少量缺陷样本情形下医疗针管刻度质量检测技术研究

A Method for Detecting Defects in Medical Syringe Scales Using Limited Defective Samples

摘要:

针对医疗针管刻度缺陷检测存在样本收集难且缺陷种类不确定的问题，提出一种少量缺陷样本情形下的医疗针管刻度质量检测方法。利用真实生产线上采集的大量正常样本，训练深度刻度分割模型，通过以针管像素分块为单位构建拉普拉斯矩阵来挖掘针管刻度间相关性，并利用模糊C均值进行无监督缺陷检测。实验结果表明，针管刻度质量检测方法可100%检测出所有缺陷样本，对提高医疗针管生产质量具有一定理论和实践应用价值。

Abstract:

In the syringe scales defecting, it is difficult to collect samples and determine the type of defects. To overcome this issue, a method for detecting the quality of syringe scale with a small number of defective samples is proposed. This method leverages a large number of normal samples collected from an actual production line to train a deep scale segmentation model. The method constructs a Laplacian matrix based on pixel blocks of syringe scales to mine their correlation, and uses fuzzy c-means for unsupervised defect detection. Experimental results demonstrate that this syringe scale quality inspection method can detect all defective samples with 100% accuracy, effectively enhancing the production quality of medical syringes.