Research On Orbital Foreign Object Intrusion Detection Algorithm Based On Deep Learning

With the rapid development of high-speed railway in China and the continuous increase of train speed,the safe operation of trains has become more and more important,and the problem of foreign body intrusion on track has seriously affected the safe operation of railways.So,it is necessary to detect and protect the track to prevent foreign bodies from invading the track.The traditional protection scheme has a large amount of engineering,high cost,few types of detection,low precision,and requires manual intervention.Although machine learning-based detection system reduces the engineering quantity and operation and maintenance cost to some extent,it still has great limitations,such as detecting only moving objects and large objects,which has become increasingly difficult to meet the increasing demand for safe transportation environment.Based on a brief introduction of domestic and international railway operation safety detection technology,the existing problems of orbital foreign body intrusion detection technology are analyzed and expounded in this paper.Aiming at the key technical bottlenecks and the shortcomings of traditional machine learning algorithms,the algorithm of orbital foreign object intrusion detection based on deep learning is proposed and studied in this paper.The research contents of trajectory foreign body intrusion detection algorithm based on deep learning are mainly divided into the following aspects:Firstly,the detected video is processed in a single frame,and the acquired single frame image is pre-processed.Secondly,it is necessary to denoise the image first,because a lot of noise are produced in the process of acquiring video from outdoor scene and image conversion.In this paper,the advantages and disadvantages of four commonly used image filtering algorithms are compared and analyzed.The median filtering algorithm is chosen to denoise the image.At the same time,in order to detect the train track better,a series of operations such as weighted average gray processing,binarization processing and edge detection extraction are carried out to get the optimal edge detection image.Thirdly,the railway track is detected by Hough transform algorithm,and the railway track is taken as the baseline to move to the outer side of the track,thus the invasion area of foreign body in the railway track is delineated.Fourthly,based on the two models of SSD and YOLO,the application scenarios and the characteristics of technical requirements are improved in this paper,and the model is trained by self-made samples.The experimental results and comparative analysis prove that the target detection model based on deep learning can not only accomplish the task of foreign body detection in orbit,but also solve the problem that the traditional algorithm cannot detect the slow moving target,and does not need to distinguish the train and foreign body running opposite to each other,which greatly improves the detection speed and accuracy.The algorithm has broad development prospects and practical application value in the field of railway foreign body detection,and has good reference value for the development and application of artificial intelligence technology in various cross-cutting fields.