摘要
为利用3D荧光技术实现基于贮藏室气体荧光信息的香蕉腐败早期预警,选取两批不同贮藏日期的香蕉贮藏室气体进行三维荧光数据采集。首先对荧光数据进行了预处理:为消除三维荧光仪整体漂移现象,对得到的三维荧光数据进行去除漂移处理;利用matlab中eemscat工具包对瑞利与拉曼散射进行去除处理,有效消除了瑞利散射和拉曼散射的不利影响;运用Savitzky-Golar(SG)进行数据平滑处理,减少了噪声对荧光信号的干扰。同时,对三维荧光数据进行初步筛选,去除了荧光强度接近于0的发射波长,以及利用三阶高斯混合分布对不同激发波长下的发射光谱进行拟合去除了离散性较大的激发波长。然后针对荧光数据的特征表征,提出了一种基于WilksΛ统计量融合间隔偏最小二乘法(iPLS)的荧光数据特征选择策略。具体是:用WilksΛ统计量进行特征激发波长的选取,初步选出了5个特征激发波长;根据初选的特征激发波长用iPLS结合pH值及相对电导率进行了特征发射波段的选取,结果每个特征激发波长下均选出包含14个波长的特征发射波段。为进一步减少分析变量个数,根据选出的特征发射波长,运用WilksΛ统计量再次进行特征激发波长反选,最终得到了3个特征激发波长。考虑到各特征激发波长下对应14个特征发射波长,故可选出42个特征发射波长。最后基于香蕉在贮藏中其品质变化具有时变特点,根据42个特征发射波长使用系统聚类分析法(SCA)进行香蕉腐败基准界定,得到两批香蕉均在贮藏的第5天出现品质突变。因此选用第5天贮藏室气体荧光信息来表征香蕉的腐败情况。另外,利用主成分分析(PCA)初步探索了用第1主成分实现香蕉早期腐败的预警。结果表明:文中提出的三维荧光数据特征波长的选择策略是能够有效降低光谱数据的复杂度,同时给出的早期腐败预警方法是可行的。
In order to use three-dimensional(3D)fluorescence technology to realize early warning of banana spoilage based on storage room gas,the storage room gas corresponding to two batches of banana with different storage dates were tested to collect 3D fluorescence data.Firstly,the 3D fluorescence data was pre-processed:to eliminate the overall drift of the scanning data of the 3D fluorescence instrument,the drift of the obtained 3D fluorescence data was processed;the removal and interpolation of Rayleigh and Raman scattering were handled by using the eemscat toolbox in matlab platform,which effectively eliminated the adverse effects of Rayleigh scattering and Raman scattering;and the Savitzky-Golar(SG)method was employed for data smoothing to reduce the influence of noise on the fluorescence signal.Meanwhile,the 3D fluorescence data were preliminarily screened,the emission wavelengths with fluorescence intensity close to 0 were removed,and the more discrete excitation wavelengths were removed by using a third-order Gaussian mixture distribution to fit the emission spectra at different excitation wavelengths.Then,aiming at the feature selection of 3D fluorescence data,a feature wavelength selection strategy based on WilksΛstatistic combined with interval partial least squares(iPLS)was proposed.The specific steps are:step 1,using WilksΛstatistics to select feature excitation wavelengths,and five feature excitation wavelengths were preliminarily selected;step 2,based on the initially selected feature excitation wavelengths,the iPLS method was used to select the feature emission bands in combination with pH and relative conductivity,and feature emission band including 14 wavelengths was selected at each feature excitation wavelength;step 3,in order to further reduce the number of analysis variables,according to the selected feature emission band,WilksΛstatistics was used againto select the feature excitation wavelengths inversely,and 3 feature excitation wavelengths were finally obtained.Combined with 14 emission wavelengths at each feature excitation wavelength,a total of 42 feature emission wavelengths were selected.Finally,considering the time-varying characteristic of banana quality during storage,with the help of the 42 feature emission wavelengths,systematic cluster analysis(SCA)was employed to define the benchmark for banana spoilage,and the cluster results showed that both batches of bananas had abrupt changes in quality on the 5th day of storage.Therefore,the fluorescence information of the storage room gas on the 5th day was used to characterize the banana spoilage.In addition,these feature wavelength variables were computed by principal component analysis(PCA),and the first principal component was preliminary explore to realize early warning of banana spoilage.The research results show that the selection strategy of feature wavelengths of the 3D fluorescence data proposed in this paper can effectively reduce the complexity of the spectral data so as to facilitate subsequent analysis and the early warning method of banana spoilage is also feasible.
作者
李孟丽
殷勇
袁云霞
李欣
刘雪茹
LI Meng-li;YIN Yong;YUAN Yun-xia;LI Xin;LIU Xue-ru(College of Food and Bioengineering,Henan University of Science and Technology,Luoyang471023,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2021年第2期558-564,共7页
Spectroscopy and Spectral Analysis
基金
国家重点研发计划项目(2017YFC1600802)资助。
关键词
香蕉
腐败预警
三维荧光
WilksΛ统计量
间隔偏最小二乘
系统聚类分析
Banana
Spoilage warning
Three-dimensional fluorescence
WilksΛstatistic
Interval partial least squares
Systematic cluster analysis
