• 首页
  • 期刊简介
  • 编委会
  • 投稿指南
  • 收录情况
  • 杂志订阅
  • 联系我们
引用本文:樊翔宇,代 琦.基于改进YOLOv5的菌落计数算法研究[J].软件工程,2024,27(10):34-38.【点击复制】
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  
←前一篇|后一篇→ 过刊浏览
分享到: 微信 更多
基于改进YOLOv5的菌落计数算法研究
樊翔宇, 代 琦
(浙江理工大学生命科学与医药学院, 浙江 杭州 310020)
1871541711@qq.com; daiqi@zstu.edu.cn
摘 要: 针对菌落计数问题,人工计数方法存在效率低、精度不高的问题。为了解决这些问题,提出了一种改进YOLOv5的模型,即YOLOES。该模型通过添加小目标检测层,并将Kmeans算法替换为Kmeans++算法,以更好地适应不同尺寸的目标;同时,采用Focal-EIoU损失函数解决难易样本的问题,引入了SPPCSPS(Spatial Pyramid Pooling Convolutional Spatial Pyramid Convolution)模块以增强特征表示能力,并在特征提取阶段引入了置换注意力机制。通过在大肠杆菌菌落数据集进行实验验证,结果显示相较于初始的YOLOv5模型,YOLOES的mAP@0.5提升了17.3百分点,表明YOLOES在菌落检测任务上具有更优越的性能。
关键词: YOLOv5;图像识别;Kmeans++;Focal-EIoU;SPPCSPS;置换注意力机制
中图分类号: TP391    文献标识码: A
基金项目: 浙江省高层次人才特殊支持计划(2021R52019)
Research on Colony Counting Algorithm Based on Improved YOLOv5
FAN Xiangyu, DAI Qi
(College of Li f e Sciences and Medicine, Zhejiang SCI-TECH University, Hangzhou 310020, China)
1871541711@qq.com; daiqi@zstu.edu.cn
Abstract: Aiming at the low efficiency and low accuracy of manual colony counting, this paper proposes an improved model based on YOLOv5, named YOLOES. This model incorporates a small object detection layer and replaces the Kmeans algorithm with the Kmeans + + algorithm to better accommodate targets of various sizes. Additionally, it employs the Focal-EIoU loss function to tackle the problem of hard and easy samples, introduces the SPPCSPS ( Spatial Pyramid Pooling Convolutional Spatial Pyramid Convolution ) module to enhance feature representation capability, and integrates a permutation attention mechanism in the feature extraction phase. Experiments conducted on a dataset of Escherichia coli colonies indicate that YOLOES achieves a 17.3 percentage points improvement in mAP@0.5 compared to the original YOLOv5 model, demonstrating its superior performance in colony detection tasks.
Keywords: YOLOv5; image recognition; Kmeans++; Foca-l EIoU; SPPCSPS; permutation attention mechanisms


版权所有:软件工程杂志社
地址:辽宁省沈阳市浑南区新秀街2号 邮政编码:110179
电话:0411-84767887 传真:0411-84835089 Email:semagazine@neusoft.edu.cn
备案号:辽ICP备17007376号-1
技术支持:北京勤云科技发展有限公司

用微信扫一扫

用微信扫一扫