A Material Spectral Feature Degradation Characterization Model Based on Weighted Joint Regression Network

Abstract

A characterization model of material spectral feature degradation based on weighted joint regression network was proposed. This model fully combines the advantages of several classical regression networks and takes full account of L1 and L2 regularization. It overcomes the phenomenon that the model parameters obtained by Lasso regression are approaching 1 or 0, which leads to too many features being sparse to 0, and Ridge regression has insufficient regularization degree and the regression coefficient decays too slowly. It not only ensures the correctness of the calculation of training data, but also ensures the generalization ability of test data, and takes into account the fitting ability and prediction ability of the model. The test results show that, based on the verified weighted joint regression network, the degradation law of the spectral image fusion degree between the material sample and the desert natural environment is modeled, and the correlation coefficient is better than 0.99.