Research On Military Target Detection Technology Based On Deep Learning

In today's world,with the development of artificial intelligence,a variety of intelligent weapons are constantly emerging and applied to the battlefield,and the traditional war mode has gradually turned to a modern,intelligent war mode.The use of smart weapons greatly improves combat effectiveness.In the use of smart weapons,it is required to be able to effectively attack specific targets.Therefore,how to make various smart weapons automatically recognize various military targets like the human brain has become the primary problem to be solved.This article aims at several common military targets.Based on the current most popular deep learning method,the improved YOLOv3 target detection algorithm is used to intelligently identify several common military targets.Based on the introduction of the CNN network composition structure and working principle,this paper analyzes the typical algorithm of target detection algorithm based on candidate regions and end-to-end target detection algorithm.Then,based on the in-depth analysis of the defects of the NMS algorithm used in the current deep learning algorithm,for the defect that the confidence is directly set to zero,creatively put forward improved measures to adjust the confidence of the piecewise linear decay function,and the defect of greedy screening,Then creatively use the area comparison method and the overlapping area length judgment method to optimize the traditional NMS algorithm,get the improved NMS algorithm,and experimentally verify the improved NMS algorithm on the data set.The results show that the improved The performance of the later NMS algorithm has been improved.In view of the currently undisclosed military target data set,this paper builds a data set for common military target detection.After the construction of the military target data set,the YOLOv3 network model was trained on the military target data set through the configured system,and the detection experiments were carried out on the two test sets.The experimental results show that the proposed YOLOv3 algorithm proposed in this paper Optimization and improvement can better achieve the detection of military targets.