Vehicle model identification method based on convolutional neural network

Abstract

The invention discloses a vehicle model identification method based on a convolutional neural network, and belongs to the technical field of vehicle model identification in intelligent transportation.According to the method, a data set is amplified by adopting a data enhancement strategy, so that the data set is more suitable for the actual situation, the weight and the offset are updated by adopting an Adam optimization algorithm for vehicle model identification on the basis of an Inception-V3 network, and the efficiency of vehicle model identification is improved by referring to a transferlearning thought. Besides, the AMSoftmax loss function is adopted to replace the traditional Softmax loss function on the basis of the loss function, so that the intra-class difference is reduced, theinter-class difference is expanded, and the vehicle model identification precision in different background environments is improved.

Description

technical field [0001] The invention relates to the technical field of vehicle identification in intelligent transportation, in particular to a vehicle identification method based on a convolutional neural network. Background technique [0002] Due to the increasing traffic pressure in recent years, the importance and urgency of the intelligent transportation system has become increasingly evident. As an important branch and key technology of the intelligent transportation system, vehicle type recognition is mainly for accurate identification of vehicles in specific locations, which can provide Provide technical support for intelligent parking lot charging system, vehicles illegally occupying roads, etc. [0003] The main methods of vehicle recognition research at home and abroad can be summarized as: methods based on image processing and methods based on physics. Convolutional neural network is a kind of neural network. It consists of convolutional layer, pooling layer, ac...

AI Technical Summary

Problems solved by technology

However, in the actual traffic environment, there are factors such as illumination changes and image blurring in vehicle recognition, which affect the accuracy of convolutional neural network for vehicle recognition.

To sum up, facing the problem of relatively low recognition accuracy of car models in reality, the present invention adopts the convolutional neural network model and introduces the AMSoftmax loss function to classify the pictures of car models, which solves the shortcoming of low recognition accuracy of car models in the prior art

Images