基于高光谱成像技术的柠檬片含水量检测研究  

Detection of Water Content of Lemon Slices Based on Hyperspectral Imaging

作　　者：周鹏程 罗强 张艮 汲新鹏 朱洪达 辛琪 崔海波 蒋文豪 ZHOU Pengcheng;LUO Qiang;ZHANG Gen;JI Xinpeng;ZHU Hongda;XIN Qi;CUI Haibo;JIANG Wenhao(College of Mechanical Engineering,Chongqing Three Gorges University,Chongqing 404100,China)

机构地区：[1]重庆三峡学院机械工程学院,重庆404100

出　　处：《食品科技》2025年第2期291-299,共9页Food Science and Technology

基　　金：重庆市科委自然科学基金面上项目(cstc2020jcyj-msxmX0143)。

摘　　要：含水量是影响柠檬片干制品品质的重要指标,传统的含水量检测方法存在耗时长、无法批量检测,对物品会造成损伤等问题。为快速检测柠檬片含水量,实现批量快速检测,文章基于光谱成像技术研究一种快速、准确的含水量检测方法。将240份柠檬片置于80℃热风干燥箱中测定其含水率,同时采集干燥后柠檬片光谱数据,利用算法筛选出柠檬片水分的特征波段,分别利用BP(Back propagation)神经网络算法、多元回归分析(Multiple regression,MLR)算法和偏最小二乘法(Partial least squares,PLS)算法,建立柠檬水分预测模型,以相关系数R2和均方根误差(Root mean squared error,RMSE)对比分析试验中样本预测集与真实值的数据,得到结论。结果表明:3种模型的剩余预测残差(Residual prediction residuals,RPD)均大于6,证明建模效果很好。具体而言,PLS模型的相关系数(R^(2)=0.9953)与RMSE(0.0245)具有极强的相关性,预测精度高。MLR模型的相关系数R2为0.9922,RMSE为0.0084,在3种模型中最低;而BP神经网络模型的相关系数R2为0.9877,RMSE为0.0404。因此,相较于BP神经网络模型,PLS模型和MLR模型更适用于柠檬片的水分预测。Water content is an important index affecting the quality of dried lemon slices.The traditional water content detection is time-consuming,unable to batch detection,and may cause damage to items,etc.In order to quickly detect the water content of lemon slices and achieve rapid batch testing,this paper researches a fast and accurate water content detection method based on spectral imaging technology.In this article,240 lemon slices were placed in a hot air drying oven at 80℃to determine their moisture content and the spectral data of the dried lemon slices were collected at the same time.The characteristic bands of the lemon slices were screen out using an algorithm,and the lemon moisture predicition model was established using the back propagation(BP)neural network algorithm the multiple regression(MLR)algorithm and the partial least squares(PLS)algorithm respectively,to analyze the water content of the samples in the experimental comparison and analysis.The correlation coefficient R~2 and the root mean squared error(RMSE)were used to compare and analyze the data of the sample prediction set and the real value in the experiment,and leading to conclusions.The results show that the residual prediction residuals(RPD)of the three models are all greater than 6,indicating good accuracy and stability.Specifically,the correlation coefficient(R^(2)=0.9953)and RMSE(0.0245)of the PLS model have extremely strong correlation and high prediction accuracy.The MLR model has a correlation coefficient of 0.9922and a lowest RMSE(0.0084),while the BP neural network model has a correlation coefficient(R~2=0.9877)and a RMSE(0.0404).Therefore,compared to the BP neural network model,the PLS model and the MLR model are more suitable for predicting the moisture content of lemon slices.

关 键 词：柠檬片 高光谱成像技术 BP神经网络算法 多元回归分析 偏最小二乘法 

分 类 号：TS207.3[轻工技术与工程—食品科学]