基于改进的YOLOv5加拿大一枝黄花检测方法
A method for detecting Canada Goldenrod based on an improved YOLOv5 network
章朋 1张磊 2尹世铭2
作者信息
- 1. 苏州健雄职业技术学院 人工智能学院,江苏 苏州 215411
- 2. 常州大学 应用技术学院,江苏 常州 213164
- 折叠
摘要
为提高加拿大一枝黄花的治理效率,实现对其智能化监测,提出一种改进的YOLOv5(You Only Look Once)网络的检测方法.首先,在YOLOv5 网络中添加通道注意力模块,在维持模型轻量化的基础上提高了目标检测精度.其次,在主干网络中添加SPD-Conv(Space-to-Depth Convolution)层,提高了网络对低分辨率图像以及小目标检测的效果.最后,采用自适应激活函数替换原网络的SiLU(Sigmoid Linear Unit)激活函数,在网络快速收敛的同时提高了网络的泛化能力.实验结果表明,改进的YOLOv5 网络精度均值较原网络提高了 5.8%,改进后的网络收敛速度更快,检测精度更高,能够更好地满足加拿大一枝黄花检测与防治应用的需求.
Abstract
In order to improve the detection efficiency of Canada goldenrod,an improved YOLOv5 network is proposed.Firstly,an attention module was added to the YOLOv5 network to improve the target detection accuracy while maintaining the light weight of the model.Secondly,an SPD-Conv layer to the backbone network was added to improve the network effective-ness in detecting low-resolution images as well as small targets.Finally,the SiLU(Sigmoid Linear Unit)activation function of the original network was replaced by the meta-Acon(Activate or Not)adaptive activation function,which improved the general-ization ability of the network while the network converges quickly.The experimental results show that the mAP(mean Average Precision)of the improved YOLOv5 is 5.8%higher than the original network.The improved network has faster convergence speed,higher accuracy,which can better meet the needs of invasive Canada Goldenrod detection and control.
关键词
YOLOv5/卷积神经网络/加拿大一枝黄花/注意力机制Key words
YOLOv5/convolutional neural network/Canada Goldenrod/attention module引用本文复制引用
基金项目
江苏省高校"青蓝工程"优秀青年骨干教师项目(202327)
出版年
2024
NETL
NSTL
万方数据