基于YOLOv5深度学习模型的动态靶标识别跟踪方法

Target Recognition and Tracking Method Based on YOLOv5 Deep Learning Model

针对火炮身管动态靶标识别跟踪精度和实时性不高的问题,提出了一种基于YOLOv5深度学习模型的靶标识别跟踪方法。分析了靶标识别跟踪过程和基本思想,通过网格化模型对靶标样本图像进行多尺度处理,并利用金字塔模型进行融合预测;搭建了YOLOv5网络模型,对组件设置优化;对比了损失函数对锚框识别效果的影响,并选取优化后的CIOU作为模型损失函数;最后对模型进行训练,并利用训练好的模型对动态靶标进行识别跟踪。实验结果可视化分析显示,靶标动态识别跟踪率可达到99.3%,动态实时跟踪效果较好。

Aiming at the low accuracy and real-time performance of dynamic target recognition and tracking of gun barrel,a target recognition and tracking method based on yolov5 deep learning model was proposed.The process and basic idea of target recognition were analyzed.The multi-scale processing of target sample image was carried out through grid model,and the fusion prediction was carried out by pyramid model.The research built the yolov5 network model and optimized the component settings.The influence of loss function on the recognition effect of anchor frame was compared,and the optimized CIOU was selected as the loss function of the model.Finally,the model was trained,and the trained model was used to identify and track the dynamic target.The visual analysis of the experimental results shows that the target dynamic recognition and tracking rate can reach 99.3%,and the dynamic real-time tracking effect is good.