基于机器学习的缺陷咖啡生豆检测

Detection of Defective Green Coffee Beans Based on Machine Learning

为实现缺陷咖啡生豆无损、快速及准确检测,提高咖啡生豆品质及咖农经济效益,提出基于局部二值模式(LBP)的多模型Stacking集成缺陷咖啡生豆检测方法。该方法采用机器视觉技术提取咖啡生豆(8,1)和(16,2)2种尺度、3种算子(统一算子、旋转不变算子、旋转不变统一算子)下的LBP特征向量,并联合同种算子下不同尺度的LBP特征,选用LightGBM、XGBoost、CatBoos、SVM作为基分类器,Logistics作为次级学习器进行Stacking模型集成。结果表明,采用(8,1)尺度统一算子Stacking集成模型检测的准确率和F1值分别为91.9%、92.3%,均高于其他尺度、算子和不同类型特征的检测模型。与LightGBM、XGBoost、CatBoost、SVM相比,Stacking集成检测模型的准确率分别提高了0.6%、1.7%、2.0%、1.2%,整体检测性能更优。

To achieve nondestructive,rapid,and accurate detection of defective green coffee beans,improve the quality of green coffee beans and the economic benefits of coffee farmers,a multi model Stacking integrated defect detection method based on local binary mode(LBP)is proposed.The method adopts machine vision technology to extract the LBP feature vectors of raw coffee beans at two scales(8,1)and(16,2)and three operators(uniform operator,rotation invariant operator,rotation invariant uniform operator),and combines the LBP features at different scales under the same operator,and selects LightGBM,XGBoost,CatBoos and SVM as the base classifiers,and Logistics as the secondary learner for Stacking model integration.The results show that the accuracy and F1 value of using the(8,1)scale unified operator Stacking integrated model for detection are 91.9%and 92.3%,respectively,which are higher than detection models with other scales,operators,and different types of features.Compared with LightGBM,XGBoost,CatBoost,and SVM,The accuracy of the Stacking integrated detection model has been improved by 0.6%,1.7%,2.0%,and 1.2%respectively,resulting in better overall detection performance.