Algorithm And Technology Implementation Of Vehicle Color Recognition Based On Deep Learning

The vehicle color is one of the important features of the vehicle.It is an important means to combat evasion of fees,inspection of black vehicles,and tracking of deck vehicles.For this reason,the paper relies on provincial transportation science and technology projects to carry out research on vehicle color recognition algorithms,which has important theoretical significance and application valueThe paper describes the research status of vehicle color recognition at home and abroad,and analyzes the main problems or limitations that exist at present.The main manifestations are:near-color cars are difficult to distinguish;color cast due to heavy fog,strong light/insufficiency,vehicle color Features are difficult to extract accurately.In order to solve the above problems,the paper proposes a vehicle color recognition algorithm that combines preprocessing and deep learning technologies.The original color of the vehicle body is restored by the preprocessing technology,and the image quality is improved.The vehicle color features are extracted from the deep learning network to realize the classification and recognition.For the problem of vehicle color cast,the paper adopts pre-processing technology to correct color cast,restore vehicle body color,and improve image detection accuracy.The vehicle fogging is eliminated by the defogging method based on the dark channel prior principle;the vehicle color cast caused by insufficient illumination is corrected based on the Retinex theory;and the correction algorithm for high light detection combined with the YUV space and Criminisi algorithm is also used.The highlight color of the vehicleFor the problem that the vehicle's color features can't be selected accurately,the paper uses deep learning to extract features:using AlexNet network to train the pre-processed data sets and implement the classification,which greatly improves the accuracy compared with the traditional vehicle color identification methods;through the residual network ResNet-18 Improve the"degradation" of the network and improve the classification accuracy.ResNet-18 network has a deeper level and better recognition than AlexNetWhen the pre-processing method and the deep learning network were selected,the accuracy and real-time of the vehicle's color recognition were taken into account.After a large number of comparative experiments,the paper validated the superiority of the proposed vehicle color recognition algorithmThe innovation and characteristics of the paper are? Proposed vehicle color recognition algorithm combining image preprocessing technology and deep learning to improve the recognition accuracy;? Optimized the deep learning recognition network and optimized the original AlexNet network into the ResNet-18 network,again improving the recognition accuracy.