摘要
果园黄桃采收的成熟度决定其鲜食与加工价值,果园实时精准检测对产业链质量管控至关重要。针对现有模型计算复杂、难以在移动设备中进行实时部署的问题,提出一种基于改进YOLOv8s的轻量化黄桃成熟度检测方法。在主干网络中采用DWR-DRB替代C2f模块,增强模型在复杂背景下对目标特征的提取能力,提高检测精度;颈部网络用Slim-Neck替换原特征融合层,构建密集多尺度特征融合与下采样路径;将原检测头替换为轻量级检测头LSCD,引入共享卷积机制并有效减少参数量,增强特征图间的全局信息融合能力,使模型在复杂果园环境中兼顾检测速度与多尺度的识别精度。结果表明,优化后的模型平均检测精度达99.2%,与原模型比,检测精度提高了2.9%,计算量降低了23.7%,模型参数减小了26.1%。该模型在保持高精度的同时实现了轻量化,为果园环境智能化、实时化的成熟度分选提供了有效的技术方案。
The ripeness of yellow peach harvest determines its fresh food and processing value,real-time accurate detection of orchard is critical to quality control of industrial chain.Aiming at the complexity of existing model calculation and difficult to deploy in mobile devices in real time,a lightweight yellow peach ripeness detection method based on improved YOLOv8s is proposed.DWR-DRB is used to replace C2f module in backbone network to enhance the extraction ability of target features under complex background and improve detection accuracy.The neck network replaces the original feature fusion layer with Slim-Neck to construct dense multi-scale feature fusion and down-sampling paths;replaces the original detection head with lightweight detection head LSCD,introduces shared convolution mechanism,and effectively reduces the number of parameters,and the global information fusion ability among feature maps is enhanced,so that the model can give consideration to detection speed and multi-scale recognition accuracy in complex orchard environment.The results showed that the average detection accuracy of the optimized model reached 99.2%,and compared with the original model,the detection accuracy was improved by 2.9%,and the calculation amount decreased by 23.7%;the model parameters decreased by 26.1%.This model achieves lightweight while maintaining high accuracy,providing an effective technical solution for intelligent and real-time ripeness sorting in orchard environments.
作者
胡鑫凤
王佳明
杨柳
宋少云
曹梅丽
HU Xinfeng;WANG Jiaming;YANG Liu;SONG Shaoyun;CAO Meili(School of Mechanical Engineering,Wuhan Polytechnic University,Wuhan 430023,China;Hubei Grain and Oil Machinery Engineering Technology Research Center,Wuhan 430023,China)
出处
《武汉轻工大学学报》
2025年第6期65-73,102,共10页
Journal of Wuhan Polytechnic University
基金
湖北省自然科学基金项目(编号:2022CFB944)
江苏省食品先进制造重点实验室开放基金(编号:FM-202103)
湖北省重点研发(编号:2022BBA0047)。
