The Design Of Train Car Information Detection System Based On Target Detection Algorithm

The port mainly records the train car information(train number,train tare weight,etc.)entering and leaving the port and the actual information entering and leaving the port,and is used for settlement based on the difference.At present,there is a problem of inconsistency between the information of the train car and the information in the system.The manual transcription is labor-intensive and errors cannot be verified.So how to quickly and accurately detect train car information is the key to research.According to the actual environment and the characteristics of the train car information target,a regression-based convolutional neural network target detection algorithm is used to complete the detection of the train car information.Firstly,a data acquisition system is designed to obtain clear video images,and through the preprocessing and target detection algorithm,the train car information is detected.The main work contents are as follows:(1)Comparatively studied five kinds of convolutional neural network target detection algorithms,compared and analyzed the accuracy and speed of the test results of other algorithms,and selected the higher speed and higher accuracy YOLOv3 algorithm.(2)According to the size of the train car and the distance between the shooting point and the train car,select the Genie TS-C2048 camera and VS-0618H1 lens,select the LDR2-32SW2 ring light source to increase the fill light,ensure the clarity of the video image,through the network cable transfer the data to the computer.LabelImg software is used to label the pre-processed images to obtain the sample library required by the network.(3)Based on the sample library,combined with the K-means ++ clustering algorithm to adjust the size of the anchor frame,according to the characteristics of the target interruption,the YOLOv3 algorithm is improved in scale and feature fusion mechanism,reducing the amount of calculation and improving The generalization ability of the model.(4)Analyzed the project requirements and system indicators,conducted training and testing,and proposed solutions to increase the sample library and threshold comparison for the problems of missed detection and false detection during the testing process,and finally completed the project that met the requirements system design.