矿井数量的增加以及地下煤炭资源的高强度开采造成生态环境破坏、地面产生塌陷以及人民生命财产安全受到威胁等一系列问题。因此，为确保矿区生态环境与经济的可持续发展，对矿区地表形变进行及时有效的沉陷监测与灾害防治变得尤为关键。合成孔径雷达干涉测量技术（InSAR）凭借其受天气因素限制小以及覆盖范围广的优势，广泛应用于矿区地表形变监测。由于煤层的埋藏深度受地质条件与构造影响，不同埋深煤层开采造成的地表沉陷盆地在InSAR监测结果上表现也不同。本文应用SBAS-InSAR技术分别对浅埋、中-深埋煤层进行开采沉陷监测，探究不同埋藏深度下煤层开采造成的沉陷盆地的表现形式与形变规律，结合矿区实际情况采用不同的方法实现浅埋煤层开采沉陷盆地参数提取与中-深埋煤层形变预测，并基于两种不同埋深下的InSAR监测结果，利用深度学习模型对沉陷区域进行精确识别和范围提取，为矿区地表形变监测提供了一种新的技术手段。本文主要研究内容如下：

（1）针对浅埋煤层，以陕西省神木市红柳林煤矿某工作面作为研究对象，采用SBAS-InSAR技术对覆盖研究区SAR影像进行时序处理，获取监测时段内矿区时间序列累积形变量和开采沉陷范围。受限于InSAR技术自身特性以及浅埋煤层开采造成的地表移动具有沉降量级大和形变速率快等特点，导致InSAR监测结果中沉陷盆地中心表现为失相干。因此，基于InSAR监测结果中下沉盆地边界高相干点形变值与工作面走向中线实际GNSS监测数据，并结合采煤工作面具体信息，利用麻雀搜索算法反向求取沉陷盆地概率积分模型参数，得到模拟的沉陷盆地轮廓。将实际监测值与模型结果进行对比，得到沉陷盆地走向中线与盆地边界的均方根误差分别为40.9mm与9.6mm，表明该模型能够较好反映地表沉陷盆地的实际情况，有效弥补了InSAR技术在大梯度沉陷监测中的不足。

（2）针对中-深埋煤层，以贵州省盘州市某矿区为例进行研究，采用SBAS-InSAR技术获取该区域研究时段地表累积沉降量与年均形变速率结果，并结合该地区矿产资源分布图进行一致性检验。提取典型区域的剖线与特征点形变结果，进一步分析中-深埋煤层地表沉陷盆地动态变化特征。由于缺少该矿区工作面详细开采信息，因此无法获得开采沉陷参数。本文将时序InSAR结果与WOA-SVR算法相结合，建立矿区开采沉陷预测模型，并与传统SVR模型和GM（1,1）模型预测结果进行对比。结果表明，WOA-SVR最大相对误差均控制在2%以下，RMSE与MAE较传统模型分别最大提升93%与92%，拟合结果较好，有利于进一步分析沉陷盆地动态变化趋势，为矿区开采沉陷预测提供一种可行方法。

（3）采用Stacking-InSAR技术获取神木市与盘州市两个研究区的长时间地表形变速率，基于此结果，参考阈值分割法进行沉陷区域的样本及标签制作。采用PSPNet、U-Net和DeepLabV3+三种经典语义分割模型对沉陷区进行识别以及范围提取。结果表明，三个模型都能有效地从InSAR监测结果中提取出沉陷区的图像特征。基于准确率（P）、召回率（R）和平均交并比（mIoU）这三个关键性能指标，发现DeepLabV3+模型在该场景下性能表现最佳。表明InSAR技术结合语义分割模型能够实现沉陷区域的自动化识别与范围提取，验证了此方法在矿区沉陷智能监测和管理领域的实际应用潜力。

The increase in the number of mines and the high-intensity exploitation of underground coal resources have led to a series of problems such as ecological damage, ground subsidence and threats to the safety of people's lives and property. Therefore, in order to ensure the sustainable development of the ecological environment and economy of the mining area, Timely and effective subsidence monitoring and disaster prevention and control of surface deformation in mining areas have become particularly crucial. Synthetic aperture radar interferometry (InSAR) is widely used in monitoring surface deformation in mining areas due to its advantages of being less restricted by weather factors and having a wide coverage. Since the effect of geological conditions and tectonics on the depth of coal seams, the surface subsidence basins caused by the mining of coal seams with different depths are also different in the InSAR monitoring results. In this paper, SBAS-InSAR technology is applied to monitor the mining subsidence of shallow and medium-deep buried coal seams respectively, aiming to explore the manifestation and deformation law of subsidence basins caused by mining of coal seams at different depths of burial, which extracts the parameters of subsidence basins of shallow-buried coal seams and predict the deformation of medium-deep-buried seams with different methods in combination with the actual conditions of the mines. According to the results of the two different depths of burial InSAR monitoring, a deep learning model is used for the prediction of subsidence basins in shallow-buried seams. Based on the results of InSAR monitoring at two different burial depths, the deep learning model is utilized to accurately identify and extract the range of the subsidence area, which provides a new technical means for the monitoring of surface deformation in the mining area. The main research content of this paper is as follows:

(1) A working field in Hongliulin mining area, Shenmu City, Shaanxi Province, is regarded as the research subject for shallow buried coal seams, and SBAS-InSAR technology is adopted to process the SAR images covering the research area in time sequence to obtain the cumulative deformation of the mining area and the extent of the mining subsidence during the monitoring period. Due to the characteristics of the InSAR technology and the surfacemovement caused by mining of shallow buried coal beds, which is characterised by large subsidence magnitude and fast deformation rate, the centre of the subsidence basin is out of coherence in the InSAR monitoring results. Therefore, based on the deformation values of the high coherence points of the subsidence basin boundary in the InSAR monitoring results and the authentic GNSS monitoring data of the strike centre line of the working face, and combining with the specific information of the coal mining face, the sparrow optimization algorithm is used to inversely extract the parameters of the probability integral model of the subsidence basin, and get the simulated contour of the subsidence basin. Comparing the actual monitoring values with the model results, the root-mean-square errors of the centre line of the strike of the subsidence basin and the boundary of the basin are 40.9mm and 9.6mm respectively, which indicates that the model can greatly reflect the authentic situation of the subsidence basin on the surface, and effectively bridge the gaps of the InSAR technology in the monitoring of large-gradient subsidence.

(2) As for the medium-deep buried coal seam, a mining area in Panzhou City, Guizhou Province is taken as a sample for the study, and SBAS-InSAR technology is used to obtain the results of the cumulative surface subsidence and the annual average deformation rate in the study period in the area, and the consistency test is carried out by combining with the distribution map of the mineral resources in the area. The results of profile line and characteristic point deformation in typical areas are extracted to further analyse the dynamic change characteristics of the surface subsidence basin of the medium-deep buried coal beds. As the mining subsidence parameters cannot be obtained due to the lack of detailed mining information of the working face in this mining area, the time-series InSAR results are combined with the WOA-SVR algorithm to establish a mining subsidence prediction model for the mining area, and compared with the prediction results of the traditional SVR model and the GM(1,1) model. The results show that the maximum relative errors of WOA-SVR are controlled below 2%, and the RMSE and MAE are improved by 93% and 92% respectively compared with the traditional model, which is a better fitting result, and is conducive to further analysing the dynamic trend of the subsidence basin and providing a feasible method for the prediction of mining subsidence in the mining area.

(3) Stacking-InSAR technique is adopt to obtain the long time surface deformation rate in the two study areas of Shenmu City and Panzhou City, based on which the samples and labels of subsidence areas were made concerning the threshold segmentation method. Three classical semantic segmentation models, PSPNet, U-Net and DeepLabV3+, are used to identify the subsidence areas as well as range extraction. The results show that all three models can effectively extract the image features of sunken areas from InSAR monitoring results. Based on the three key performance indicators, accuracy (P), recall (R) and average cross-to-union ratio (mIoU), it was found that the DeepLabV3+ model performed the best in this scenario. It presents that the InSAR technology combined with the semantic segmentation model can achieve the automated identification and range extraction of subsidence areas, which verifies the practical application potential of this method in the field of intelligent monitoring and management of mine subsidence.