Deep Learning Classification Method Of Vehicle Images Based On Frequency Segmentation Enhancement

With the development of society,vehicles have been widely used today.While providing convenience for people,they also bring many safety issues that cannot be ignored.Especially when intelligent transportation is gradually enriching the social governance system,the classification and identification of vehicles has gradually become one of the key areas of intelligent transportation.Among them,the vehicle information collected by the camera can speed up the event processing process,but the image collected by the camera is easily affected by the environment,resulting in a significant reduction in the amount of information.Therefore,it is of great practical significance to study the enhancement processing of blurred images and the classification methods of vehicle images.According to the research on image enhancement algorithms and the in-depth analysis of deep learning classification networks,based on vehicle image data sets,combined with traditional image enhancement algorithms and improved Res Net,a deep learning classification method for vehicle images based on frequency segmentation enhancement is proposed.The main work of this paper is as follows:In the field of intelligent transportation,an image enhancement algorithm based on frequency segmentation is proposed for the effect of image blur on vehicle classification.A mathematical model is established based on the correlation between image pixels,and the metric matrix is obtained by defining the frequency mapping of the mean square value of the pixels and Gaussian weighting.The metric matrix is normalized to obtain the feature map of the image,and then the feature map is segmented into a low-frequency image,a medium-frequency image,and a high-frequency image.The segmented sub-images are respectively enhanced with gamma correction,contrast-limited adaptive histogram equalization,and Laplace enhancement,and finally the three part feature maps are merged to obtain the overall enhanced image.In the open source image enhancement data and synthetic image data,two evaluation methods,subjective and objective,are used to measure the performance of the algorithms.Compared with other algorithms,it proves the effectiveness of this algorithm.In the subsequent classification experiments,the effect of the enhancement algorithm is also explained in a deeper level.For the application of vehicle image classification,a vehicle image classification method based on improved Res Net is proposed.By adding a random down-sampling mechanism on the basis of the traditional Res Net50 network.And by adding activation functions and batch normalization layers to improve the network’s hierarchical structure.Using the random downsampling rate,during each network iteration,average pooling downsampling is added between different residual blocks.The scale of the feature map is randomly reduced according to the sampling rate,so that the network can achieve higher classification accuracy while having a small amount of calculation,so as to achieve the balance and unity of accuracy and calculation amount in network performance.It adopts a clear data set containing ten types of vehicles,a fuzzy data set,and a data set processed by the enhanced algorithm of this paper.These three types of data are separately trained and compared with other classification networks.The experimental results show that the training loss of the classification network in this paper is lower.According to the evaluation of parameters,calculation amount and accuracy,it proves that the vehicle classification network proposed in this paper has obvious advantages.And the enhancement algorithm proposed in this paper can weaken the negative influence caused by image blur in the preprocessing of the classification network.