基于机器学习算法的缺损米粉块在线快速检测

目的:实现缺损米粉块的快速在线检测。方法:提出运用机器学习算法对缺损米粉块检测数据进行分析。通过相机对米粉块进行非接触式数据采集,图像上传及处理后,获取米粉块的轮廓周长和面积、近似轮廓的周长和面积、近似轮廓点数、轮廓外接圆半径6个特征数据,依据米粉块样本数据特点,运用支持向量机(SVM)分类算法对米粉块的多特征数据组成的样本集进行分析。结果:通过与5种算法测试对比,GBDT分类算法平均准确率89%,用时1.10 s;KNN分类算法平均准确率88%,用时0.23 s;Logistic Regression分类算法平均准确率88%,用时0.68 s;Random Forest分类算法平均准确率87%,用时0.47 s;tree分类算法平均准确率87%,用时0.084 s;SVM分类算法检测平均准确率最高,达95%,平均用时最短,为0.000 97 s。结论:利用SVM分类算法进行米粉块缺损检测准确率高、用时短,适用于生产线的在线检测。

On line fast detection of defective rice flour based on machine learning algorithm

Objective: To realize the rapid on-line detection of defective rice flour. Methods: Non-contact data acquisition of rice flour blocks through cameras, image upload and processing, obtained the contour perimeter and area, approximate contour perimeter and area, approximate contour points and radius of the contour circle. According to the characteristics of rice flour block sample data, the SVM classification algorithm was used to analyze the sample set composed of multi feature data of rice flour block. Results: Compared with five algorithms, the average accuracy of GBDT classification algorithm was 89% with elapsed time of 1.10 s. The average accuracy of KNN classification algorithm was 88% with elapsed time of 0.23 s. The average accuracy of logistic regression classification algorithm was 88% with elapsed time of 0.68 s. The average accuracy of random forest classifica-tion algorithm was 87% with elapsed time of 0.47 s. The average accuracy of tree classification algorithm was 87% with elapsed time of 0.084 s. SVM classification algorithm had the highest average detection accuracy, up to 95%, and the shortest average elapsed time of 0.000 97 s. Conclusion: SVM classification algorithm has the characteristics of high accuracy and low elapsed time, which adapt the on-line detection of defective rice flour.