Research On Intelligent Identification Method Of Moisture Content Of Loess On Convolutional Neural Networks

With the advancement of my country's major loess engineering construction,a series of geological disasters have been triggered,which has brought serious threats to the safety of people's lives and properties.Among the preventive measures for many geological disasters,in-situ accurate detection of loess moisture content in deep layers is an effective means to reveal the disaster mechanism of major loess engineering and disaster warning.Therefore,this thesis proposes an intelligent recognition method based on convolutional neural network(CNN)to realize the detection of deep layers.Long-term uninterrupted detection and identification of in-situ loess moisture content.First of all,to address the problem of lack of data sets when training CNN,this thesis has built an indoor experimental platform that fully simulates the environmental conditions of deep in-situ detection of loess holes,collected a large number of loess images with different moisture content,and constructed a neural network training platform.data set.The loess water content image data set constructed in this thesis is sampled in Yan'an,Lantian and Lanzhou and other typical areas of major loess project construction to ensure the extensiveness and representativeness of the constructed data set.Secondly,to test the feasibility of the CNN based on the loess image to identify the water content,this thesis builds a deep CNN model,and uses the loess water content image data set for training,and the test accuracy rate is 99%.In order to reduce the weight parameters of the network and improve the calculation speed of the network,a lightweight convolutional neural network model is built.While ensuring that the lightweight convolutional neural network model achieves 99% test accuracy,the weight parameters of the network are reduced by more than 10 times,and the calculation speed is increased by 2to 3 times.The lightweight convolutional neural network model built in this thesis provides feasible algorithmic support for identifying the moisture content of loess based on mobile detection equipment.Finally,to test the reliability of the method of intelligently identifying the moisture content of loess based on the CNN proposed in this thesis,a full-scale test platform was built for experimental verification,and the error of the test results was analyzed and inferred.The method of identifying the moisture content of loess based on the convolutional neural network proposed in this thesis enriches the research methods of geological exploration,overcomes the technical bottleneck of the existing loess moisture content detection technology that cannot be used for deep in-situ and long-term uninterrupted detection,and is used to clarify the special structure loess engineering slope strength change law and deterioration process,reveal the disaster mechanism of major engineering loess slopes to provide technical conditions,and through real-time observation of changes in deep soil moisture content,the key data for geological disaster forecasting is enriched.