| 摘要: |
| 本文以玉米叶斑病检测为研究对象,针对现有目标检测模型普遍存在的计算复杂度高、部署困难等问题,基于YOLOv8算法提出了一种改进方案。主要创新点包括：构建DWCA轻量化注意力模块提升特征提取能力,引入SIoU损失函数优化目标框定位精度,采用基于BatchNorm层的模型剪枝策略降低计算复杂度。在robflow玉米病虫害数据集上的实验结果表明,改进后的模型检测精度mAP50达到88.8%,较原始YOLOv8提升0.7个百分点；同时模型参数量和计算量分别减少33.1%和31.4%,推理速度提升20.5%。该方法在保持较高检测精度的同时显著提升了模型效率,为农作物病虫害智能检测提供了新的技术思路。 |
| 关键词: YOLOv8算法 玉米叶斑病检测 DWCA注意力模块 SIoU损失函数 模型剪枝 轻量化网络 |
| DOI： |
| 分类号:TP18 |
| 基金项目:国家自然科学基金青年基金(62302307)；广西人机交互与智能决策重点实验室开放(GXHIID2209) |
|
| A review of structural light 3D imaging technology |
|
Liu Ennian1, Li Luqun1, Li Shuang2
|
|
1.College of Information,Mechanical and Electrical Engineering,Shanghai Normal University;2.College of Information, Mechanical and Electrical Engineering, Shanghai Normal University
|
| Abstract: |
| In this paper, corn leaf spot detection is taken as the research object, and an improved scheme based on YOLOv8 algorithm is proposed to solve the problems of high computational complexity and difficult deployment in existing target detection models. The main innovations include: constructing DWCA lightweight attention module to improve feature extraction ability, introducing SIoU loss function to optimize the positioning accuracy of target frame, and adopting model pruning strategy based on BatchNorm layer to reduce computational complexity. The experimental results on robflow corn pest data set show that the detection accuracy of the improved model mAP50 reaches 88.8%, which is 0.7 percentage points higher than that of the original YOLOv8. At the same time, the parameters and calculation of the model are reduced by 33.1% and 31.4% respectively, and the reasoning speed is increased by 20.5%. This method significantly improves the model efficiency while maintaining high detection accuracy, and provides a new technical idea for intelligent detection of crop diseases and insect pests. |
| Key words: YOLOv8 algorithm Detection of maize leaf spot DWCA attention module SIoU loss function Model pruning Lightweight network |
