基于SKNet改进YOLOv5s的无人机对道路小目标的检测

Improving YOLOv5s Based on SKNet for Unmanned Aerial Vehicle Detection of Small Road Targets

针对无人机航拍图像中出现的道路小目标检测精度较低的问题,提出以融合SKNet卷积核注意力机制与YOLOv5s的目标检测模型,提高检测模型对地面小目标特征信息提取识别能力。在此改进基础上,基于Visual Studio Code配置的Pytorch深度学习开发环境,对SKNet+YOLOv5s的性能进行测试试验。结果表明:以VisDrone2019作为数据集训练时,相较于几种常规注意力机制的改进方法,如SENet+YOLOv5s、CBAM+YOLOv5s,SKNet+YOLOv5s的检测精度有所提升。

Aming at addressing the issue of low detection accuracy in detecting small road targets in drone aerial images,this work proposes a target detection model based on the fusion of SKNet convolutional kernel attention mechanism and YOLOv5s,which has enhanced ability to extract and recognize feature information of small ground targets.Based on this improvement,a Python deep learning development environment configured with Visual Studio Code was used to conduct performance testing experiments on SKNet+YOLOv5s.The experimental results show that when using VisDrone2019 as the dataset for training,compared to several conventional attention mechanisms such as SENet+YOLOv5s and CBAM+YOLOv5s,the detection accuracy of SKNet+YOLOv5s is higher.