| 陈教料,王振舵,潘立.基于INC4-YOLO的菌落计数方法研究[J].高技术通讯(中文),2025,35(8):901~910 |
| 基于INC4-YOLO的菌落计数方法研究 |
| A counting method for colony forming unit based on INC4-YOLO |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 08. 009 |
| 中文关键词: 菌落计数； 目标检测； 改进YOLOv5； Inception模块； 小目标检测 |
| 英文关键词: colony forming unit counting, object detection, improved YOLOv5, Inception module, small object detection |
| 基金项目: |
| 作者 | 单位 | | 陈教料 | (浙江工业大学机械工程学院杭州 310012) | | 王振舵 | | | 潘立 | |
|
| 摘要点击次数: 74 |
| 全文下载次数: 68 |
| 中文摘要: |
| 针对菌落图像中小菌落易漏检的问题，提出了一种基于INC4-YOLO(you only look once)的计数方法，实现精准的菌落计数。采用带残差结构的Inception模块(Inception module with residual connection, IncRes)替换YOLOv5骨干网络中的Bottleneck模块，以增强图像特征提取能力。从网络的浅层特征中引出一个小目标检测头，以增强算法在训练过程中对小菌落的注意力。分别在标注微生物自动识别数据集(annotated germs for automated recognition, AGAR)和真实菌落计数场景下对INC4-YOLO进行计数性能测试。实验结果表明，在AGAR测试集中，提出的算法在小菌落的平均百分比绝对值计数误差(mean absolute percentage error, MAPE)比其他先进目标检测算法降低了2%；真实菌落计数场景下，INC4-YOLO的MAPE相比YOLOv5降低了7%，表明该算法可帮助菌落计数设备实现精准计数。 |
| 英文摘要: |
| An INC4-YOLO-based counting method for colony forming unit is proposed to handle miss detection of small colony forming units and improve counting accuracy. The Bottleneck module in backbone network of YOLOv5 is replaced by an Inception module with residual connection (IncRes) to enhance image feature extraction capability. An additional small object detection head for the early-layer feature is used to focus on small colonies during network training. The experiments are conducted on test set of annotated germs for automated recognition (AGAR) dataset and real bacterial colonies scenario respectively. The results on AGAR demonstrate that INC4-YOLO outperforms other advanced detection algorithms with a 2% decrease in mean absolute percentage error (MAPE) for small bacterial colonies; In real colony counting scenario, the MAPE of INC4-YOLO surpasses that of YOLOv5 by 7%.INC4-YOLO can be a useful tool for bacterial colony counting equipment to count bacterial colony accurately. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
