Research On The Key Technology Of Tank And Armored Vehicle Target Detection Based On Yolov5

Compared with the traditional image processing and recognition technology,the target recognition technology based on deep learning has more strong adaptability and has achieved good recognition results in many fields.In the military field,due to the influence of factors such as artificial camouflage of targets and the complexity of the ground battlefield environment,as well as the need to identify distant targets in many cases,these targets in the process of recognition due to distance factors lead to the size of the image target becomes small,increasing the difficulty of target recognition.In this paper,a target detection algorithm for tank and armored vehicles based on YOLOv5 is established,and related research is carried out to solve the above problems.The main work is as follows.(1)To solve the problem of insufficient and imperfect image features of target objects generated by small targets and mutual occlusion in current target detection,an improved YOLOv5 algorithm based on attention mechanism is proposed.The experimental analysis showed that the CBAM attentional mechanism module was superior to the CA attentional mechanism module.The use of CBAM attentional mechanism improved the connection of each target feature in the channel and space,which was more conducive to the perfect extraction of effective features of the target,and then effectively solved the identification problem of small targets and overlapping condensed targets.(2)Aiming at the problem of how to achieve faster detection while ensuring high detection accuracy in current target detection,an improved YOLOv5 algorithm based on network model lightweight is proposed.Firstly,Mosaic data enhancement and HSV color space amplification were used to enrich the samples and shorten the training time.Secondly,several experiments were carried out to obtain the optimal network depth and width.Finally,Tensor RT was applied to accelerate deployment and double detection speed.(3)Making a data set of tanks and armored vehicles.In order to solve the problem of lack of special database of tank and armoured vehicle pictures,a data set with 9028 pictures and more than 11,000 tank and armoured vehicle targets was made independently.The data set contains all kinds of complex scenes(such as mountains,deserts,forests,fog,snow and other extreme conditions),as well as multi-scale and various forms of pictures of tanks and armored vehicles,which greatly enhances the robustness of the algorithm.This paper mainly studies the principle and implementation process of YOLOv5 algorithm,and builds the implementation environment of YOLOv5 algorithm on Ubuntu18.04 system with Py Torch framework.(4)Development of human-computer interaction interface.Aiming at the problem that the current target detection is mainly carried out under the command line window,a Web page version of friendly human-computer interaction interface based on Web Flask is designed and developed,which realizes the visual operation of image target detection and can realize the unrestricted access of multiple terminals at the same time.(5)Verification of the algorithm robustness.Based on the existing data,the brightness experiment,occlusion experiment,rotation experiment,tensile experiment and noise experiment were designed and carried out to verify the robustness of the algorithm.Combined with the self-built tank and armored vehicle dataset,the target detection model of tank and armored vehicle based on YOLOv5 is trained and tested.The detection accuracy can reach 96.6%,the m AP can reach 97.7%,and the FPS can reach up to 200.In order to solve the problem of missing detection of fuzzy images and wrong detection of targets with similar features and not obvious features,a solution is given.There are 59 figures,18 tables and 44 references.