摘要
针对复杂背景下小目标草莓检测识别率低的问题,本文提出一种改进的YOLOv8n模型,来提升草莓目标识别的精度。首先在模型结构中引入SPD-Conv模块,用以增强模型对小物体和低分辨率图像的处理能力,提高在复杂场景下的鲁棒性,随后整合YOLOv10提出的PSA注意力机制,以低计算成本提升模型的全局表示学习能力,进一步增强模型性能,最后使用WIoU损失函数以替代CIoU损失函数,解决原损失函数的局限性。改进后的YOLOv8n模型相比原始模型,精度提升0.9个百分点,召回率增加4.3个百分点。此外,mAP0.5和mAP0.5:0.95分别提高了3个百分点和3.5个百分点。改进的YOLOv8n模型显著提升了草莓目标检测的精度,在复杂背景下对小目标的草莓具有更优异的检测表现。
Addressing the low recognition rate of small strawberry targets in complex backgrounds,this study proposed an improved YOLOv8n model to enhance the accuracy of strawberry target recognition.In the experimental process,the SPD-Conv module was incorporated into the model structure to improve the model's ability to handle small objects and low-resolution images,thereby increasing robustness in complex scenes.The PSA attention mechanism proposed by YOLOv10 was then integrated to embed global representation learning capability at a low computational cost,further enhancing model performance.Lastly,the WIoU loss function replaced the CIoU loss function to address its limitations.Compared to the original model,the improved YOLOv8n model achieved 0.9%increase in precision and 4.3%increase in recall.Additionally,mAP50 and mAP50-95 was improved by 3%and 3.5%,respectively.The improved YOLOv8n model significantly enhanced the accuracy of strawberry target detection and demonstrated superior detection performance for small strawberry targets in complex backgrounds.
作者
李东辉
余宏杰
LI Donghui;YU Hongjie(College of Mechanical Engineering,Anhui Science and Technology University,Fengyang 233100,China;College of Information and Network Engineering,Anhui Science and Technology University,Bengbu 233000,China)
出处
《安徽科技学院学报》
2025年第1期48-59,共12页
Journal of Anhui Science and Technology University
基金
安徽省重点研究与开发计划项目“基于区块链技术的乳制品质量安全监测关键技术的研究与应用”(202204c06020065)。
