基于机器视觉的红提串无损检测及分级

Non-destructive Firmness Detection and Grading of Bunches of Red Globe Grapes Based on Machine Vision

为实现整串红提的紧实度无损检测和分级,提出基于机器视觉检测的分级方法,通过工业摄像头采集新鲜红提串的彩色(red green and blue,RGB)和近红外图像,对整串红提RGB图像的三通道进行提取,采用归一化GB色差法提取红提图像中的果梗,运用形态学重构及局部亮度极大值方法定位红提串中的各个果粒中心,同时提取每串红提的质心,选取果梗面积、红提果粒个数与果串面积之比、红提果粒与红提质心距离之和与果粒个数之比等特征参数,分别建立基于线性判别分析、集成学习算法和支持向量机的紧实度分类模型,经检验支持向量机模型分类效果最佳,应用该模型对130串红提进行紧实度检测和分类,分级正确率94.6%。结果表明该方法可为后续葡萄品质及产量预测提供参考。

A machine vision-based method for the non-destructive firmness detection and grading of whole bunches of Red Globe grapes was proposed in this study.Red,green and blue(RGB)and near infrared images of the samples were acquired with an industrial camera.The three color channels in the RGB images were extracted,and the normalized GB color difference method was applied to locate the fruit stems based on the RGB images.Meanwhile,the center of mass of each fruit in a bunch was located by morphological reconstruction based on local maximum brightness and the center of mass of the whole bunch was extracted.The area of the fruit stalk,the ratio of the number of grapes to the area of the whole bunch of grapes,and the ratio between the sum of distance from the surface of each grape to the center of mass of the whole bunch of grapes and the number of grapes were selected as characteristic parameters to establish a classification model for predicting the firmness of grape bunches using linear discriminant analysis(LDA),integrated learning algorithm(ILA)or support vector machine(SVM).The results of validation showed that the SVM model had the best classification performance.This model was applied to 130 bunches of Red Globe grapes with a grading accuracy of 94.6%.These results will contribute to predicting grape quality and yield in the future.