矿产资源的开采在促进区域经济发展的同时，也造成了地表破坏，对生态环境造成了影响。利用遥感技术及时获取矿区土地利用覆盖和生态环境的变化信息，可以辅助矿区生态保护和生态恢复。针对传统变化检测方法中需要提取大量人工设计的图像特征，提出一种改进UNet孪生网络结构。用卷积层代替UNet结构中的池化层，加入中心环绕、双通道图像输入的孪生结构和特征金字塔模块，对遥感影像进行端到端的矿区变化检测。首先，对经过数据预处理的前后两时期遥感影像利用中心环绕的方法进行裁剪，获取图像的多尺度信息，将裁剪后的中心区域和环绕区域的影像分别输入到网络的孪生结构编码端，通过权值共享的孪生结构提取前后两时期影像的差异信息;然后，将同一特征层上的特征进行相减，获取前后两时期影像在不同卷积层上的差值影像并进行特征融合，把融合后的图像送入特征金字塔模块获取图像多尺度上下文信息。最后，将编码端和解码端对应层跳跃连接，进行端到端的预测，得到前后两时期矿区遥感影像的变化二值图像。实验结果表明:改进的变化检测网络方法与传统方法相比可以自动的提取图像的低层特征和高层语义特征，避免了人工提取图像特征的繁琐。此外，在检测结果上，改进的变化检测方法与对比方法相比，总体精度和Kappa系数有明显提高，同时也降低了检测的误检率和漏检率。

The exploitation of mineral resources,while promoting regional economic development,has also caused some damage on land surface and has an impact on the ecological environment. The remote sensing technology for timely ob- taining the change information on land cover and ecological environment in mining areas can be used in practical ap- plications such as guiding ecological protection and restoration of mining areas. Aiming at the need to extract a large number of artificially designed image features in the traditional change detection method,an improved UNet siamese network is proposed. The convolution layer is used to replace the pooling layer in the UNet structure,and the siamese structure with dual channels,the feature pyramid module and central surround module was added. Firstly,the preprocessed remote sensing images of the two periods are cropped by central surround and the multi-scale information of the image is acquired,and the images of the central area and the surround area are respectively input into the encoder part of the network structure. The different information between the two periods is extracted by the weight shared siamese structure,then,the features on the same feature layer are subtracted,and the different images on different convolutional layers are obtained and the feature fusion is performed. The merged image is sent to the feature pyramid module to ob- tain image multi-scale context information. Finally,the corresponding features of the encoder and the decoder are fused by skip connection,and the end-to-end prediction is performed to obtain the change binary image of the mining area remote sensing image in the two periods. The results show that the improved change detection network method can au- tomatically extract the low-level features and high-level semantic features of the image compared with the traditional methods,which avoids the cumbersome manual extraction of image features. In addition,from the detection results,the improved change detection method has a significant improvement in overall accuracy and Kappa coefficient compared with the comparison method,and also reduces the commission error and omission error of the detection result.