MMFN: Multi-Scale Mask Fusion Network For Remote Sensing Change Detection

In recent years, the integration of CNN and Transformer network architectures has played a pivotal role in the field of change detection in remote sensing. The most recent research has showcased remarkable advancements. However, we have encountered several challenges concurrently. Traditional methodologies predominantly rely on conventional semantic segmentation approaches, overlooking the inherent geometric features unique to the task of change detection, particularly those related to buildings. Secondly, the extensive use of convolutional operations can mask the inherent spatiotemporal characteristics of the data. Addressing the aforementioned issues, this paper introduces a multi-scale fusion method named MMFNet, designed to extract features at different levels.