坦克目标的准确识别定位是信息化战争中一项重要研究，针对传统检测算法抗干扰性差、难应用于大视野复杂环境下的问题，提出了 一种基于改进YOLOv5坦克自动识别的检测算法。利用YOLOv5模型对大视野复杂战场环境下坦克目标进行识别:在YOLOv5基础模型中引入Attention-based information fusion模块，提高模型检测精度和识别能力;使用Pre-segment multi-scale fusion模块解决骨干网络中池化操作所造成的信息丢失问题;使用Swin Transformer机制降低小目标坦克漏检误检的问题。在坦克数据集上进行实验，结果表明:与YOLOv5原始模型相比，改进模型的召回率、平均精度分别提高了 9。1％、5。1％。改进后的YOLOv5模型可以很好地对大视野复杂环境下坦克目标进行精确识别，改善了坦克目标检测中小目标漏检的问题。
Tank target detection algorithm based on improved YOLOv5
Accurate identification and positioning of tank targets is an important research in information warfare.Aiming at the problems of insufficient timeliness and low accuracy of traditional detection algorithms,a detection algo-rithm based on improved YOLOv5 tank automatic identification was proposed.The YOLOv5 model is used to identify tank targets in complex battlefield environment.The Attention-based information fusion module is introduced into the basic model of YOLOv5 to improve the detection accuracy and identification ability of the model.The Pre-segment multi-scale fusion module is used to solve the problem of information loss caused by pooling operations in backbone network.Use Swin Transformer to reduce the leakage rate of small target tanks.The experimental results show that compared with the original YOLOv5 model,the accuracy rate,recall rate and average accuracy of the improved model are increased by 0.9％,11％and 5.7％,respectively.The improved YOLOv5 model can accurately identify tank tar-gets in a complex environment with a large field of vision,reducing the problem of tank small targets missing detection.
万方数据
维普
