高光谱技术结合网格搜索优化支持向量机的桃缺陷检测  

Hyperspectral technology combined with grid search optimized support vector machines to detect defects of peach

作　　者：张立秀 张淑娟[1] 孙海霞[1] 薛建新[1] 任锐 刘文俊 ZHANG Lixiu;ZHANG Shujuan;SUN Haixia;XUE Jianxin;REN Rui;LIU Wenjun(College of Engineering,Shanxi Agricultural University,Jinzhong 030801,China)

机构地区：[1]山西农业大学农业工程学院,山西晋中030801

出　　处：《食品与发酵工业》2023年第16期269-275,共7页Food and Fermentation Industries

基　　金：国家自然科学基金青年基金项目(31801632);山西省重点研发计划项目(201903D221027)。

摘　　要：为快速区分完好桃、疮痂桃、腐烂桃(虫咬桃、鸟啄桃),实现久保桃外部缺陷的无损检测,该研究利用高光谱技术对久保桃的外部缺陷进行了研究。共采集302个久保桃样本(120个完好桃样本、120个缺陷桃样本、62个验证桃样本),对比经光谱学、基线校正、中值滤波(median filter,MF)等5种预处理方法建立偏最小二乘法模型的准确率,选取经MF预处理后的光谱数据进行后续建模研究。采用回归系数法、竞争性自适应重加权算法(competitive adaptive reweighted sampling,CARS)提取特征波长,建立网格搜索法优化支持向量机(grid search optimized support vector machines,GS-SVM)模型、遗传算法优化SVM模型、粒子群算法优化的SVM模型并进行对比分析。结果表明,CARS-GS-SVM模型预测效果最好,其训练集的判别率为93.33%,预测集的判别率为96.77%,验证集的判别准确率为91.94%,运行时间为11.5 s。该研究利用高光谱技术结合CARS-GS-SVM模型实现了久保桃外部缺陷的检测,为开发水果的分级分选设备提供了理论基础。To quickly distinguish intact peaches,scab peaches,and rotten peaches(insect-eaten peaches and bird-pecked peaches) and realize nondestructive testing of external defects of Kubo peaches,hyperspectral technology was used to study the external defects of Kubo peaches.A total of 302 Kubo peach samples(120 intact peach samples,120 defective peach samples,and 62 verified peach samples) were collected,and the five pretreatment methods such as spectroscopy,baseline correction,and median filtering(MF) were used for spectral pretreatment and partial least squares method(PLS) model was carried out to obtain the better performance.Finally,MF was selected as the pretreatment method,and based on the pre-treated spectral data.The regression coefficient method and competitive adaptive reweighting algorithm(CARS) were used to extract the important wavelength.The grid search method optimized SVM(GS-SVM) model,genetic algorithm optimized SVM model,and particle swarm optimization algorithm optimized SVM model were established and compared.Results showed that the CARS-GS-SVM model had the best prediction effect on the three types,the discrimination rate of the training set was 93.33%,the discrimination rate of the prediction set was 96.77%,the discrimination accuracy of the validation set was 91.94%,and the running time was 11.55 s.In this study,hyperspectral technology combined with the CARS-GS-SVM model was used to detect the external defects of Kubo peach,which provided a theoretical basis for the development of fruit grading and sorting equipment.

关 键 词：高光谱 久保桃 外部缺陷 网格搜索法优化支持向量机 检测 

分 类 号：S662.1[农业科学—果树学] TP18[农业科学—园艺学]