Research On Automatic Identification Model Of Debris Flow Based On Neural Network

Taking the north line of the Sichuan-Tibet Highway as the research area,the paper discriminates the debris flow ditch in the study area by means of remote sensing interpretation and field investigation,and analyzes the development characteristics of debris flow in the area.On this basis,a debris flow sample dataset for neural network training is constructed.Then,four kinds of neural network frameworks are established,and the Squeeze-and-Excitation channel attention mechanism is added respectively;using the constructed data set,the training of the neural network model is completed based on the matlab2021 a platform.The following results were obtained through the above research:1.Collect the environmental background information of debris flow formation in the study area and the survey data of existing debris flow disasters;analyze the development characteristics of debris flow accumulation fans,provenance and vegetation in the area,and lay the foundation for the establishment of subsequent neural network training data sets.2.Establish the sign of remote sensing interpretation of debris flow,and carry out remote sensing interpretation of the watershed in the study area;verify and revise the interpretation signs and results through on-the-spot investigation.On this basis,a remote sensing image sample dataset for the training of the debris flow neural network model is constructed,and the dataset is processed to form a dataset sample that can be trained.3.Analyze the structure of the neural network,based on the convolutional neural network,and use resnet50,resnet101,inceptionv3 and xception as the basic network framework to establish an automatic identification model of debris flow that conforms to the scale of the sample data set;then,the SE channel attention The mechanism is applied to the above network to establish a new network model.4.Select appropriate training parameters to train and verify the obtained network model.The results show that each model of the neural network can identify debris flow,and the accuracy rate is greater than 74%,and the accuracy rate can be further improved after the Squeezeand-Excitation channel attention mechanism is added.Among them,the important areas determined by the network model for the automatic identification of the incoming debris flow are different,the discriminative effect of resnet50 is poor,and the discriminative effect of se-resnet50 is better.The above research shows that neural network can be applied to remote sensing interpretation of debris flow to achieve the purpose of automatic interpretation.This study provides a new idea for remote sensing interpretation of debris flow.