisition in which case the effects of uninterested signatures cannot be properly annihilated via orthogonal projection. This constraint, referred to as band number constraint (BNC) is generally not applied to hyperspectral images because the number of signatures resident within the images is usually far less than the total number of spectral bands. In this paper, a new approach, called unsupervised vector quantization-based target subspace projection (UVQTSP) is presented which can be implemented in an unknown environment with all required information obtained from the data to be processed. The proposed UVQTSP has practical advantages over OSP, specifically, it relaxes the band number constraint (BNC) so that it can be applied to multispectral imagery. The UVQTSP uses vector quantization to find a set of clusters representing the unknown signatures and interferers which will be eliminated prior to target detection and classification. The number of clusters can be determined by constraints such as the intrinsic dimensionality or the number of spectral bands. This process is carried out in an unsupervised manner without training data. The superiority of UVQTSP is demonstrated through real data including SPOT and HYDICE images.
[1]
C.-C. Jay Kuo,et al.
A new initialization technique for generalized Lloyd iteration
,
1994,
IEEE Signal Processing Letters.
[2]
Chein-I Chang,et al.
Hyperspectral image classification and dimensionality reduction: an orthogonal subspace projection approach
,
1994,
IEEE Trans. Geosci. Remote. Sens..
[3]
Robert M. Gray,et al.
An Algorithm for Vector Quantizer Design
,
1980,
IEEE Trans. Commun..
[4]
A. Goetz,et al.
Terrestrial imaging spectroscopy
,
1988
.
[5]
Chein-I Chang,et al.
A posteriori least squares orthogonal subspace projection approach to desired signature extraction and detection
,
1997,
IEEE Trans. Geosci. Remote. Sens..
[6]
Shang-Liang Chen,et al.
Orthogonal least squares learning algorithm for radial basis function networks
,
1991,
IEEE Trans. Neural Networks.
[7]
Mark L. G. Althouse,et al.
Least squares subspace projection approach to mixed pixel classification for hyperspectral images
,
1998,
IEEE Trans. Geosci. Remote. Sens..