Application of Non-Negative Sparse Matrix Transformation in Hyperspectral Analysis

A variety of pictures in hyperspectral fields requires a reduction in dimensionality, which often needs unique algorithms such as principal component analysis and minimum noise fraction (MNF). This article investigates the improved method of non-negative sparse matrix transformation based on the maximum likelihood covariance estimation and the Frobenius norm to better achieve dimensionality reduction. Non-negativity is presented based on the sparse matrix, which reduces the calculation time and improves efficiency. In order to verify the non-negative sparse matrix transforms (n-SMT) algorithm, samples eroded by disease were selected in the experiment and classified to identify the different parts of leaves after dimension reduction. Besides the n-SMT method, the MNF algorithm is also applied to all the samples. This article compares the two algorithms' operating time and verifies the accuracy of classification after the n-SMT algorithm.