Overall framework of MSTLF. (a) MSTLF mainly includes two essential components, i.e., the Multi-Scale Texture Extractor (MSTE) and the Aggregation Module (AM). We denote the selection of the aggregation rule with the logic operators demux and mux. (b) MSTE module extracting a textural representation from the input images using a texture descriptor extracted from the GLCMs at different spatial and angular scales. (c) Dynamic aggregation by Self-Attention mechanism (SA) that combines the extracted representation into a scalar loss function