基于机器学习的速冻水饺缺陷检测  

Defect detection of quick⁃frozen dumpling based on machine learning

作　　者：刘一阳 马子领 Liu Yiyang;Ma Ziling(Mechanical College of North China University of Water Resources and Electric Power,Zhengzhou 450045,China)

机构地区：[1]华北水利水电大学机械学院,郑州450045

出　　处：《现代计算机》2023年第24期1-9,39,共10页Modern Computer

摘　　要：速冻水饺在冷冻过程中,容易造成冻裂、破损的现象,采用人工方法对破损水饺进行剔除耗时耗力,因此,设计一种基于机器学习的速冻水饺缺陷检测方法。首先,分别提取R,G,B,H,S,V颜色空间的灰度共生矩阵(GLCM)和颜色矩共20维特征向量,其次采用主成分分析法(PCA)对特征向量进行优化,得到6个最优特征向量作为支持向量机输入,采取交叉验证超参数寻优确定支持向量机参数,最后用训练好的支持向量机模型对缺陷饺子进行分类识别,对有缺陷的饺子外接最小矩形,并输出外接矩形的中心点坐标。结果证明,该方法可以对有缺陷的饺子进行快速识别,识别准确率为95.5%,平均每个饺子的检测耗时为13.13ms。In the freezing process,quick‑frozen dumplings are easy to cause freeze cracking and damage,and it is time‑consuming and labor‑intensive to manually remove broken dumplings,so designing machine learning‑based defect detection method for quick‑frozen dumplings.First,the grayscale coexistence matrix and color moment of R,G,B,H,S,V color space are extracted with a total of 20‑dimensional eigenvectors,which are optimized by the method of PCA.Six optimal eigenvectors are obtained as the input of the support vector machine.The parameters of the support vector machine are determined by the cross‑algorithm hyperparametric optimization,and the defective dumplings are classified and identified by the trained support vector machine model.The defective dumplings are externally connected with the minimum circumscribed rectangles,and the center point coordinates of the minimum circumscribed rectangle are output.The results show that the method can quickly identify the defective dumplings with the recognition accuracy of 95.5%and the time‑consuming of 13.13 ms per dumping.

关 键 词：速冻食品缺陷检测 灰度共生矩阵 PCA 支持向量机 

分 类 号：TS207.3[轻工技术与工程—食品科学] TP181[轻工技术与工程—食品科学与工程] TP391.41[自动化与计算机技术—控制理论与控制工程]