| 摘 要: 针对茄子果实病虫害特征复杂、自然环境下检测精度低和检测模型较大等问题,提出一种基于改进YOLOv8的轻量化网络模型HCI-YOLO。该模型以YOLOv8作为基础框架,将特征增强网络替换为多级特征融合金字塔(High-level Screening-feature Fusion Pyramid Network)结构,减少了网络参数以及解决了应对尺度问题;在骨干网络中的第11层嵌入坐标注意力机制(Coordinate Attention),提升了模型的回归性能;使用Inner-SIoU损失函数替代原损失函数,提升了模型的检测精度。实验结果表明,HCI-YOLO 模型参数量和计算量相比于原始版本YOLOv8n分别降低了36.67百分点和14.81百分点,精确度和mAP@0.5分别提高了1.3百分点和2.1百分点。这证明HCI-YOLO在茄子果实病虫害检测方面取得了显著的性能提升。 |
| 关键词: 图像识别;轻量化网络;注意力机制;多级特征融合;病虫害检测 |
|
中图分类号: TP391.4
文献标识码: A
|
|
| HCI-YOLO: A Detection Model for Eggplant Fruit Diseases and Pests Based on Improved YOLOv8 |
|
GUO Zhen, LI Yue
|
(School of Inf ormation Science and Technology, Gansu Agricultural University, Lanzhou 730070, China)
1020622618@qq. com; liyue@gsau.edu.cn
|
| Abstract: This paper proposes a lightweight network model, HCI-YOLO, based on improved YOLOv8 to address issues of the complex characteristics of eggplant fruit diseases and pests, low detection accuracy in natural environment, and the large size of detection models. This model uses YOLOv8 as the foundational framework and replaces the feature enhancement network with a High-level Screening-feature Fusion Pyramid Network structure. This modification reduces the network parameters and addresses scale issues. A Coordinate Attention mechanism is embedded in the 11th layer of the backbone network, enhancing the model ' s regression performance. Additionally, the Inner-SIoU loss function is used to replace the original loss function, improving the model's detection accuracy. Experimental results indicate that the number of parameters and the computational complexity have decreased by 36.67 percentage points and 14.81 percentage points compared to the original version YOLOv8n, while the accuracy and the mAP @ 0.5 have improved by 1.3 percentage points and 2.1 percentage points. This demonstrates that HCI-YOLO achieves significant performance improvements in the detection of eggplant fruit diseases and pests. |
| Keywords: image recognition; lightweight network; attention mechanism; multi-level feature fusion; pest and disease detection |
