Multiscale Convolution Fusion Filters with Heterogenous Remote Sensing Images for Land Cover Change Detection

Deep learning approaches are becoming increasingly prominent in the field of land cover change detection (LCCD) utilizing Remote Sensing Images (RSIs). But present networks frequently focus primarily on deep feature learning, ignoring the critical role of multiple feature attention and fusion algorithms in performance of detection. To address the issue, this research proposes a Multiscale Convolution Fusion Filters (MCFFs) in RSI for LCCD. A Compound Loss Function (CLF) is created to efficiently balance the contribution of features at different levels of backpropagation error. A Convolutional Block Attention Module (CBAM) is also being developed to detect more unique features, which will aid in image segmentation into modified and unchanged areas. Experimental assessments of four datasets show that the suggested MCFF-CLF-CBAM strategy outperforms existing methods. The experiments support the feasibility and superiority of MCFF-CLF-CBAM, with quantitative results showing significant improvements across all assessment measures.