| 摘 要: 为了有效应对玉米地杂草对玉米产量和品质的影响,实现玉米与杂草的快速、准确检测,提出了一种基于改进YOLOv8n(You Only Look Once Version 8 nano)的玉米与杂草检测模型。首先,提出了ACMConv(Accurate and Computationally Minimal Convolution)新型卷积方式,显著减少了模型计算量,使模型更加轻量化;其次,使用SELU激活函数,引入非线性因素,有效缓解了梯度消失问题;最后,引入FocalLoss作为边界框损失函数,使模型更加容易收敛。实验结果表明,相较于原始 YOLOv8n模型,改进后的 YOLOv8n模型的平均精度均值提升了1.3百分点,计算量降低了7.3%,实现了对玉米与杂草的高效、准确检测。 |
| 关键词: 深度学习;杂草识别;YOLOv8n;激活函数;FocalLoss |
|
中图分类号: TP391.41
文献标识码: A
|
| 基金项目: 贵州省科技计划[黔科合支撑(2021)一般176] |
|
| Weed Detection in Corn Fields Based on Improved YOLOv8n |
|
WEN Tao, WANG Tianyi
|
(School of Big Data and Information Engineering, Guizhou University, Guiyang 550025, China)
1815714855@qq.com; tywang@gzu.edu.cn
|
| Abstract: In order to effectively address the impact of weeds on corn yield and quality, and to achieve rapid and accurate detection of corn and weeds, this paper proposes a corn and weed detection model based on the improved YOLOv8n (You Only Look Once Version 8 nano). Firstly, a novel convolution method called ACMConv (Accurate and Computationally Minimal Convolution) is introduced, which significantly reduces the computational load of the model, making the model more lightweight. Secondly, the SELU activation function is used to incorporate non-linear factors, effectively alleviating the vanishing gradient problem. Finally, Focal Loss is introduced as the bounding box loss function to facilitate the convergence of the model. Experimental results show that, compared to the original YOLOv8n model, the improved YOLOv8n model achieves a mean Average Precision increase of 1.3 percentage points and a 7.3% reduction in computational load, realizing efficient and accurate detection of corn and weeds. |
| Keywords: deep learning; weed recognition; YOLOv8n; activation function; Focal Loss |
