Research On Deep Learning Extraction Method In Open Mining Area Based On Multi-source Remote Sensing Images

Mineral resources are an important material basis for China's social and economic development The use of remote sensing technology for information extraction and monitoring of open pit mining areas has become an important means to solve the natural environment problems of mines.This paper took Tongling City,Anhui Province as the research area,used high-resolution remote sensing images as experimental data,enhanced the feature extraction of different types of remote sensing image open-pit mining areas by improving the full convolutional neural network with densely connected blocks.The open-pit mining area extraction model for multi-source remote sensing image data was trained through the construction of the open-pit mining area sample library.The automatic extraction of the open-pit mining area in the study area was realized,and the comparative experiment was used to verify.The research work in this paper is as follows:(1)The domestic GF-1,GF-2 and GoogleEarth remote sensing images in the study area were collected and preprocessed.The quality and interpretability of the images was improved by radiometric calibration,atmospheric correction,orthorectification,image fusion and other methods;(2)The automated batch production process of remote sensing image deep learning sample library in open pit mining area was designed,and a set of multi-source and multi-scale remote sensing open-pit mining training sample library including GF-1,GF-2 and GoogleEarth images was constructed;(3)Based on the improved full convolutional neural network with densely connected blocks,an intelligent extraction model for open-pit mining areas adapted to multi-source remote sensing images was constructed and trained to realize automatic extraction of open-pit mining areas.The extraction of features from open source areas of different data sources was enhanced;(4)In order to verify the advantages of this model to the boundary extraction of open-pit mining areas,compared with traditional classification methods(maximum likelihood method,decision tree classification,support vector machine)and other deep learning methods(Deep lab,U-Net,Segnet),the method of this paper has some improvement in pixel-based and object-based evaluation,in which pixel precision PA:0.9767,intersection ratio IoU:0.7207,comprehensive evaluation index F1:0.8377,Kappa coefficient:0.8251,recall rate:0.9130,missed alarm rate:0.0870,false alarm rate:0.5333:(5)In order to verify the applicability of the model to other data source data,the GoogleEarth image of Tongling District of Anhui Province and Wushan District of Ruichang County,Jiujiang City,Jiangxi Province was extracted from the open-pit mining area.The experimental results prove that the model has good applicability to GoogleEarth images in different regions and the practicability of the model was enhanced;(6)Using the deep extraction learning model of multi-source remote sensing image open-pit mining area,combined with other image processing techniques,the integrated extraction process of boundary extraction in open-pit mining area was studied,and a set of automatic extraction tools for the boundary of open pit mining areas was designed and developed through Python;(7)Taking the open-pit mining area under the jurisdiction of Tongling City of Anhui Province as an example,GF-1 of the Tongling City area with 5 scenes from 2013 to 2017 was used to study the temporal and spatial variation characteristics of the open-pit mining area by using the superposition analysis in the traditional spatial analysis,and the time and space change monitoring of the open pit mining area was realized.