WSCDNet: A Window Structural Similarity Guided Deep Feature Recalibration Method for Remote Sensing Image Change Detection

With the advancement of deep learning, numerous promising and effective neural network models have emerged in the field of remote sensing image change detection, finding widespread applications in urban planning and environmental monitoring. In this study, we propose a detection framework predicated on the EfficientNet-B5 architecture. To address the issues of insufficient communication and feature confusion between bi-temporal images, we incorporate information interaction modules designed with aggregation and distribution mechanisms at strategic intervals. This enables the model to better capture temporal information during feature extraction. Furthermore, we use a dense ladder -shaped fusion method, making full use of the multi-level feature maps in the feature pyramid. In the decoding stage, we innovatively designed a difference calculation method based on the structural similarity index between windows, recalibrated the deep features extracted by the network, and then decode layer by layer. This approach can effectively alleviate the problem of susceptibility to noise and image style in pixel-by-pixel differential calculations. Our model was tested on four remote sensing datasets: LEVIR-CD, SYSU-CD, PX-CLCD and CLCD, and all showed excellent performance. The code and pre-trained model are available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/Vgrant0/WSCDNet.git</uri>.