Research On Objection Detection Algorithm For Train Components Based On Structured Knowledge

As an important facilities of transportation,the development of railway train is related to the national economy and people’s livelihood.With the development of trains,the safe operation of trains has gradually grasp the focus of people’s attention.The traditional train safety inspection relies on manual inspection by examining the bottom of the train box one by one,which is a time-consuming and labor-consuming process,and may lead to wrong inspection due to visual fatigue or other factors.Therefore,it is very necessary to improve the intelligent automatic train detection system.As an important part of intelligent automatic train detection system,object detection is mainly responsible for the detection of train components.Due to the complexity of scene and the variability of object scale in the industrial field,the design of object detection algorithm will be greatly challenged.The purpose of this paper is to further improve the object detection algorithm for intelligent detection system of train components.Two intelligent algorithms are introduced for intelligent detection of train components by integrating structured knowledge and deep learning technologies.The research contents of this paper mainly include:1.This paper proposed an intelligent object detection algorithm for train components based on structured knowledge.Aiming at the strong correlation and scale variability among the train components,the algorithm is based on the cascade convolutional neural network,which utilizes the structured knowledge of the spatial relationship between objects to detect the train components.The core idea of the algorithm is to detect large and easily discoverable components first,and then discover the region of interest that may contain small components for further process.Next,the optimization mechanism analyzes and optimizes these regions through the basis of structured knowledge.The optimized regions is then cropped,enlarged and input into the following model for further detection.Finally,the experimental results show that the precision of this algorithm achieved 92.30%(m AP),which is an excellent result and can be applied in the detection system with high precision requirement.2.An object detection algorithm are proposed,which integrates structured knowledge and scene information.The algorithm can mainly divided into three modules.The first module is to optimize the original features by using the constructed spatial relationship knowledge to get the spatial features.The second module is to determine the scene category of the input image and extract the scene features.The third module uses the classified scene categories and the corresponding structured knowledge to supervise the generation of relationship features.Finally,the algorithm designed a feature fusion module to fuse theses three features and then put them into the final classification and regression network for detection.The experimental results demonstrates that the accuracy of the proposed algorithm is superior to the based object detection algorithms,which is 87.63%.Although the accuracy of this algorithm is lower than that of the first algorithm,the detection speed of this algorithm is faster,up to 0.23 second per frame.It is suitable for the system which requires a faster detection time.