Building change detection using the parallel spatial-channel attention block and edge-guided deep network

Building change detection in high-resolution satellite images plays a special role urban management and development. Recently, methods for building have been greatly improved by developing deep learning. Although learning technologies, especially Siamese convolutional neural networks, successful popular, they usually problems extracting features that are not discriminative enough also cause the loss of shape details at edges. To address these problems, dual-branch network parallel spatial-channel attention mechanism were suggested to extract spatial spectral dependencies more features. The unit measured rich context local features, distinction between changed objects backgrounds was increased using channel module adjusted weight channels acted as selection process. Mixing two attentions mode made practical, useful information learned robustly. Moreover, dual function proposed which edge-based consistency constraints used first part converge edges training predicted data. weighted binary cross-entropy added second function. method implemented on remote sensing datasets, results evaluated with state-of-the-art methods. With model, F1-score 2.43% 1.83% respectively.