Wavelet-Feature Greedy Clustering for Remote Sensing Identification

The wavelet transform can capture the abrupt features by accurately describing the spectral abrupt features. In contrast, the wavelet correlation coefficient can evaluate the similarity or distance between remotely sensed samples in the superdimensional space. Remote sensing identification is used to search the pixels with the lowest correlation, and the best clustering centers with the lowest redundancy can be achieved through greedy searching within the boundaries of wavelet correlation coefficients. This wavelet-feature greedy clustering algorithm is unsupervised without knowing the number and distribution of classes corresponding to various land covers. Its favorable identification results were achieved on the AVIRIS hyperspectral images.