摘要
本文针对航拍图像检测中可提取目标特征少、小目标多导致检测精度低的问题,提出了一种基于原始YOLOv5算法的改进目标检测算法。首先,将CBAM注意力机制部署到原始YOLOv5的特征提取和特征融合网络的C3模块中,以提升主干网络中网络局部特征捕获与融合能力,其次,针对在不断采样过程中导致浅层次特征信息丢失的问题,在检测头部分,增加了一个检测头,用于检测小目标。最后,对原始YOLOv5的Mosaic数据增强方式进行改进,将4张图片拼接改为9张图片拼接,然后根据拼接图片的矩形面积对灰色背景进行裁剪,加快模型收敛,提高训练效率,在不增加模型计算量的前提下,改善背景复杂等问题。本文使用VisDrone2019数据集,并基于改进的算法与原始算法分别进行了实验,结果表明本文的算法能够针对航拍图像目标进行有效的检测。
In this paper, based on the original YOLOv5 algorithm, an improved target detection algorithm is proposed to solve the problem of low detection accuracy caused by few extracted target features and many small targets in aerial image detection. Firstly, CBAM attention mechanism is deployed in the C3 module of the original YOLOv5 feature extraction and feature fusion network to improve the local feature capture and fusion capability in the backbone network. Secondly, to solve the problem that shallow level feature information is lost in the process of continuous sampling, a detection header is added to the detection header, used to detect small targets. Finally, the Mosaic data enhancement method of the original YOLOv5 was improved, and the Mosaic of 4 images was changed to 9 images, and then the gray background was cropped according to the rectangular area of the Mosaic images, In this way, the model convergence can be accelerated, training efficiency can be improved and complex background can be improved without increasing the calculation amount of the model. In this paper, we use VisDrone2019 data set to carry out experiments based on the improved algorithm and the original algorithm respectively. The results show that the proposed algorithm can effectively detect aerial image targets.
出处
《软件工程与应用》
2023年第3期455-462,共8页
Software Engineering and Applications
