Research On Intelligent Detection Method Of Color Steel House Along Railway Line Based On UAV Inspection Image

Rail transit has become the main form of people’s travel and freight transport in China,and its security risks will increase with the increase of operating time and mileage.As a typical hidden danger building outside the railway operation environment,the color steel house along the railway will cause safety accidents such as railway shutdown and casualties without regular detection and timely treatment.At present,the detection methods of color steel house along railway lines are mainly based on manual inspection of personnel in the railway engineering section.However,due to the complex external environment and vast space along the railway,manual inspection of color steel house has the disadvantages of low efficiency,limited inspection perspective and lack of automatic data recording means.Existing railway line color steel house survey methods cannot meet the growing external environment along the railway security needs.Therefore,this dissertation proposes the method of automatic inspection by unmanned aerial vehicle(UAV)for the investigation of color steel house along the railway line.This method improves the efficiency of the investigation of color steel house and the means of recording inspection data.At the same time,based on deep learning technology,a method of image classification and color steel house detection along railway line is proposed for collecting UAV inspection image data.The main research contents are as follows:1.Briefly describes the research background and significance of this dissertation,analyzes the necessity of color steel house detection along railway line and the possibility of applying UAV inspection and deep learning to color steel house detection.The application status of UAV inspection,the detection status of color steel house along the railway line and other building structures,and the development status of deep learning are deeply studied.Based on the railway line environment in Handan,the UAV inspection method is proposed for the detection of color steel house along the line.The selection of flight platform,cloud camera and ground station in the UAV inspection equipment is completed.The UAV route and inspection scheme are designed.The field UAV inspection experiment is carried out,and the rich UAV inspection images along the railway line are collected.2.In order to make the collected UAV inspection image more suitable as the input of convolutional neural network,the left and right overlapping decomposition method is proposed to preprocess the image and adjust the pixel ratio.At the same time,the split image is enhanced by geometric transformation and pixel transformation to increase the sample size and sample diversity.3.In order to classify the UAV inspection images along the railway line efficiently and accurately according to whether there are color steel houses,a classification method of UAV inspection images based on MobilenetV1 convolution neural network is proposed.The images are classified by labels as the input of the network,and the classification effects under different width factors are compared.Based on a data set of flights,it is verified that the proposed method can be effectively classified according to whether there is a colored steel house,and the method is applied to screen UAV inspection images with colored steel house in other flights.4.In order to realize the intelligent detection of color steel house along railway line in UAV inspection image,a color steel house detection method based on improved RetinaNet target detection algorithm is proposed.Aiming at the problem that the pixel proportion of some color steel houses is too small,which leads to the unsatisfactory detection effect,a larger scale feature map is used to detect color steel houses by moving down feature pyramid network.The preset anchor frame size is improved by K-means clustering algorithm to make it more suitable for color steel house detection.LabelImg software was used for data labeling,and the test verified that the improved RetinaNet algorithm could effectively detect the color steel buildings along the railway.In order to integrate LabelImg and RetinaNet algorithm detection program and realize automatic labeling of color steel house data,Label-RetinaNet labeling software is developed.5.In order to further improve the detection rate and accuracy of color steel house detection along railway line in UAV inspection images,a color steel house detection method based on improved Mask R-CNN instance segmentation algorithm is proposed.By adding a bottom-up feature map fusion path to construct a two-way feature pyramid network,the transmission of positioning information of color steel house between feature maps of different scales is enhanced.The CBAM convolution attention mechanism is introduced to improve the attention of the algorithm on the channel dimension and spatial dimension of the color steel house area in the image,and the anchor frame in the regional proposal network is improved.LabelMe software is used for data annotation.The experiment verifies that the improved Mask R-CNN algorithm has higher detection rate and accuracy,and can segment the color steel house and image background.Through the research in this dissertation,the equipment type of railway line inspection by UAV is selected and the inspection scheme is formulated.The abundant inspection data of UAV along the railway are obtained,and the methods of image preprocessing and data enhancement are explored.The proposed image classification method based on deep learning and color steel house detection method can effectively solve the related problems existing in manual inspection,and realize efficient,high precision and intelligent detection of color steel house along the railway.It plays a positive role in ensuring the safety and stability of the railway system and improving the economic and social benefits of rail transit operation.