Combining high-resolution remote sensing data with deep learning methods,the effective extraction of land damage information in open-pit mines can be realized,and the mastery of the current situation and changes of mine geo-logical environment can be improved.Based on the images data of GF-2 satellite,the U-Net model was used to extract the land damage information of open-pit mines in three typical ore concentration areas in Hubei Province.According to the accuracy evaluation results of the model,the more data contained in the data training set,the better the information ex-traction effect;when the number of backbone model layers is too deep,over-fitting will occur,which will reduce the accu-racy of information extraction results.After comprehensive consideration,the rotation angle of the data training set is set to 30°,and the backbone model is set to ResNet34.Finally,a better information extraction effect is obtained,which verifies the feasibility of U-Net model for land damage information extraction in open-pit mines.
关键词
深度学习/卷积神经网络/U-Net/ResNet/露天矿山土地损毁/信息提取
Key words
deep learning/convolutional neural networks/U-Net/ResNet/land damage in open-pit mines/information extraction
维普
万方数据