Research On Extraction Method And Application Of Coal Mining Subsidence Area In Radar Differential Interferometric Image Based On U-Net

China has large reserves coal resources,underground coal mining caused by the ground subsidence will have a great impact on the surrounding environment,efficient and accurate coal mining subsidence extraction means for the sustainable development of mining areas play a vital role.The extraction method of coal mining subsidence area based on timing InSAR not only needs complicated data processing,but also interfered with the space-time loss of baseline during SAR image processing,which leads to incomplete identification of coal mining subsidence area,resulting in the extraction area samller than the actual area.On the other hand,when the artificial carrying visual interpretation of the deformation rate graph obtained by the small baseline set method,different interpretation standards between different interpreters will lead to different extraction results.In recent years,deep learning has made great success in the field of remote sensing,but because the shape of coal mining subsidence area is slow and irregular,the characteristics of interference imaging are obvious,which makes it difficult to make great progress in the study of automatic extraction of coal mining subsidence area at home and abroad.Based on these problems,this paper uses differential interference images to establish a sample library for extraction of coal mining subsidence areas and trains the U-Net network.Taking Shanxi Province as an example,the trained model is used to identify and count coal mining subsidence areas.The specific work and conclusions are as follows:(1)Research on the correction method of differential interferogram.Joint ground control points to optimize the interference baseline;use the power-low atmospheric model to adaptively reduce the atmospheric delay error related to elevation.Finally,the baseline error and the altitude-related atmospheric delay error are removed from the original differential interferogram,so that the deformation characteristics of the coal mining subsidence area are more obvious in the interferogram,which provides a data source for subsequent extraction of the coal mining subsidence area.(2)Establish a sample library for extraction of coal mining subsidence areas and train U-Net network.Select 34 sentinel images within the scope of Shanxi Province from November to December from 17-18.After differential interference and baseline optimization,the 17-scene differential interference image is obtained,and Arc GIS is used for visual interpretation,and more than 5000 512×512 are obtained.Large and small samples,establish a coal mining subsidence area to extract a sample set and train the U-Net network to obtain the optimal model weight.The accuracy of the sample is evaluated,and the accuracy of the sample is better than 95%,which proves the reliability of the sample set.The extraction results of differential interference images before and after optimization are compared and analyzed,and the extraction accuracy is improved by 2%after optimization.Compared with the results of the small baseline set,the subsidence area is more than 85%consistent.(3)Statistics on the area of coal mining subsidence in Shanxi Province.Using the trained optimal model,the coal mining subsidence area in the entire Shanxi Province is identified.According to the extraction results,the total area of the coal mining subsidence area in Shanxi Province in the winter of 2019 is about 1,343.91 km~2.