| 摘 要: 杂草是百合生长过程中的一大危害,会干扰百合生长并吸收其营养,导致产量下降。文章以百合及其伴生杂草为主要研究对象,将 YOLOv8模型引入百合与杂草的检测中,并进行了针对性的改进。首先,构建基于BiFormer双 层 路 由 注 意 力 机 制 的 C2f_BF 模 块;其 次,在 头 部 网 络 Neck 端 引 入 GSConv(Grouped Shuffle Convolution)和Slim-neck(轻量化特征融合网络)技术;最后,使用 MPDIoU(Multi-Perspective Distance)损失函数克服CIoU(Complete Intersection over Union)损失函数的局限性。实验结果表明,改进后的YOLOv8-LWD(Lily Weed Detection)模型的平均精确率为90.3%,相比于原始 YOLOv8n检测模型的平均精确率提升了2.9百分点。该方法可以为百合草害防治提供重要的技术支持,具有实际的应用价值。 |
| 关键词: 百合;杂草检测;YOLOv8模型;卷积神经网络 |
|
中图分类号: TP39
文献标识码: A
|
| 基金项目: 自然科学基金-甘肃省科技计划资助(24JRRA656);2022年横向课题:农产品物资销售模式的数据统计分析(loonG20220201) |
|
| Weed Detection Method in Lily Fields Based on Improved YOLOv8 Model |
|
WANG Yao1, ZHAO Xia1, CHENG Hong2
|
(1.College of Information Science and Technology, Gansu Agricultural University, Lanzhou 730070, China;
2.Vegetable Research Institute, Gansu Academy of Agricultural Sciences, Lanzhou 730070, China)
18693428559@163.com; 58892778@qq.com; chenjn@yeah.net
|
| Abstract: Weeds pose a significant threat to the growth of lilies, interfering with their development and absorbing their nutrients, leading to reduced yields. This study focuses on lilies and their associated weeds, introducing the YOLOv8 model into the detection of lilies and weeds and implementing targeted improvements to the model. Firstly, a C2f_BF module based on the BiFormer dua-l layer routing attention mechanism is constructed. Secondly, the head network incorporates GSConv (Grouped Shuffle Convolution) and Slim-neck (lightweight feature fusion network) technologies at the Neck end. Lastly, the MPDIoU (Mult-i Perspective Distance) loss function is employed to overcome the limitations of the CIoU (Complete Intersection over Union) loss function. Experimental results demonstrate that the improved YOLOv8-LWD (Lily Weed Detection) model achieves an average precision rate of 90.3% , representing a 2.9 percentage points improvement over the original YOLOv8n detection model. This method provides significant technical support for lily weed control and has practical application value. |
| Keywords: lily; weed detection; YOLOv8 model; Convolutional Neural Network (CNN) |
