基于DeepLabv3+与GF-2高分辨率影像的露天煤矿区土地利用分类  

Recognition of land use on open-pit coal mining area based on DeepLabv3+and GF-2 high-resolution images

作　　者：张成业 李飞跃 李军[1,2] 邢江河[1] 杨金中 郭俊廷 杜守航 ZHANG Chengye;LI Feiyue;LI Jun;XING Jianghe;YANG Jinzhong;GUO Junting;DU Shouhang(College of Geosciences and Surveying Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China;State Key Laboratory of Coal Resources and Safe Mining,China University of Mining and Technology(Beijing),Beijing 100083,China;China Aero Geophysical Survey and Remote Sensing Center for Natural Resources,Beijing 100083,China;State Key Laboratory of Water Resource Protection and Utilization in Coal Mining,Beijing 102209,China)

机构地区：[1]中国矿业大学(北京)地球科学与测绘工程学院,北京100083 [2]中国矿业大学(北京)煤炭资源与安全开采国家重点实验室,北京100083 [3]中国自然资源航空物探遥感中心,北京100083 [4]煤炭开采水资源保护与利用国家重点实验室,北京102209

出　　处：《煤田地质与勘探》2022年第6期94-103,共10页Coal Geology & Exploration

基　　金：煤炭开采水资源保护与利用国家重点实验室2020年开放基金课题(GJNY-20-113-14);中央高校基本科研业务费项目(2022YQDC08,2022JCCXDC04)。

摘　　要：遥感与深度学习为及时掌握露天煤矿区土地利用情况提供了高效率的技术手段。基于国产高分二号(GF-2)卫星高分辨率遥感影像,利用深度学习DeepLabv3+模型实现露天煤矿区土地利用识别,并与U-Net、FCN、随机森林、支持向量机、最大似然法等方法进行对比。首先,制作高分辨率影像样本数据,通过敏感性测试确定适合研究区露天煤矿场景的样本最佳裁剪尺寸和方式;然后,训练深度神经网络DeepLabv3+模型,进行土地利用识别实验;最后,比较不同方法的识别结果。结果表明:研究区露天煤矿场景下的样本最佳裁剪尺寸为512像素×512像素,最佳裁剪方式为随机裁剪。采用的DeepLabv3+模型对露天煤矿区土地利用识别的总体精度、Kappa系数分别为80.10%、0.73,均优于U-Net、FCN、随机森林、支持向量机、最大似然法等方法的识别精度。DeepLabv3+模型的识别速度与上述5种方法保持在同一数量级,验证了DeepLabv3+模型和GF-2卫星影像在露天煤矿区土地利用识别中的可行性,对露天煤矿区生态环境监测与修复规划具有重要意义。A highly efficient means is provided by remote sensing and deep learning to keep tracking of land use in open-pit coal mining area.Based on the high–resolution images from the domestic GF-2 satellite,a DeepLabv3+model was utilized to achieve recognition of land use on open-pit coal mining area.In addition,a comparison was made among Deeplabv3+,U-Net,FCN,Random Forest,Support Vector Machine,and Maximum Likelihood Method.Firstly,samples data from high-resolution images were produced and sensitivity tests were conducted to determine the optimal cutting size and mode of the sample.Then,the deep neural network model(DeepLabv3+)was trained for conducting experi-ments of recognition of land use.Finally,the recognition results of different methods were compared.The results show that the optimal cutting size of the sample on the open-pit coal mining of the study area is 512 pixel×512 pixel.The op-timal cutting mode of the sample is random cropping.The overall accuracy and Kappa coefficient of the DeepLabv3+for recognition of land use on open-pit coal mining area are 80.10%and 0.73,respectively,which are better than the re-cognition accuracy of the U-Net,FCN,Random Forest,Support Vector Machine and Maximum Likelihood Method.The DeepLabv3+is kept in the same order of magnitude as the above five methods.The feasibility of DeepLabv3+model and GF-2 images in recognition of land use on open-pit coal mining area is verified,which is important for monitoring and restoration of eco-environment on open-pit coal mining area.

关 键 词：露天煤矿区 土地利用 高分辨率影像 深度学习 神经网络 高分二号卫星 自动识别 识别精度 

分 类 号：P237[天文地球—摄影测量与遥感] TP753[天文地球—测绘科学与技术]