A new feature extraction method for remote sensing image was proposed based on a novel wavelet transform algorithm. Different from binary wavelet transform partition the frequency domain by constant Q criteria

the method can partition the frequency domain freely through setting the ratio of bandwidth of adjacent wavelet. Feature extraction based on discrete cosine transform of the wavelet energy was performed. The results of C-means clustering and RBF neural networks classification experiments show that

the proposed feature of wavelet transform can effectively describe spectral curve

and has better classification rate than the traditional wavelet transform algorithm.