Liquid Level Detection of Test Tube Sample Based on YOLOv5
WU Zhuqing1, TAN Fengfu2, LI Dongxiu3, TANG Zhipeng4
1. Guilin URIT Medical Electronic Co., Ltd., Guilin Guangxi 541004, China; 2. Key Laboratory of Atmospheric Optics,
Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Hefei Anhui 230031, China;
3. Department of Emergency, Affiliated Hospital of Guilin Medical University, Guilin Guangxi 541001, China;
4. Guilin STRAS Science and Technology Co., Ltd., Guilin Guangxi 541004, China
Abstract：Objective To design a liquid level detection algorithm of test tube sample based on YOLOv5 for complex and changeable
practical application scenarios, due to the defect of low recognition rate of the algorithm based on traditional image processing
technology. Methods Firstly, Labelme was used to label the test tube liquid level data set, the format of json annotation files were
converted, the YOLOv5 network model training environment was built, the pre-training model configuration files were modified,
the model weight data migration training was started, the optimal model parameters and the iteration change curves of the evaluation
index were output. Results The test data set was loaded to test the effect of the training optimal model algorithm. The algorithm
could recognize the sample liquid levels for both normal and abnormal sample images. Through the aging comparison test with
the recognition algorithm based on traditional image processing technology, it had higher recognition accuracy and precision.
Conclusion The liquid level detection algorithm of test tube sample based on YOLOv5 has high detection accuracy and strong robust
generalization ability, which can meet the actual detection requirements of medical equipment, and has a good market application
prospect.
吴祝清1，谭逢富2，李冬秀3，唐智鹏4. 基于YOLOv5的试管样本液位检测[J]. 中国医疗设备, 2023, 38(4): 61-67.
WU Zhuqing1, TAN Fengfu2, LI Dongxiu3, TANG Zhipeng4. Liquid Level Detection of Test Tube Sample Based on YOLOv5. China Medical Devices, 2023, 38(4): 61-67.
