基于改进YOLOv8卷积神经网络的蟹味菇检测方法  

Crab Flavored Mushroom Detection Method Based on Improved YOLOv8 Convolutional Neural Network

作　　者：林宗缪[1] 马超[2] 胡冬 LIN Zongmiao;MA Chao;HU Dong(Shanghai Institute of Quality Inspection and Technical Research,Shanghai 201114,China;Agricultural Information Institute of Science and Technology,Shanghai Academy of Agricultural Science,Shanghai 201403,China)

机构地区：[1]上海市质量监督检验技术研究院,上海201114 [2]上海市农业科学院农业科技信息研究所,上海201403

出　　处：《农业工程》2024年第3期27-31,共5页AGRICULTURAL ENGINEERING

基　　金：上海市市场监督管理局科研计划项目(2022-52)。

摘　　要：针对蟹味菇生产过程中更好地预估产量,对生长状态做到实时检测的问题,提出了一种基于改进YOLOv8卷积神经网络的蟹味菇识别检测方法。该方法参照PASCAL VOC数据集格式,构建了蟹味菇目标检测数据集,采用添加CBAM注意力机制对原算法进行改进,并且与Faster R-CNN、SSD(single shot multibox detector)、原始YOLOv8等算法进行模型性能的试验对比。试验结果表明,改进的算法明显优于其他算法,其在测试集上的平均精度均值(mean average precision,mAP)和检测速度分别达到95%和91帧/s。此检测精度与检测时间满足蟹味菇的实时识别检测任务,为预估蟹味菇产量,提高生产管理水平提供了理论技术支持。A crab flavored mushroom recognition and detection method based on improved YOLOv8 convolutional neural network was proposed to address issue of better yield estimation and real-time detection of growth status in production process of crab flavored mush-rooms.This method refered to PASCAL VOC dataset format and constructed a crab flavored mushroom target detection dataset.Origin-al algorithm was improved by adding CBAM attention mechanism,and model performance was compared with Faster R-CNN,SSD(single shot multibox detector),original YOLOv8 and other algorithms for experimental testing.Experimental results showed that improved algorithm was significantly superior to other algorithms,with an mean average precision(mAP)and detection speed of 95%and 91 frames/s on the test set,respectively.This detection accuracy and detection time met real-time recognition and detection task of crab flavored mushrooms,providing theoretical and technical support for estimating yield of crab flavored mushrooms and im-proving production management level.

关 键 词：YOLOv8 卷积神经网络 蟹味菇 目标检测 CBAM 

分 类 号：S126[农业科学—农业基础科学]