高光谱成像技术在核桃壳仁检测中的应用  

Application of hyperspectral imaging technology in identification of walnut shell and kernels

作　　者：陈彩虹 张淑娟[1] 孙海霞[1] 李成吉 邢书海 Chen Caihong;Zhang Shujuan;Sun Haixia;Li Chengji;Xing Shuhai(College of Engineering,Shanxi Agricultural University,Taigu 030801,China)

机构地区：[1]山西农业大学工学院,山西太谷030801

出　　处：《山西农业大学学报（自然科学版）》2018年第11期27-32,共6页Journal of Shanxi Agricultural University(Natural Science Edition)

基　　金：国家自然科学基金(31271973);晋中市科技重点研发计划(Y172007-4)

摘　　要：[目的]为了实现对核桃壳、仁及分心木快速、准确识别。[方法]以礼品2号核桃的核桃壳、仁及分心木为研究对象,采用高光谱成像系统采集样本的光谱信息。对所提取的光谱信息分别用一阶微分处理(1stDer),基线校正(Baseline)、标准归一化(Standard Normalized Variate,SNV)及多元散射校正(Multiplicative Scatter Correction,MSC)进行预处理并建立偏最小二乘(Partial Least Squares,PLS)模型进行判别。用竞争自适应重加权算法(Competitive Adaptive Reweighted Sampling,CARS)、回归系数法(Regression Coefficient,RC)和连续投影法(Successive Projections Algorithm,SPA)提取特征波长,建立最小二乘支持向量机(Least Squares-Support Vector Machine,LS-SVM)判别模型。[结果]建立的PLS模型表明一阶微分处理为最佳预处理。CARS提取的特征波长具有较好的预测结果。LS-SVM建模效果好,对不同特征波长提取下的核桃壳、仁及分心木的判别准确率分别达到了100%、100%、99%。[结论]用高光谱成像技术对核桃壳、仁及分心木进行分选判别是可行的,为核桃深加工和壳、仁在线分选及相关设备的开发提供理论依据。[Objective]The study was proposed to rapidly and accurately identify walnut shells,kernels,and distractions.[Methods]The shells,kernels,and distraction wood of walnut variety Gift#2 were selected as the research material,and the spectral information of the sample was collected by hyperspectral imaging system.The extracted spectral information was preprocessed with first-order differential processing(1st Der),baseline correction(Baseline),Standard Normalized Variate(SNV)and Multiplicative Scatter Correction(MSC)to establish a Partial Least Squares(PLS)model for object discrimination.The feature wavelengths were extracted using Competitive Adaptive Reweighted Sampling(CARS),Regression Coefficient(RC),and Successive Projections Algorithm(SPA),and Least Squares-Support Vector Machine(LS-SVM)discriminant model was established.[Results]The analysis results from established PLS model indicated that the 1st Derwas the best pre-processing.The characteristic wavelengths extracted by CARS produced better prediction results,and the modeling effect of LS-SVM was acceptable.The discriminative accuracy of walnut shell,kernel and distraction wood under different characteristic wavelengths were 100%,100%and 99%,respectively.[Conclusion]It is feasible to sort and distinguish walnut shell,kernel and distraction wood using hyperspectral imaging technique,and present study provided theoretical basis for further walnut products processing and on-line sorting of shells and kernels,as well as related equipment development.

关 键 词：高光谱成像技术 核桃壳 核桃仁 检测 

分 类 号：S123[农业科学—农业基础科学] S664