Research On The Method Of Truck Axle Type Recognition Based On Deep Learning

Truck axle type recognition is widely used in areas such as highway toll collection and truck overweight detection.Traditional truck axle type recognition methods based on axle type detectors have the problems of low accuracy,road damage during installation,difficult maintenance,and high cost.In response to these problems,this paper studies the truck axle type recognition method based on deep learning.Only the image information of the truck is needed to directly identify the truck axle type,which has the advantages of high accuracy,low cost and convenient installation and maintenance.The main contents of this article are as follows:(1)Research a method for identifying truck axle types based on target detection.The method first uses the target detection algorithm YOLOv3 to detect truck axles,and then recognizes truck axle types based on the results of the truck axle detection.Aiming at the problems of network structure redundancy and low accuracy in detecting truck axles in YOLOv3,the network structure and loss function of YOLOv3 are improved.Experimental results show that the average accuracy of this method in identifying truck axle types is 98.42%,and it only takes0.096 seconds to identify truck axle types on the CPU side.(2)Research a method for identifying truck axle types based on image classification,which uses image classification algorithms to identify truck axle types.In view of the fact that some truck axle types are very similar,the image classification algorithm has low accuracy in identifying truck axle types.This method draws on the idea of transfer learning and uses two pre-trained convolutional neural networks to extract truck wheel axle features and truck contour features.,And then fusion of the wheel axle features and contour features,and finally use the classification output method to identify the truck axle type.Experimental results show that the average accuracy of this method in identifying truck axle types is 99.26%,and it only takes 0.177 seconds to identify truck axle types on the CPU side.(3)Test the truck axle type identification method studied in this paper in a practical application scenario(a truck overload monitoring station of a coal mine enterprise in Lishi District,Luliang City,Shanxi Province).The test results show that the truck axle type recognition methods based on target detection and image classification in this paper have achieved high average accuracy,reaching 98.60%and 95.24% respectively,which proves the effectiveness of the truck axle type recognition method in this paper.