In laser simulation training,target devices with intelligent strike functions can more realistically simulate the actual combat confrontation process to meet the needs of intelligent military training.However,most of the current targets do not have intelligent strike or counterattack functions,targets can only achieve the target hit function or non intelligent trigger counterattack function,the target can only realize the function of being struck as a target or non intelligent trigger counterattack function.The simulation training format is monolithic,with target movement or strike function showing regularity,and does not realistically simulate the ability of an enemy target to initiate an attack.Therefore,in order to adapt to the needs of simulation training,it is necessary to simulate the enemy targets through intelligent targets for combat.In the development process of intelligent targets,one of the first problems to be solved is object detection.The ultimate goal of this research is to detect and make attack activities on individual targets in the battlefield environment,so the detection speed and accuracy of individual targets have greater requirements.In this paper,an algorithm based on Yolov5 neural network architecture to improve the accuracy of target detection and accelerate the speed of target detectionand to optimize the traditional target detection algorithm is proposed using theimproved the network structure to ensure that under the input visible light image,it can achieve rapid detection and ensure the accuracy of detection.