Research Of Vehicle Classification Under Hazy Environment Based On Convolutional Neural Network

Vehicle recognition system is an important part of urban intelligent transportation system.As the rapid development of some technology,such as digital image processing,computer vision and machine learning,vehicle recognition technology based on image content has been widely concerned.Currently,vehicle recognition technology is mainly based on statistical classification.First of all,a large number of vehicle images are needed as a dataset.Secondly,feature information of the vehicles are extracted by feature extraction technology.Finally,the features are learnt and classified using machine learning classification technology to finish the vehicle recognition.In many recognition algorithms based on image,only few can be actually applied to the Intelligent Transportation System duing to the influence of multi-scale,multi-angle and complex background.At the same time,affected by the fog and haze,haze reduction vehicle recognition has become a difficult point.Therefore,it is of great significance to study how to accurately and efficiently classify the vehicle models in complex environment.At present,in view of multiple influence factors of vehicle recognition based on images,the main contents are as follows:(1)Setting up a database based on vehicle images fetched from the Internet.Because of the complexity,multi-angle,multi-scale and multi-type of vehicle background,the variety of the database is ensured.Types of vehicles consist of minibus,pickup truck,sports car,car,suv and van.(2)To recover the clarity of vehicle images imaged in the fog and haze weather using the atmospheric scattering model combined with the scattering characteristics of haze particles.(3)A vehicle recognition algorithm under the environment of haze based on Convolutional Neural Network is proposed,which is combined with Softmax classifier to recognize vehicles.We adopt the improved haze removal algorithm and the feature extraction network structure optimized by Alex Net,which are compared with current vehicle recognition algorithms based on HOG feature,PCA+SIFT feature or CNN feature.Through the tests,It is proved that the vehicle recognition algorithm improved by AlexNet proposed in this paper is more accurate than other algorithms in both sunny and hazy weather.(4)A vehicle recognition system is achieved in complex environment,which consists of vehicle image preprocessing module and vehicle classification module.The image haze removal algorithm proposed in this paper is used in vehicle image preprocessing module and vehicle recognition algorithm proposed in this paper is used in vehicle classification module.Through the combination of theory and practice,prove the effectiveness of the algorithm in this paper and the significance.