摘要
为实现自然环境下对野生菌种类的精确识别,提出一种基于改进YOLOv8s的野生菌识别算法YOLOv8s—EAS。首先,使用EIE模块改进原C2f模块的Bottleneck结构,进一步捕获目标的空间信息和边缘特征信息,加强特征提取能力;其次,通过引入ADown模块,增强模型的特征提取能力;然后,提出一种融合MultiSEAM注意力机制的检测头,实现多尺度的野生菌检测,使模型能够更准确地检测出被遮挡情况下的野生菌,从而提升识别精确率;最后,替换损失函数为WIoU函数,加快模型的收敛,进一步提升模型的检测性能。结果表明,改进的YOLOv8s—EAS模型在野生菌识别任务中,其mAP@0.5和mAP@0.5∶0.95较原YOLOv8s模型分别提高2.1个百分点、1.6个百分点,精确率和召回率也分别提升5.2个百分点、1.3个百分点。对比目前主流的目标检测模型,YOLOv8s—EAS模型对自然环境下的野生菌识别精度具有明显优势,能够更好地满足实际农业需求。
To achieve precise recognition of wild mushroom species in natural environments,an improved YOLOv8s—based recognition algorithm named YOLOv8s—EAS was proposed.Firstly,the Bottleneck structure of the original C2f module was improved by using the EIE module to further capture spatial and edge feature information of the target,thereby strengthening feature extraction capability.Secondly,feature extraction capability was enhanced through the introduction of the ADown module.Thirdly,a detection head incorporating the MultiSEAM attention mechanism was proposed to achieve multi-scale wild mushroom detection,allowing the model to detect wild mushrooms in occluded conditions more accurately and thereby consequently improving recognition accuracy rate.Finally,the loss function was replaced with the WIoU function to accelerate model convergence and further enhance detection performance.Experimental results demonstrated that compared to the original YOLOv8s model,the improved YOLOv8s—EAS model achieved increases of 2.1 and 1.6 percentage points in mAP@0.5 and mAP@0.5∶0.95,respectively,on the wild mushroom recognition task.Precision and recall rates were also improved by 5.2 and 1.3 percentage points,respectively.Compared to current mainstream object detection models,the YOLOv8s—EAS model demonstrated significant advantages in recognition accuracy for wild mushrooms in natural environments and can better meet practical agricultural requirements.
作者
向琪琪
陈中举
李嘉诚
李和平
许浩然
王亮
Xiang Qiqi;Chen Zhongju;Li Jiacheng;Li Heping;Xu Haoran;Wang Liang(School of Computer Science,Yangtze University,Jingzhou,434023,China;Jingzhou Civil Air Defense Command Information Support Center,Jingzhou,434000,China)
出处
《中国农机化学报》
北大核心
2026年第1期100-107,F0002,共9页
Journal of Chinese Agricultural Mechanization
基金
湖北省教育厅科学技术研究项目(B2021052)
中国高校产学研创新基金——新一代信息技术创新项目(2023IT269)。
关键词
边缘特征
野生菌识别
深度学习
目标检测
自然环境
edge features
wild mushroom recognition
deep learning
target detection
natural environment
