Research On The Method And Application Of Truck Type Recognition Based On Deep Learning From The Perspective Of Unmanned Aerial Vehicle

Truck is a common means of transportation in people’s daily life.In order to reduce the occurrence of truck overrun transportation,it is necessary to effectively supervise the truck.According to the national standard,each type of truck has different limit values of overall dimensions,so it is necessary to identify the type of truck in advance before the overrun detection of the truck.In recent years,the theory of deep learning has become more and more mature,and its application in the field of image recognition has attracted widespread attention.It can quickly and accurately identify truck types.At present,the traditional overrun detection station mode has the problems of poor mobility and inflexibility.The unmanned aerial vehicle has the characteristics of low cost,flexibility and portability.Therefore,on the basis of full investigation,the scheme of building truck type recognition system based on deep learning theory with unmanned aerial vehicle as flight carrier is established.The main work and conclusions of this paper are as follows:(1)The self-built truck type recognition data set is established.Up to now,there is no public data set of truck side images dedicated to truck type recognition.In order to enable the normal training and iteration of the target detection model,a total of 18640 truck images,including 30963 truck targets,were collected,and the image labeling work was completed,and the truck type recognition image data set was established.(2)A single-stage Mobile Net V3-YOLOv5 s network model for truck type recognition is constructed.Aiming at the problem of truck type recognition to be studied,three classic target detection algorithms based on deep learning are compared and analyzed.Based on the application scenarios and requirements of this study,YOLOv5 s algorithm is selected as the basic algorithm to build a lightweight single-stage Mobile Net V3-YOLOv5 s network model that conforms to the characteristics of the application scenarios in this paper.In view of the shortcomings of YOLOv5 s using GIo U loss function,DIo U loss is used as the regression loss of the bounding box.At the same time,K-means clustering method is used to reset the prior frame to speed up training and effectively improve accuracy.(3)The evaluation and application of the lightweight single-stage Mobile Net V3-YOLOv5 s network model was completed.The Mobile Net V3-YOLOv5 s network model was comprehensively evaluated by selecting reasonable evaluation indicators.When the reasoning speed of Mobile Net V3-YOLOv5 s network model is 91.48 frames per second,the recognition m AP value of truck types still reaches 89.6%.Compared with YOLOv5 s,the m AP value only decreased by 2%,but the detection speed increased by about 27%,achieving the balance between detection accuracy and speed,and the parameter quantity of the model decreased by about 45% compared with the original YOLOv5 s model.(4)The hardware platform construction,model deployment and software interface design of the truck type recognition system are completed,which provides the possibility for the practical application of the truck type recognition system.The portable truck type recognition system based on UAV has the advantages of wide detection coverage,low cost,flexibility and mobility.UAVs can be deployed in key areas of interest without additional infrastructure costs,so the development space of portable truck type identification system based on UAVs is becoming increasingly large.