Research On Recognition Algorithm Of Orbit Obstacles Based On Deep Learning

Railway is the most popular medium-and long-distance travel tool in China.In recent years,with the increase of railway mileage and the improvement increase of train speed,the requirements for the train running environment have become higher and higher.In view of the fact that the railway running track is mostly open,the emergence of track obstacles is often sudden,so railway traffic accidents caused by this have occurred frequently in recent years.Therefore,quickly and efficiently detect and identify obstacles on the track and deal with them in time,which contributes to the safety of railway operations.This paper firstly introduces the current domestic and foreign railway safety operation technology,and then analyzes and clarifies the incompatibility of contact obstacle detection and recognition technology to the current situation of China’s railway operation,and the traditional image processing detection and recognition algorithm based on video image in non-contact technology often can’t meet the requirements of real-time and accuracy at the same time.Based on these problems,this paper proposes and designs an algorithm of track obstacle detection and recognition based on deep learning.From the perspective of deep learning,this paper divides the algorithm into two parts.In the first step,the track image preprocessing detects and identifies the track line by selecting the combination of several image processing algorithms,and extend outward to delimit the ROI track region without excessive background interference;in the second step,the target detection and recognition network is used to classify and locate the obstacles in the track ROI region.The delineation of track ROI region is different from that most algorithms,which use Hough-transform detection track line,but combines edge detection operator,perspective transformation and curve fitting algorithm to fit out the track line,and then the track line is transferred to the outside of the track as the baseline to delineate the track ROI region.This algorithm can detect not only linear orbital zone,can also detect corners orbital zone.Finally,based on the general target detection and recognition algorithm model,the paper improvs it according to the actual application requirements,integrates the track ROI region demarcation algorithm,and fine-tunes on the self-made data set to obtain the detection and recognition models for orbital obstacle detection,and tests and analyzes the new model experimentally.The analysis of the experimental results of the algorithm proves that the algorithm can take into account the requirements of real-time and high detection accuracy at the same time,and complete the task of track obstacle intrusion detection and recognition.The track obstacle detection and recognition algorithm based on deep learning is applied to the field of track safety detection,which can achieve real-time monitoring and real-time detection,and realize the role of timely detection and early response.And it also has a broad development prospect in the field of track obstacle detection and recognition.