Research On Early Warning Method Of ERT Pollution Area Based On Time-space Domain

Electrical Resistivity Tomography(ERT)is one of the important technologies applied in the process of land pollution early warning,restoration and reuse.Among them,land pollution identification and precise determination of the contaminated area are the urgent problems to be solved by the ERT detection system.The current ERT monitoring results for the identification of contaminated areas are mainly concentrated in the spatial domain and lack continuity in the time domain,which leads to poor universality of the identification algorithm.In order to realize real-time monitoring and early warning of pollution,it is not only necessary to improve the accuracy of identification,but also to realize change detection of contaminated sites in the time domain.This thesis focuses on the early warning of ERT pollution areas in the time domain and space domain,involving the collection and construction of ERT monitoring data sets,the detection methods of ERT time domain changes,using theoretical research,simulated site research and actual site research,and carried out the following work:First,the current development status of ERT technology and the application research of image segmentation algorithms are reviewed.On this basis,the key problems and technical difficulties faced by the recognition of contaminated areas based on ERT detection data are explained.Secondly,a simulation model of heavy metal contaminated sites with resistivity method was established to simulate the monitoring data of heavy metal contaminated sites at different times and in different contaminated areas,providing data support for the development of timedomain change detection algorithms;And carried out the actual site ERT detection,and constructed the site pollution ERT detection data set used in this article.Then,an ERT time-domain change detection method based on discrepancy matrix and sparse autoencoder is proposed.This method first uses the difference operator and the ratio operator to describe the data difference.By comparing the value in the difference matrix with the set threshold,it is judged that the data set measured at time k+1 on the same measurement line is relative to the data measured at time k.Whether the collection is abnormal;Aiming at the problem of insufficiently accurate threshold selection,the threshold is optimized by fusing neighborhood features and sparse autoencoders.The results show that the optimized threshold is closer to the actual situation of the site than the threshold before optimization.Finally,a method for identifying polluted areas in the spatial domain based on the combination of covariance clustering and semi-supervised support vector machine algorithm is proposed.Aiming at the problem of inaccurate identification when there is media stratification,a clustering method based on covariance is proposed to solve the problem of the influence of media stratification on the identification of contaminated areas;On this basis,in order to improve the recognition accuracy in the depth direction,a semi-supervised training method is introduced,which combines the semi-supervised support vector machine(S4VM)and semisupervised Ladder Network(SSLN)are respectively integrated with C-FCM.The experimental results show that this method not only overcomes the influence of medium layering,but also solves the problem of inaccurate recognition when the difference between the background resistivity value and the contaminated area resistivity value decreases,and improves the recognition accuracy of the algorithm.