Research On Vehicle Logo Image Detection Technology Based On Deep Learning

Computer vision and image target detection are two important research directions in the field of machine learning.In the field of the modern Intelligent Transportation System(ITS),a large amount of image data is generated every day.Detecting the targets that need to be identified in these images can effectively help the daily traffic management law enforcement.Therefore,it is particularly important to constantly explore efficient target detection methods.Vehicle Manufacturer Recognition(VMR)is an important application in ITS.Since the car logo is the clearest and most typical feature of major car manufacturers,most methods for identifying vehicle brands are based on Vehicle Sign Recognition(VSR).Currently,it is still difficult to accurately segment the logo from images containing vehicles of various colors.In various imaging environments,VMR technology becomes more difficult due to geometric distortion or various imaging conditions.The VSR has high requirements for robustness,so vehicle identification is still a challenging identification problem.In the application of traditional target recognition methods,manual extraction of image features requires many limitations.Therefore,Convolutional Neural Networks(CNN)acquired many achievements in many machine vision tasks in recent years.Inspired by the he excellent performance of CNN,this paper designs a vehicle identification system based on convolutional neural network and conducts in-depth research.The CNN algorithm is applied to the vehicle identification system,which can eliminate the requirements for accurate detection and segmentation of the vehicle standard in the traditional machine learning method.At the same time,in order to solve the problem of high computational cost of the convolutional neural network during the process of training,this paper introduces an effective network improvement strategy to enable the network to be implemented in real-world applications.In this paper,the Faster R-CNN algorithm is used to train the network,which significantly increases the training speed while reducing the computational cost.At the same time,this paper uses this method to help prevent over-fitting and avoid the algorithm falling into local optimum.In the experiment of this paper,10,000 identification image data sets containing 10 automobile manufacturers were produced,of which 8000 images were used for training,while 2000 images were used for testing,and the improved algorithm was used for data sets and evaluation of algorithm performance.Experiments results show that the average accuracy of the algorithm is 92%.Compared with the classical convolutional neural network and the traditional machine learning algorithm,the algorithm has the best recognition effects and the highest recognition accuracy.This indicates that the method has high classification ability and is robust to various poor imaging environments.The algorithm can be further applied in real-world applications.