Research On Machine Vision Detection Method For Bird’s Nest Invasion Of High-speed Railway Catenary

In recent years,my country’s high-speed railways have developed rapidly and have built a high-speed railway network with the longest operating mileage and the highest operating speed in the world.High-speed rail has gradually become the preferred mode of travel for people.Therefore,how to ensure the safe operation of high-speed rail has become very important.Among them,high-speed rail foreign body intrusion detection technology has played a very important role in the safe operation of high-speed rail.The high-speed rail catenary is a special form of transmission line that is erected along the railway line to supply power to electric locomotives.Most trains in my country use this form of power supply,and the intrusion of foreign objects on it is one of the important factors affecting the safe operation of high-speed rail.In a variety of invading foreign objects,the nests built by birds are prone to failures such as short-circuiting of transmission lines and breakdown of insulators,which can cause great harm.Therefore,how to locate and identify the bird’s nest invaded by the high-speed rail contact network in real time and accurately is an important issue to be solved in ensuring the safe operation of the high-speed rail.In this paper,the research on machine vision detection method for bird’s nest intrusion of high-speed railway catenary network is carried out.The main innovations are as follows:Aiming at the current problems of the bird’s nest data with few samples and slow detection speed,it is proposed to speed up the processing speed of subsequent algorithms through image gray-scale processing before accurately detecting the bird’s nest,and use the Retinex algorithm to dehaze and enhance the image.Two methods are used to extract the region of interest that may appear in the bird’s nest: for the catenary equipment with simple structure and not much difference in the parameters of the cantilever pillar,the region of interest is extracted based on the LSD line segment detection algorithm;for the catenary equipment with complex structure and much difference in the parameters of the cantilever pillar,the region of interest is extracted based on the lightweight network YOLOv3-Tiny,so as to adapt to various types of catenary equipment and reduce the impact of environmental factors on the detection results.Finally,add the bird’s nest image to the extracted region of interest and reduce the weight of the easy-to-classify samples in the loss function to solve the problem of imbalance between positive and negative samples.Aiming at the current low accuracy and slow detection speed of bird’s nest intrusion detection of high-speed railway catenary.Firstly,according to the size of the area occupied by the bird’s nest in the image,the K-means algorithm is used to obtain the number and size of the prior frame,and the multi-scale fusion operation is used to design the network detection branch.Secondly,an attention mechanism based on global maximum pooling and global average pooling is introduced into the model to learn the correlation between feature map channels.Then use DIo U-Loss and DIo U-NMS to optimize the loss function to speed up the convergence of the network.Finally,a residual network structure based on depth separable convolution is designed,which reduces the size of the network model while ensuring the detection accuracy and detection speed are basically unchanged.Comparative experiment results show that the model proposed in this paper improves the average detection accuracy by 3.56%,the detection speed by3.03%,and the model size is reduced by 9.83%.