基于改进Yolov5n的无人机对地面军事目标识别算法

Recognition Algorithm for UAV Ground Military Targets Based on Improved Yolov5n

针对目前主流的目标检测算法在真实航拍战场数据背景下识别精度低、误检率与漏检率高等问题,对Yolo目标识别算法进行了研究,提出一种基于改进Yolov5n的轻量化航拍军事目标检测模型;首先,采用ECA注意力机制与主干网络C3模块融合,以解决航拍图像背景复杂且存在相似目标干扰问题;其次,引入归一化高斯瓦萨斯坦距离(NWD)代替CIoU损失函数,提高对模糊小目标的检测识别;最后,采用GSConv轻量化卷积代替标准卷积,减轻模型重量;经过实验测试,改进后的算法模型平均检测精度达到81.5%,提升0.9个百分点,模型大小为3.4 MB,减轻0.4 MB,识别速度为每秒113帧;实验结果表明该模型在轻量化的同时保持着高精度的航拍军事目标检测。

In response to low recognition accuracy,high false detection rate and missed detection rate of mainstream target detection algorithms in real aerial battlefield data backgrounds,research was conducted on the Yolo target recognition algorithm,and a lightweight aerial military target detection model based on improved Yolov5n was proposed;Firstly,the efficient channel attention(ECA)mechanism is integrated with the C3 module of the trunk network to solve the interference from complex backgrounds and similar targets in aerial images;Secondly,the normalized Gaussian Wasserstein distance(NWD)is introduced to replace the CIoU loss function,improving the detection and recognition of fuzzy small targets;[JP2]Finally,the GSConv lightweight convolution is used to replace standard convolution to reduce the weight of the model;After experimental testing,the improved algorithm model reaches an average detection accuracy of 81.5%and improves 0.9 percentage points,with the model size of 3.4 MB,reduction of 0.4 MB,and recognition speed of 113 fps;Experimental results show that the model has high accuracy in aerial military target detection while being lightweight.[JP]