Multi-Scale LBP Texture Feature Learning Network for Remote Sensing Interpretation of Land Desertification

Land desertification is a major challenge to global sustainable development. Therefore, the timely and accurate monitoring of land status can provide scientific decision support for control. The existing automatic interpretation methods are affected by factors such as “same spectrum different matter”, “different same object”, staggered distribution areas, wide ranges ground objects. We propose an method remote sensing that incorporates multi-scale local binary pattern (MSLBP) spectral features based on above issues. First, convolutional LBP feature extraction network designed obtain spatial texture images fuse them with enhance representation capability model. Then, considering continuity kind objects in space, we adaptive median filtering process probability map extreme learning machine (ELM) classifier output improve classification accuracy. Four typical datasets were developed using GF-1 multispectral imagery Horqin Left Wing Rear Banner study area. Experimental results four show proposed solves problem ill omission classifying desertification, effectively suppresses effects “homospectrum” “heterospectrum”, significantly improves accuracy desertification.