基于图像局部方差的亮度矫正下番茄表面缺陷检测方法  

Tomato surface defect detection method based on image local variance considering brightness correction

作　　者：何婷婷 姚继超 张钟莉莉 芦天罡 岳焕芳 HE Tingting;YAO Jichao;ZHANG Zhonglili;LU Tiangang;YUE Huanfang(Shanxi Agricultural University,Taigu,Shanxi 030801,China;Intelligent Equipment Technology Research Center,Beijing Academy of Agriculture and Forestry Sciences,Beijing 100097,China;Liaocheng Water Conservancy Development and Guarantee Center,Liaocheng,Shandong 252000,China;Beijing Digital Agriculture Promotion Center,Beijing 100101,China;Beijing Agriculture Technology Extension Station,Beijing 100029,China)

机构地区：[1]山西农业大学,山西太谷030801 [2]北京市农林科学院智能装备技术研究中心,北京100097 [3]山东省聊城市水利事业发展和保障中心,山东聊城252000 [4]北京市数字农业农村促进中心,北京100101 [5]北京市农业技术推广站,北京100029

出　　处：《食品与机械》2023年第9期128-133,161,共7页Food and Machinery

基　　金：现代农业产业技术体系(编号:BAIC10-2022-E02);北京市乡村振兴科技项目(编号:20220818)。

摘　　要：目的:解决番茄表面缺陷颜色复杂多变、纹理变化不规则导致的缺陷提取不准确的问题。方法:提出一种考虑亮度矫正下基于图像局部方差的番茄表面缺陷分割方法。在采用直方图阈值分割方法分割花萼与茎疤以及领域像素加权和替代原像素的方法完成亮度矫正的基础上,将番茄表面灰度图像划分成若干图像块,使用图像像素方差对各图像块的颜色进行表征,将缺陷区域与健康区域区分开。使用SVM模型对番茄表面缺陷区域面积占原图中番茄面积的比例进行检测。结果:考虑亮度校正后对番茄缺陷区域提取准确率提高了27.74个百分点,在此基础上,与全局阈值、动态阈值、区域生长算法相比,基于图像局部方差的缺陷提取方法能够实现番茄表面缺陷的准确定位与完整提取,以缺陷面积比为输入的高斯-SVM模型对番茄表面缺陷检测的精度达96%。结论:考虑亮度矫正下,基于图像局部方差的SVM缺陷提取方法适用于番茄表面缺陷检测。Objective:In order to solve the problem of inaccurate defect extraction caused by complex and variable color and irregular texture changes of tomato surface defects,the defect segmentation method based on image local variance with brightness correction was proposed.Methods:On the basis of using histogram threshold segmentation method to segment calyx and stem scar and the method of domain pixel weighted sum to replace the original pixel to complete the brightness correction,the gray image of tomato surface was divided into several image blocks,and the color of each block was characterized by image pixel variance,and then the defect and healthy area were separated.SVM model was used to detect the proportion of tomato surface defect area in the original tomato area.Results:Considering the brightness correction,the accuracy of tomato defect area extraction could be improved by 27.74%.On this basis,compared with the global threshold,dynamic threshold and regional growth algorithm,the defect extraction method based on image local variance could accurately achieve the quasi-deterministic and complete extraction of tomato surface defects,and the accuracy of the Gauss-SVM model with the defect area ratio as the input for tomato surface defect detection reached 96%.Conclusion:Considering brightness correction,the SVM defect extraction method based on image local variance is suitable for tomato surface defect detection.

关 键 词：机器视觉 缺陷检测 花萼 茎疤 亮度矫正 图像局部方差 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] S641.2[自动化与计算机技术—计算机科学与技术]