Feature extraction of hyperspectral images using wavelet and matching pursuit

Since hyperspectral images contain rich and fine spectral information, an improvement of land use/cover classification accuracy is highly expected from the utilization of such images. However, the traditional statistics-based classification methods which have been successfully applied to multispectral data in the past are not as effective as to hyperspectral data. One major reason is that the number of spectral bands is too large relative to the number of training samples. This problem is caused by curse of dimensionality, which refers to the fact that the sample size required for training a specific classifier grows exponentially with the number of spectral bands. A simple but sometimes very effective way to overcome this problem is to reduce the dimensionality of hyperspectral images. This can be done by feature extraction that a small number of salient features are extracted from the hyperspectral data when confronted with a limited size of training samples. In this paper, a new feature extraction method based on the matching pursuit (MP) is proposed to extract useful features for the classification of hyperspectral images. The matching pursuit algorithm uses a greedy strategy to find an adaptive and optimal representation of the hyperspectral data iteratively from a highly redundant wavelet packets dictionary. An AVIRIS data set was tested to illustrate the classification performance after matching pursuit method was applied. In addition, some existing feature extraction methods based on the wavelet transform are also compared with the matching pursuit method in terms of the classification accuracies. The experiment results showed that the wavelet and matching pursuit method exactly provide an effective tool for feature extraction. The classification problem caused by curse of dimensionality can be avoided by matching pursuit and wavelet-based dimensionality reduction.