摘要
利用高光谱成像技术(HSI)采集了300个不同贮藏时间的番茄高光谱图像,在所界定的有效波段中提取特征波长的基础上,构建了番茄贮藏过程中品质变化的马氏距离监测模型,实现了番茄贮藏过程中的品质变化监测。首先,采用多元散射校正(MSC)结合Savizky-Golay卷积平滑(SG)方法对高光谱原始数据进行了预处理,以消除基线漂移及噪声信号等影响。其次,基于光谱曲线在不同波段范围的变化趋势,并结合全波段下Wilks Λ最小值对应主成分在各个波长下的权重系数定义了可凸显番茄贮藏过程中品质变化的有效波段。随后,分别在全波段和有效波段采用连续投影算法(SPA)、竞争性自适应加权算法(CARS)和Wilks Λ统计量融合主成分分析3种筛选特征波长的方法进行特征波长的提取,通过对比分析3种方法提取的特征波长数量,指出了基于Wilks Λ统计量融合主成分分析可以有效降低数据维度,简化运算过程。接着对基于Wilks Λ统计量融合主成分分析在全波段和有效波段筛选出的主成分进行分析,指出了在有效波段基于Wilks Λ统计量融合主成分分析进行特征波长的提取可以很好地避免冗余信息对有效信息的掩盖作用,还可以进一步降低数据维度。最后,对分别以贮藏第1日和番茄腐败临界日为监测基准日构建番茄贮藏过程中品质变化的马氏距离监测模型的优缺点进行了分析,指出以贮藏第1日为监测基准日构建的模型更具有有效性和可靠性。研究结果表明:基于Wilks Λ统计量融合主成分分析提取的特征波长个数为最少(5个),且筛选出的主成分能够有效地表征番茄贮藏过程中品质变化的差异性。为利用高光谱成像技术监测番茄贮藏过程中的品质变化提供了一种有效的特征波长提取方法。
In order to realize the rapid monitoring of quality change during tomato storage,300 tomato hyperspectral images with different storage data were collected by hyperspectral imaging technology(HSI),and a Mahalanobis distance monitoring model for quality change during tomato storage was constructed based on extracting feature wavelengths from the defined effective wavelength bands.Then the quality change monitoring during tomato storage was realized.Firstly,the hyperspectral raw data is preprocessed using multiplicative scatter correction(MSC)combined with Savizky-Golay convolutional smoothing(SG)to eliminate the effects of baseline drift and noisy signals and so on.Secondly,based on the variation trend of the spectral curve in different bands and combined with the weight coefficient of principal components corresponding to the minimum value of Wilks Λ statistic in the whole band,the effective band that can highlight the quality change in the tomato storage process is defined.Thirdly,three feature wavelength screening methods,namely,successive projections algorithm(SPA),competitive adaptive reweighted sampling(CARS)and principal component analysis based on Wilks Λ statistics,are used to extract feature wavelengths in the full band and effective band,respectively;By comparing and analyzing the number of feature wavelengths extracted by the three methods,it is pointed out that the fusion principal component analysis based on Wilks Λ statistics can effectively reduce the data dimension and simplify the operation process.Then,the principal components screened in the full band,and effective band based on Wilks Λ statistics coupled with principal component analysis are analyzed.It is pointed out that the extraction of feature wavelength based on Wilks Λ statistics fusion principal component analysis in the effective band can avoid the masking effect of redundant information on effective information effectively and further reduce the data dimension.Finally,the advantages and disadvantages of the Mahalanobis distance monitoring model for quality change during tomato storage constructed based on the first storage day and the critical day of tomato spoilage are analyzed,and it is pointed out that the model constructed by the first storage day as the monitoring benchmark has higher effectiveness and reliability.The results show that the number of feature wavelengths extracted based on Wilks Λ statistic combined with principal component analysis under the effective band is the least(5 feature wavelengths),and the selected principal components can effectively represent the difference between the quality change during tomato storage.At the same time,it also provides an effective feature wavelength extraction method for monitoring the quality changes during tomato storage by hyperspectral imaging technology.
作者
贾梦梦
殷勇
于慧春
袁云霞
王志豪
JIA Meng-meng;YIN Yong;YU Hui-chun;YUAN Yun-xia;WANG Zhi-hao(College of Food and Bioengineering,Henan University of Science and Technology,Luoyang 471023,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2023年第3期969-975,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
高光谱成像技术
特征波长提取
WilksΛ统计量
贮藏
番茄
马氏距离
Hyperspectral imaging technology
Feature wavelength extraction
WilksΛstatistic
Storage
Tomato
Mahalanobis distance
