Study On Key Technologies Of Green Traffic Management Based On Image Recognition

As a vital Chinese benefit farming policy,the "green channel" not only guarantees the supply of agricultural products,but also promotes the economic development of rural areas.However,with the in-depth implementation of the policy,some problems in the management of green traffic are gradually exposed.For example,the number of green traffic has increased dramatically and some truck drivers choose to disguise themselves as green traffic to evade toll by the profit temptation,which brings challenges to vehicle inspection.Meanwhile,lack of data acquisition standardization,the inspection work is facing enormous difficulties under the poor image quality of green traffic.On the basis of the inspection requirements of green traffic,an image classification model,integrated with image recognition,target detection and other technologies,is designed to automatically distinguish the validity of vehicle images such that the quality of image database can be improved.Further,vehicle types are classified and identified,and different types of green traffic cheating are studied to provide auxiliary reference for toll station inspectors.Firstly,the image data characteristics of green traffic,the type and reason of the inefficiency of the green traffic image are investigated through the field investigation and analysis of the current situation of green traffic management.In addition,an artificial criterion for judging the validity of green traffic image is formulated on the basis of the subjective evaluation method of image quality,and tags are made to provide data support for the training of image validity classification and recognition model.Secondly,an image classification model is designed with the image data characteristics of green traffic based on Convolution neural network,including image classification module and image preprocessing module.A method of artificial synthesis of minority samples is adopted in the image preprocessing module to enhance the image data and prevent the network from over-fitting,which can handle the problem of unbalanced data sets of green traffic images in the database of Expressway Green Traffic Management platform.Meanwhile,an image preprocessing method based on object detection and image clipping is proposed for the problem of complex image information,where the redundant information besides classifying and recognizing objects can be eliminated,and the classification accuracy can be greatly improved.The experimental results of various convolutional neural network models for effective classification of pre-processed images are compared and analyzed,and the highest accuracy rate is 93.76%.Finally,two indicators,axle type and carriage-loading type,are selected as the key criteria for the classification of green traffic vehicles based on the attributes of freight cars,and the green-through vehicles are divided into 8 categories according to the axle type,and are classified into 6 categories according to the carriage-load type.With this criteria,vehicle types are identified and classified by image validity classification method,where the highest accuracy rate of axle classification experiment can reach 97.20%,and the highest accuracy rate of axle classification experiment can reach 98.82%.By this contribution,the cheating situation of different types of vehicles is futher studied from the requirement of green traffic inspection,which can provide auxiliary reference for the differential inspection of green traffic.