Classification Of Debris Flows And Valleys In Remote Sensing Images Based On Deep Migration Learning

Debris flow is a natural disaster that is dominated by heavy rainfall and frequently occurs in areas with steep terrain.It has the characteristics of sudden outbreak,wide impact range,and strong destructiveness.Once a mudslide disaster occurs,it will cause serious loss of life and property to the surrounding residents.How to accurately identify debris flow valleys and reduce the threat of debris flow disasters has become an urgent problem to be solved.Remote sensing technology has the advantages of high efficiency,low cost,high temporal and spatial accuracy,and dynamic monitoring.It has become an effective method to identify the disaster-generating area of debris flow after the traditional field survey method.The rise and application of deep learning has achieved significant results in the recognition and classification of remote sensing images,which has further promoted the development and popularization of remote sensing images.However,there is still a problem of insufficient image data required for deep learning.In response to this problem,this paper improves the VGG16 convolutional neural network,combines the transfer learning method,selects remote sensing data of debris flow valleys in Nujiang Lisu Autonomous Prefecture for training and verification,and proposes a deep transfer learning-based remote sensing image of debris flow valley classification model.The specific research work and content of this article are as follows:1.It briefly introduces the current research background,significance and current research status in the field of debris flow disasters,remote sensing images and deep migration learning,and outlines the relevant knowledge and theories required by the experimental research.And use Arc GIS tools to make a simple analysis of the distribution of debris flow disasters in the study area.2.Obtain the valley image data set in DEM and remote sensing images.In the MATLAB programming environment,the region growing algorithm is used to perform rough river recognition processing on the DEM image,and a rectangle is constructed on the recognized DEM image,and the valley image of the four bands of the remote sensing image is divided by the corresponding latitude and longitude to construct the follow-up Data set required for the experiment.3.Based on the VGG16 convolutional neural network,explore the factors that affect the accuracy of the model.First,carry out classification experiments on the debris flow valleys in remote sensing images under different channels,and determine the near-infrared channel of remote sensing images with the highest classification accuracy.Then use different Dropouts to analyze the classification results of the model,and get the most suitable Dropout value for the model.4.Construct a classification model of debris flow valleys in remote sensing images based on deep migration learning.First,the debris flow gully remote sensing data set is enhanced.Based on the VGG16 model,the second fully connected layer is removed,the third fully connected layer is optimized,and Dropout is introduced to avoid overfitting.Then combined with the migration learning strategy,the convolutional layer of the first three layers of convolutional network is frozen,and the remaining convolutional layer,the top network and the Softmax layer are fine-tuned to optimize the network parameters.Finally,the confusion matrix,single-class and average classification accuracy are used to evaluate the model.The experimental results show that the improved model can converge faster and classify debris flow valleys more accurately.The kappa coefficient of the classification result is 0.8762,which has higher consistency.The average classification accuracy can reach 94.74%,which is higher than that of the unimproved model.The VGG16 model increased by 3.07%.5.Develop the debris flow valley classification system at the B/S end.The debris flow valley classification model based on deep migration learning,following the software engineering development specifications and processes,researches and develops the system,and realizes the functions of uploading and identifying debris flow valley images.