Abstract: As a method of deep learning interpretability, class activation mapping (CAM) is efficient and convenient for extracting geographic objects supervised by image-level labels. However, in ...