基于改进YOLOv7的微生物细胞识别算法  

Microbial cell recognition algorithm based on improved YOLOv7

作　　者：吕彦朋 赵颖彤 苏晓明 刘占英 LÜYanpeng;ZHAO Yingtong;SU Xiaoming;LIU Zhanying(Inner Mongolia Autonomous Region Fermentation Industry Energy Saving and Emission Reduction Engineering and Technology Research Center,Hohhot 010051,China;Inner Mongolia Autonomous Region Engineering Research Center for Green Manufacturing of Bio-fermentation,Hohhot 010051,China;College of Data Science and Application,Inner Mongolia University of Technology,Hohhot 010051,China)

机构地区：[1]内蒙古自治区发酵产业节能减排工程技术研究中心,内蒙古呼和浩特010051 [2]生物发酵绿色制造内蒙古自治区工程研究中心,内蒙古呼和浩特010051 [3]内蒙古工业大学数据科学与应用学院,内蒙古呼和浩特010051

出　　处：《现代电子技术》2025年第1期47-54,共8页Modern Electronics Technique

基　　金：国家自然科学基金项目(32060017);国家级大学生创新创业训练计划项目(202310128011);内蒙古自治区杰出青年基金项目(2022JQ10);内蒙古草原英才团队滚动支持项目;内蒙古自治区科技重大专项(2021ZD0015)。

摘　　要：针对传统及人工方法识别多种类、大量微生物细胞存在耗时长、准确率低等问题,文中提出一种改进YOLOv7的微生物细胞识别算法YOLOv7-PN。通过引入改进的路径聚合网络(PANet)提取和融合不同尺度的特征,以捕捉细胞图像中的多尺度信息,从而提高细胞的检测精度和鲁棒性;在骨干网络添加注意力模块(NAM),能够自适应地学习每个通道的权重,提高细胞的特征表示能力;将传统的IoU边界框损失函数替换为DIoU_Loss,以考虑边界框之间的距离和重叠程度,能够更准确地衡量检测框的精度,从而提高细胞的定位准确性。实验结果表明,使用BCCD数据集进行评估时,文中算法在微生物细胞识别任务中取得了显著的提升,与基准算法YOLOv7相比,YOLOv7-PN的Precision值提高了1.46%、F_(1)值提高了2.61%、Accuracy值提高了0.86%。实验结果验证了该算法的有效性和性能优势,为微生物学研究和医学诊断等领域的微生物细胞分析提供了有力支持。It is time-consuming when the traditional and manual methods are used to identify multiple species and large quantity of microbial cells,and the identification accuracy is low.In view of this,a microbial cell recognition algorithm named YOLOv7-PN based on the improved YOLOv7 is proposed.The improved path aggregation network(PANet)is introduced to enable the extraction and fusion of features at different scales in order to facilitate the capture of multi-scale information in cell images,which enhances its detection accuracy and robustness of cells.The incorporation of NAM into the backbone network enables the adaptive learning of weights for each channel,which enhances the feature representation of cells.Furthermore,the replacement of the traditional IoU bounding box loss function with DIoU_Loss allows for the consideration of the distance and overlap between bounding boxes,thereby facilitating more accurate precision measurement,which,in turn,enhances the accuracy of cell localization.The experimental results demonstrate that the algorithm presented in this paper exhibits a notable enhancement in the capacity to recognize microbial cells when evaluated with the BCCD dataset.In comparison to the benchmark algorithm YOLOv7,the YOLOv7-PN demonstrates an improvement.Its precision is increased by 1.46%,its F_(1)by 2.61%and its accuracy rate by 0.86%.The experimental results demonstrate the efficacy and superiority of the algorithm.Therefore,the algorithm can provide compelling evidence for its utility in the microbial cell analysis in microbiology research and medical diagnosis,as well as in other fields.

关 键 词：微生物细胞 YOLOv7 YOLOv7-PN PANet NAM DIoU_Loss 

分 类 号：TN911.7-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]