摘要
为提高免套袋苹果分级效率,最小化果梗与花萼对缺陷区分的影响,提出了一种基于机器学习的免套袋苹果缺陷分级方法,该方法根据缺陷的数量和面积进行缺陷程度分级。获取免套袋苹果3个不同侧面的图像,利用固定阈值分割和形态学方法提取每个图像的苹果区域。根据苹果表面缺陷在HSV(Hue saturation value,色调、饱和度、明度)颜色空间的特征提取疑似缺陷区域,用种子填充法按序标记疑似缺陷区域,并计算每个区域的大小及灰度共生矩阵特征值。将特征值输入训练后的SVM(Support vector machine,支持向量机)模型,进行果梗、花萼与缺陷的区分,计算当前图像的缺陷数量与面积,再计算苹果3个不同侧面图像的总缺陷数量与面积,实现免套袋苹果缺陷分级。结果显示,正常区域、果梗区域、花萼区域在SVM模型中的分类正确率分别为96.7%、93.3%、88.3%。利用该缺陷分级方法对60个苹果进行分级的正确率为90.0%,满足苹果分级的实际生产需求。
In order to improve the efficiency of apple grading,a grading method based on machine learning for the defects detection of non-bagged apples was designed.The method can distinguish fruit stalks,calyxes and defects,reduce the mistake probability of categorizing the stalks and calyxes as defects,and classify defect level of apples according to the number and area of the defects.Three images on different sides were taken for each apple,threshold segmentation and morphological processing were performed to get the apple area of each image.The suspected defect areas were extracted according to the features of apple surface defects in hue saturation value color space.Suspected defect areas were labeled by seed filling method,and the number of pixels and the eigenvalues of gray level co-occurrence matrix of each area were also calculated.The eigenvalues were input into the trained support vector machine model to classify the stalks,calyxes and defects,and the real number and area of the defects could be calculated.The total number and area of defects in the images of the three different sides were combined to achieve apple defect classification.The results showed that the classification accuracy of normal area, stalk area,calyx area in support vector machine model were 96.7%,93.3% and 88.3%,respectively.To validate the veracity of the proposed method,60 apples were tested and the grading accuracy was 90.0% which meets the actual production needs of apple grading.
作者
张琛
房胜
王风云
李哲
郑纪业
沈宇
ZHANG Chen;FANG Sheng;WANG Fengyun;LI Zhe;ZHENG Jiye;SHEN Yu(Science and Technology Information Institute,Shandong Academy of Agricultural Sciences,Jinan 250100,China;College of Computer Science and Engineering,Shandong University of Science and Technology,Qingdao 266000,China)
出处
《河南农业科学》
北大核心
2019年第4期154-160,共7页
Journal of Henan Agricultural Sciences
基金
山东省重点研发计划项目(2016CYJS03A01-1)
山东省农业科学院农业科技创新工程项目(CXGC2017B04)
关键词
机器视觉
苹果分级
图像处理
果梗
花萼
支持向量机
Machine vision
Apple grading
Image processing
Fruit stalk
Calyx
Support vector machine
