Tailing Pond Identification Monitoring And Risk Prediction Based On Deep Learning

The tailings pond is a necessary facility for the beneficiation of mining companies,and it is also a major source of danger to the surrounding environment.Comprehensive and detailed identification and extraction of the risk information of the tailings pond is a prerequisite for preventing tailings pond accidents and emergency work of the tailings pond.Therefore,research and exploration on tailings pond identification monitoring and environmental risks are of great significance.Due to the widely distributed,large-scale,and complex structure of tailings reservoirs,traditional ground surveys are difficult,inefficient,and inaccurate.With the rapid development of remote sensing space-based earth observation technology and geographic information technology,the use of remote sensing technology and geographic information technology methods instead of traditional methods for surface environment observation and target recognition has gradually become a trend and has attracted widespread attention.Based on the state-of-art technology of machine learning-deep learning method,this paper I used the basic principles of target recognition and semantic segmentation to identify and segment the tailings pond under the high-resolution remote sensing image,and I also selected the method of hydrological analysis to predict the risk of the research area.By considering the surface environment of Beijing,Tianjin and Hebei in the study area,the high-resolution remote sensing image and basic geographic data of gaofen-1 and Google Earth are chosen as the basic data sources to carry out the accurate identification of tailings pond and determine the distribution location of tailings pond.Then,the specific range is selected and combined with arc hydro model to simulate the risk of dam break,runoff distribution,and to analyze the safety of the dense area of tailings pond.The process chain can be divided into four steps.Firstly,a tailings pond sample set for target detection is made by analyzing the texture,color,shape and size of tailings pond on the remote sensing image.Secondly,I propose a multi-scale fusion target detection algorithm,MSF＿SSD,based on SSD algorithm,which is realized by adding deconvolution module and connection module on the basis of the original SSD target detection network.Next,based on the target detection results,the semantic segmentation samples of tailings pond are made,and the structure segmentation of tailings pond is realized by PSPnet algorithm,and the internal structure of tailings pond dam body and reservoir area are obtained.Finally,on the basis of the vector files of dam body and reservoir area,through hydrological analysis,remote sensing and geographic information technology are used to extract the upstream catchment surface of tailings pond and the possible accident runoff.The results and conclusions are summarized as follows:(1)The tailing pond sample set is generated Using the GF-1 remote sensing image and Google Earth 16 level image,which provides the basis for the training,testing and verification of tailing pond deep learning framework.(2)The MSF＿SSD target detection algorithm is studied and improved,and it is successfully applied to the extraction of Beijing Tianjin Hebei tailings pond.Under the premise of ensuring the accuracy and recall rate,the high automatic detection of tailings pond target is realized.(3)Applied the PSPnet algorithm to the classification of tailings pond,and the detected tailings pond is divided into three categories: dam body,reservoir area and background.The classification accuracy of MIo U is 98.4%,which is the premise of safety evaluation of tailings pond.(4)Comprehensive utilization of hydrological analysis in spatial analysis technology,combine the DEM data of the study area and the vector data of the dam body and reservoir area,determine the upstream catchment surface of the tailings pond and the possible path of the dam failure accident of the tailings pond to extract;carry out the buffer zone analysis of the dam failure path and the land use map obtained from the random forest to analyze the disaster area of the study area,etc.Research results can be used to analyze the loss of ground features and the area of influence caused by the dam break of tailings pond,to meet the monitoring and risk assessment of a large range of tailings pond,to improve the risk management level and emergency response capacity of tailings pond,and to provide theoretical basis for relevant departments to make decisions.