Intelligent Extraction of Remote Sensing Image Change Patches Based on Deep Learning and Human–Computer Collaboration in Coal Mine Surface Areas

Deep learning-based change detection methods are sensitive to illumination and registration errors and require a massive amount of samples during training, which is challenging to acquire in tasks related to coal mining. Hence, this study develops intelligent extraction methods and technical processes for change patches of surface remote sensing images of a coal mine area using Cesium, a Web-based geographic information system (WebGIS), and deep learning theory. Specifically, the proposed scheme utilizes a deep learning model with improved fusion spatiotemporal feature matching capabilities and a large convolutional kernel spatiotemporal attention to extract remote sensing image change patches intelligently. Then, the extracted change detection patches are visualized, managed, and edited by the constructed WebGIS platform, supported by multi-user human-computer collaboration, to improve the correctness and accuracy of the detection results. Finally, the manually edited results are deposited into the sample library as new samples and input back to the deep learning model as a new sample for autonomous learning to improve the model's accuracy further. The practical application of the developed method demonstrates that the proposed deep learning algorithm attains a better detection accuracy than similar algorithms, effectively utilizing the enhanced autonomous learning accuracy of the sample library to compensate for insufficient samples. The proposed method incorporates collaborative perception, interaction, and decision-making between humans and machines, providing a theoretical and practical reference for practical engineering applications of re-mote sensing image change detection.