近红外光谱结合模糊非相关QR分析的生菜储藏时间辨别  

Discrimination of Lettuce Storage Time Based on Near-Infrared Spectroscopy Combined With Fuzzy Uncorrelated QR Analysis

作　　者：胡彩平 傅兆民 许鸿嘉 武斌[3] 孙俊[4] HU Cai-ping;FU Zhao-min;XU Hong-jia;WU Bin;SUN Jun(School of Computer Engineering,Jinling Institute of Technology,Nanjing 211169,China;Jiangsu University Jingjiang College,Zhenjiang 212028,China;Department of Information Engineering,Chuzhou Polytechnic,Chuzhou 239000,China;School of Electrical and Information Engineering,Jiangsu University,Zhenjiang 212013,China)

机构地区：[1]金陵科技学院计算机工程学院,江苏南京211169 [2]江苏大学京江学院,江苏镇江212028 [3]滁州职业技术学院信息工程系,安徽滁州239000 [4]江苏大学电气信息工程学院,江苏镇江212013

出　　处：《光谱学与光谱分析》2024年第8期2268-2272,共5页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(31471413);金陵科技学院高层次人才科研启动项目(JIT-RCYJ-202102);江苏省重点研发计划项目(BE2022077);江苏大学大学生创新训练计划项目(202213986008Y)资助。

摘　　要：生菜是人们经常食用的蔬菜之一,生菜的储藏时间是影响生菜新鲜程度的重要因素。所以研究一种简单、快速、非破坏性的生菜储藏时间的鉴别方法是非常必要的。近红外光谱(NIR)分析能快速和准确的获取生菜的近红外光谱,从而实现无损鉴别生菜储藏时间。但是生菜的NIR数据中存在噪声信号和冗余信号,为了消除光谱的噪声信号并提取特征信息,提出了一种基于模糊非相关QR分析(FUQRA)的近红外光谱生菜储藏时间鉴别新方法。首先,需要降低原始NIR数据的维数,通过使用主成分分析(PCA)将包含1557个维度的光谱数据降至包含22个维度。然后通过模糊非相关判别转换(FUDT)计算出特征向量,利用特征向量建立鉴别向量矩阵,并进行QR分解,得到最终的鉴别向量矩阵。最后以60个新鲜生菜样本为研究样本,使用K近邻(KNN)方法进行分类,用AntarisⅡ型NIR光谱仪对生菜样品进行近红外光谱检测和数据收集。实验过程中每隔12小时对每个样本进行3次重复检测,将这些数据取平均值作为实验数据。随后利用多元散射校正(MSC)减少近红外光谱中的噪声信号。为了验证所提出方法的有效性,分别将主成分分析(PCA)结合KNN、主成分分析和模糊线性判别分析(FLDA)结合KNN、主成分分析和模糊非相关判别转换(FUDT)结合KNN以及主成分分析和模糊非相关QR分析(FUQRA)结合KNN四种算法分析结果进行比较。将权重指数m的不同取值产生的分类准确率进行比较,选出最合适的权重指数和KNN的参数K:m=2,K=3。最终得到的分类准确率分别为43.33%、96.67%、96.67%和98.33%。可以看出,相比其他三个算法,模糊非相关QR分析可以更好地实现对生菜储藏时间的鉴别。Lettuce is one of the vegetables that people often eat,and the storage time of lettuce is an important factor affecting the freshness of lettuce.Therefore,it is necessary to develop a simple,fast,and non-destructive method to identify the storage time of lettuce.Near-infrared spectroscopy(NIR)can quickly and accurately detect the near-infrared spectrum of lettuce to realize the non-destructive identification of lettuce storage time.However,noise and redundant signals are in the NIR spectral data collected by the near-infrared spectrometer.To eliminate the noise information of the spectrum and extract the feature information,a novel method was proposed to identify the storage time of lettuce based on NIR spectroscopy and fuzzy uncorrelated QR analysis(FUQRA).Firstly,principal component analysis(PCA)was used to reduce the dimension of the original spectral data from 1557 dimensions to 22 dimensions.Secondly,after the feature vectors are obtained by fuzzy uncorrelated discriminant transformation(FUDT),the discriminant vector matrix is established by using the feature vectors,and the final discriminant vector matrix is obtained by QR decomposition.Finally,the k-nearest neighbor algorithm was utilized for classification.60 fresh lettuce samples were selected as the research object.Firstly,the NIR spectral data of lettuce samples were collected by AntarisⅡnear-infrared spectrometer and detected once every 12 hours.Secondly,multivariate scatter correction(MSC)was used to reduce the noise signal in the NIR spectra.To verify the effectiveness of the proposed method,the experimental results were compared by four classification models:principal component analysis(PCA)combined with a K-nearest neighbor(KNN)algorithm,PCA and fuzzy linear discriminant analysis(FLDA)combined with KNN algorithm,PCA and fuzzy uncorrelated discriminant transformation(FUDT)combined with KNN algorithm and PCA and FUQRA combined with KNN algorithm.The classification accuracies produced by different values of the weight indexm were studied,and the m

关 键 词：近红外光谱 模糊线性判别分析 模糊非相关判别转换 模糊非相关QR分析 生菜 

分 类 号：O657.33[理学—分析化学]