基于改进YOLOv5的远距离小目标检测方法研究  

Research on long-range small target detection method based on improved YOLOv5

作　　者：钱明 杨成佳[1] QIAN Ming;YANG Cheng-jia(School of electrical and computer science,Jilin Jianzhu university,Changchun 130118,China)

机构地区：[1]吉林建筑大学电气与计算机学院,长春130118

出　　处：《吉林建筑大学学报》2024年第1期79-83,共5页Journal of Jilin Jianzhu University

基　　金：吉林省教育厅科学技术研究项目(JJKH20240366KJ).

摘　　要：针对实际情况中小目标人物在目标检测领域中存在识别率低、漏检、精度低等问题,本文提出一种基于改进YOLOv5的远距离小目标检测算法。通过提高数据集图片分辨率,增强目标的特征信息;在特征提取网络引入CABM注意力机制,提高模型提取小目标特征的能力;在特征融合网络引入BiFPN,提高了网络的多尺度特征融合能力和小目标特征识别的效果。与原YOLOv5算法相比,在神经网络模型体积大小变化不大的前提下,验证集数据集上检测精度提升了8%,召回率提高了6%。实验证明该改进方法对远距离小目标检测应用的有效性。In view of the actual situation,small and medium-sized target people have problems such as low recognition rate,missed detection,and low accuracy in the field of target detection,this paper proposes a long-range small target detection algorithm based on the improved YOLOv5.Improve the image resolution of the dataset and enhance the feature information of the target;The CABM attention mechanism is introduced in the feature extraction network to improve the ability of the model to extract small target features.BiFPN is introduced into the feature fusion network,which improves the multi-scale feature fusion ability of the network and the effect of small target feature recognition.Compared with the original YOLOv5 algorithm,the detection accuracy and recall rate on the validation set dataset are improved by 8%and the recall rate is increased by 6%under the premise that the size of the neural network model does not change much.

关 键 词：小目标检测 YOLOv5 网络优化 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]