MIP-YOLO：显微图像法注射液中不溶性微粒的检测模型

doi:10.3778/j.issn.1002-8331.2405-0323

摘要/Abstract

摘要： 准确检测注射液中的不溶性微粒种类和数量对于提高注射液质量具有重要作用。针对不溶性微粒中的小目标检测难题，提出了一种基于YOLOv8s的改进模型MIP-YOLO（microscopic insoluble particles-YOLO），用于准确检测不溶性微粒中的玻璃、橡胶等特征微粒。在颈部网络中使用部分卷积，引入重参数化模块对部分卷积进行改进，提出了DPC2f模块，增加梯度流动能力，降低漏检、误检率，并结合三重注意力机制提出了DPTC2f模块，在减少冗余信息的同时，增强模型在不同维度的特征提取能力。引入NWD（normalized Wasserstein distance）损失函数替换CIoU损失函数，降低对不溶性微粒位置偏差的敏感性。优化了颈部网络检测层结构，提升浅层信息和深层信息的融合能力，降低模型的参数量。设计了协同池化坐标注意力机制，提高模型对微粒显著特征的提取能力。实验结果表明，MIP-YOLO的精度和mAP分别达到了89.0%和90.5%，在参数量下降40%的同时，相比YOLOv8s，精度和mAP分别提升了5.4和3.0个百分点，提高了模型的检测效果。

关键词: 不溶性微粒, 显微图像, 特征检测, YOLOv8s, 小目标检测, 重参数化

Abstract: Accurately identifying the types and quantities of insoluble particles in injection plays an important role in improving the quality of injection. To address the challenge of detecting small targets within insoluble particles, this paper proposes an improved model based on YOLOv8s, named MIP-YOLO (microscopic insoluble particles-YOLO), which is designed for precise identification of characteristic particles in insoluble substances. The model employs partial convolutions in the neck network and introduces a reparameterization module to enhance these convolutions, resulting in the DPC2f module which increases gradient flow capability, thereby reducing false negatives and false positives. The DPTC2f module, incorporating a triple attention mechanism, is introduced to reduce redundant information and enhance the feature extraction ability of the model across different scales. The NWD (normalized Wasserstein distance) loss function is introduced to replace the CIoU loss function, reducing sensitivity to positional deviations of insoluble particles. The structure of the detection layers in the neck network is optimized to improve the fusion of shallow and deep information, thereby reducing the parameter count of the model. A synergistic pooling coordinate attention mechanism is proposed to enhance the extraction of significant features of particles. The experimental results show that the accuracy of MIP-YOLO is 89.0%, and the mAP is 90.5%, with a 40% reduction in the number of parameters, the model shows an improvement of 5.4 percentage points in precision and 3.0 percentage points in mAP compared to YOLOv8s, significantly enhancing the detection performance of the model.

Key words: insoluble particles, microscopic image, feature detection, YOLOv8s, small target detection, reparameterization

杨达, 刘伟, 王雅静, 张宗强, 秦福元. MIP-YOLO：显微图像法注射液中不溶性微粒的检测模型[J]. 计算机工程与应用, 2025, 61(16): 272-282.

YANG Da, LIU Wei, WANG Yajing, ZHANG Zongqiang, QIN Fuyuan. MIP-YOLO: Detection Model of Insoluble Particles in Injection Using Microscopic Images[J]. Computer Engineering and Applications, 2025, 61(16): 272-282.

参考文献

[1] 肖美慧, 席明霞, 屈婧, 等. 全闭环输液监测系统在临床输液安全管理中的应用[J]. 护理学报, 2020, 27(3): 14-17.
XIAO M H, XI M X, QU J, et al. Application of closed-loop infusion monitoring system in safety management of clinical infusion[J]. Journal of Nursing (China), 2020, 27(3): 14-17.
[2] LANGILLE S E. Particulate matter in injectable drug products[J]. PDA Journal of Pharmaceutical Science and Technology, 2013, 67(3): 186-200.
[3] LI P H, LI Q C, LAI Y J, et al. Direct entry of micro(nano)plastics into human blood circulatory system by intravenous infusion[J]. iScience, 2023, 26(12): 108454.
[4] HUANG Z, PENCZEK P A. Application of template matching technique to particle detection in electron micrographs[J]. Journal of Structural Biology, 2004, 145(1/2): 29-40.
[5] ZHAN Y T, ZHANG G Y. An improved Otsu algorithm using histogram accumulation moment for ore segmentation[J]. Symmetry, 2019, 11(3): 431.
[6] WANG J Q, ZHANG L, LU F X, et al. The segmentation of wear particles in ferrograph images based on an improved ant colony algorithm[J]. Wear, 2014, 311(1/2): 123-129.
[7] OKTAY A B, GURSES A. Automatic detection, localization and segmentation of nano-particles with deep learning in microscopy images[J]. Micron, 2019, 120: 113-119.
[8] HU Z C, LIU J, JIANG C, et al. A high-precision detection method for coated fuel particles based on improved faster region-based convolutional neural network[J]. Computers in Industry, 2022, 143: 103752.
[9] ZHANG H, ZHAO M, LIU L, et al. Deep multimodel cascade method based on CNN and random forest for pharmaceutical particle detection[J]. IEEE Transactions on Instrumentation and Measurement, 2020, 69(9): 7028-7042.
[10] LYU H, XU F X, JIN T, et al. Automated detection of multi-class urinary sediment particles: an accurate deep learning approach[J]. Biocybernetics and Biomedical Engineering, 2023, 43(4): 672-683.
[11] SANBORN D, HE R C, FENG L, et al. In situ biological particle analyzer based on digital inline holography[J]. Biotechnology and Bioengineering, 2023, 120(5): 1399-1410.
[12] WANG Z H, ZHOU D, GUO C J, et al. YOLO-Global: a real-time target detector for mineral particles[J]. Journal of Real-Time Image Processing, 2024, 21(3): 85.
[13] SZEGEDY C, LIU W, JIA Y Q, et al. Going deeper with convolutions[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2015: 1-9.
[14] DING X H, ZHANG X Y, HAN J G, et al. Diverse branch block: building a convolution as an inception-like unit[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 10881-10890.
[15] CHEN J R, KAO S H, HE H, et al. Run, don??t walk: chasing higher FLOPS for faster neural networks[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 12021-12031.
[16] MISRA D, NALAMADA T, ARASANIPALAI A U, et al. Rotate to attend: convolutional triplet attention module[C]//Proceedings of the 2021 IEEE Winter Conference on Applications of Computer Vision. Piscataway: IEEE, 2021: 3138-3147.
[17] ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 12993-13000.
[18] WANG Z, HUA Z X, WEN Y C, et al. E-YOLO: recognition of estrus cow based on improved YOLOv8n model[J]. Expert Systems with Applications, 2024, 238: 122212.
[19] WANG J W, XU C, YANG W, et al. A normalized Gaussian Wasserstein distance for tiny object detection[J]. arXiv:2110. 13389, 2021.
[20] WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of the 15th European Conference on Computer Vision. Cham: Springer, 2018: 3-19.
[21] HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[C]//Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 13708-13717.
[22] GUO Y C, ZHANG M Y. Blood cell detection method based on improved YOLOv5[J]. IEEE Access, 2023, 11: 67987-67995.
[23] CAI X H, LAI Q X, WANG Y W, et al. Poly kernel inception network for remote sensing detection[C]//Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2024: 27706-27716.
[24] CHEN Y F, ZHANG C Y, CHEN B, et al. Accurate leukocyte detection based on deformable-DETR and multi-level feature fusion for aiding diagnosis of blood diseases[J]. Computers in Biology and Medicine, 2024, 170: 107917.