Classification of Soil Texture Based on Wavelet Domain Singular Values

Abstract

Singular Value Decomposition (SVD) based novel approach using wavelet packet transformation is proposed for classification of soil textures. A procedure for classifying the textures in the presence of additive white Gaussian noise is introduced and this procedure is experimentally validated. The proposed approach extracts features such as energy, entropy, local homogeneity and min-max ratio from the singular values of wavelet packet transformation coefficients. A modified form of Probabilistic Neural Network (PNN) called Weighted PNN (WPNN) is employed for performing the classification. Compared to probabilistic neural network, WPNN includes weighting factors between pattern layer and summation layer of the PNN. Experiments have been carried out to test the performance of the proposed approach in terms of classification rate at various Peak Signal-to-Noise Ratio (PSNR), various number of training texture images, various levels of wavelet packet transformation, and various feature set dimensionality. Experimental results showed superiority of the proposed approach over wavelet domain Gray Level Co-Occurrence Matrix (GLCM) based approach, wavelet domain SVD model based approach and Hidden Markov Model(HMM) based approach.