Research On Truck Model Recognition System Based On Neural Network

In the face of the promulgation of the new standard for truck overload and overload,the traditional vehicle identification method is no longer suitable for the identification of truck models at the inspection station.It is urgent to design a new vehicle identification method suitable for over-limit detection stations.The rapid development of image-based vehicle classification system has brought new ideas to the identification of truck models.The identification of truck models based on neural network will improve vehicle detection efficiency and improve road capacity and reduce labor costs.According to the new standard,this paper classifies the nine types of trucks that are common on the road.In order to use this system for different inspection stations,various types of vehicle image information are collected on different roads,and then sorted and classified.Image database.In view of the problem that the background of the image recognition is too complicated and affects the recognition effect of the vehicle,this paper firstly carries out the vehicle target location detection.This scheme combines the feature extraction network based on dense sampling with the convolutional neural network,and analyzes several The problems of the network model in the detection and the different size and wheelbase of the vehicle model will determine the depth of the shared convolution network based on the experimental data,and apply the difficult sample mining strategy to form the vehicle detection model.In the collected vehicle image test samples,good results were obtained,and compared with the traditional RCNN series algorithm,the average detection rate increased by 10%compared with the traditional network model.After the vehicle is detected and positioned,the vehicle calibration part is cut and sent to the vehicle classifier.This paper uses the deep convolutional neural network model for pattern recognition.In order to improve the recognition rate,after designing the vehicle classification network,and a lot of experimental verification,the specific model design details such as the convolution layer number,the convolution kernel size and the activation function are determined.And through the Dropout optimization network,this paper selects he learning rate to make the network optimal,and uses the self-built vehicle side image database Truck to fine-tune the learning parameters of the vehicle classification network.In this paper,the vehicle identification test is carried out on the self-built dataset by using the above method.The results show that this method can effectively obtain higher recognition rate in different scenarios.To solve the problem of two similar vehicle identification errors and ensure the classification accuracy of the vehicles,we can adjust the camera shooting angle and enlarge the difference in the distribution of similar axles.