基于近红外高光谱技术的杧果可溶性固形物含量无损检测

林娇娇1,2,蒙庆华1,2*,吴哲锋1,2,常洪娟1,2,倪淳宇1,2,邱邹全1,2,李华荣1,黄玉清3

1南宁师范大学物理与电子学院,南宁 530001;2南宁师范大学广西信息功能材料与智能信息处理重点实验室,南宁 530001;3北部湾环境演变与资源利用教育部重点实验室·广西地表过程与智能模拟重点实验室·南宁师范大学,南宁 530001)

摘 要:目的】近红外高光谱成像技术(NⅠR-HSⅠ)在水果内部品质的无损检测方面具有快速、准确和无损的特点。旨在利用NⅠR-HSⅠ技术分析不同品种杧果的可溶性固形物含量,并探讨400~1000 nm波段范围内的光谱差异和可溶性固形物含量的响应。【方法】选择贵妃杧果和台农1 号杧果作为研究对象,使用NⅠR-HSⅠ技术获取杧果样本的光谱数据。采用CARS-PLS模型分析可溶性固形物含量与各波段光谱反射率的相关系数。为了验证模型的性能,计算了建模R2、斜率Slope、截距和RMSE等指标。【结果】得到CARS-PLS模型的性能指标:建模R2为0.880 6,斜率为0.851 5,截距为12.208,RMSE为0.636 6。这些指标表明该模型具有较高的建模拟合度和预测精度。【结论】应用NⅠR-HSⅠ技术对杧果可溶性固形物含量进行检测具有可行性。为进一步研究不同水果可溶性固形物含量的高精度模型奠定了基础。通过NⅠR-HSⅠ技术的应用,可以提供一种非破坏性且高效准确的方法,用于水果品质评估和检测。这对农产品质量控制和市场营销具有重要的意义。

关键词:杧果;近红外(NⅠR);高光谱成像(HSⅠ);可溶性固形物含量;无损检测;光谱差异

中国广西百色盛产杧果,杧果品种丰富、肉质嫩滑、营养价值高,深受人们喜爱。可溶性固形物含量是评价杧果内部品质的重要指标[1-2],可确定杧果的收获时间。传统的水果内部品质检测方法为化学分析方法[3],它是将待测水果用组织粉碎机粉碎,榨汁过滤后测定。这种破坏水果样品外观的化学分析检测方法,过程费时费力且效率低下,无法实现实时在线检测。随着成像和光谱技术的快速发展,近红外高光谱成像技术(NⅠR-HSⅠ)已经广泛应用于农产品品质的快速无损检测[4]。NⅠR-HSⅠ正越来越多地与果实分选系统、成熟度监测和贮藏果实成熟度水平的决策相结合[5]。随着人们生活水平的日益提高,对杧果分级和精深加工的要求也越来越高,因此研究一种简单、快速、非破坏性的杧果可溶性固形物含量检测技术很有必要[6]

国内外进行了水果无损检测研究,其中研究者通过高光谱成像技术开展了多项探索。高升等[7]研究了红提的可溶性固形物含量和硬度的无损检测方法,发现基于随机森林(RF)建立的模型在预测可溶性固形物含量和硬度方面效果较好。特别是针对可溶性固形物含量,他们采用遗传算法(GA)优化的随机森林模型,取得了高度准确的预测结果。Yuan等[8]针对桃果实的可溶性固形物含量提出了一种融合共识模型的策略,旨在克服遗传算法在模型优化中的不确定性,最大限度地利用光谱信息来实现快速检测。此外,Seki等[9]开发了一种可视化白草莓果肉中糖含量的方法,为未来设计非接触式质量监测系统提供了重要见解。Gao等[10]对海棠果的可溶性固形物含量和硬度指数进行了研究,利用近红外高光谱成像结合化学计量学建立了多种模型,以提高无损检测效率,结果显示这种方法用于海棠果的质量评估是可行的。最后,Riccioli 等[11]应用HSⅠ技术对橙的质量属性进行量化,找到了最佳分类策略,以获得质量属性的空间分布信息。这些研究强调了高光谱成像技术在水果无损检测中的潜力,以及通过不同方法和模型的应用,可以实现对水果内部质量的准确和高效检测,这为农产品质量控制和市场监测提供了重要的工具和方法。

然而,大多数研究者利用光谱仪检测果类品质、可溶性固形物含量、酸度等时,研究单一品种或单一类水果的检测模型其稳定性低、普适性弱,不能较好地实现科研成果产业化的发展。在对NⅠR-HSⅠ光谱数据进行建模处理的过程中,需要识别潜在的峰,以避免共线性问题,然后使用从这些相应的峰中提取的信息来校准模型[12],提高模型准确性。因此笔者在本研究中以贵妃杧果和台农1 号杧果为研究对象,利用NⅠR-HSⅠ技术,研究不同品种杧果不同可溶性固形物含量光谱差异及特征响应波段范围,为建立不同水果可溶性固形物含量高精度检测模型奠定基础。

1 材料和方法

1.1 杧果样本

实验所用样本果实采摘于广西壮族自治区百色市田阳区正义杧果园,果实大小均匀,无病虫害、机械伤和破损。为了实现NⅠR-HSⅠ技术对杧果可溶性固形物含量的无损检测,对杧果样本进行标号,台农1号杧果y1~y150,贵妃杧果g1~g134,常温静置24 h后进行高光谱图像的获取,然后进行质量属性检测,主要针对杧果的可溶性固形物含量进行检测。

1.2 高光谱图像获取

高光谱图像数据通过高光谱成像系统采集得到,高光谱成像系统由美国的Headwall Micro-Hyperspec VNⅠR A 高光谱成像仪、一个300 W 的卤素灯和一个可移动云台组成,如图1 所示。试验采集得到光谱范围是400~1000 nm,获取的高光谱图像共327 个波段。图像3-d 数据立方体由高光谱系统测量,包括样本的光谱(xy)和空间(xy)信息。为了减少暗电流噪声的影响,并降低高光谱成像系统的光照在图像中产生一定的噪声,在样本采集前需要对高光谱图像进行黑白校正。使用标准白板扫描得到的白色参考像Iwhite和无光照覆盖镜头得到的黑色参考像Idark进行校正。从理论上讲,校正后的图像R由原始的高光谱图像I根据以下公式进行变换:

图1 光谱采集图
Fig.1 Spectral acquisition map

1.3 提取光谱和图像数据

在杧果顶部、中部、底部标记区域作为杧果可溶性固形物含量测试部位,每个杧果正反各扫描一次。运用ENVⅠ的感兴趣区域提取功能,提取可溶性固形物含量测量区域的平均光谱,将此平均光谱与其对应的可溶性固形物含量建立对应关系。在提取和保存相应的光谱信息和图像信息后,使用MATLAB R2018b软件执行光谱数据建模和图像数据原始分割。由于光谱数据容易受到光线、噪音、基线漂移等因素的干扰,因此需要对原始数据进行预处理[13]。多重散射校正(MSC)算法可以有效消减光谱数据中的随机噪声,消噪效果受平滑点数的影响[14],本文中选择MSC预处理对光谱数据进行处理。

1.4 杧果可溶性固形物含量测量

采集完所有样本的光谱图像信息后,当天进行并完成杧果可溶性固形物含量测定。将杧果顶部、中部、底部标记区域的果皮削掉,取出适量果肉压汁,随后用数字折射计测定可溶性固形物含量,读出该样本的可溶性固形物含量理化值示数。每个样本以3 次平行测定结果的算术平均值作为该杧果样本的可溶性固形物含量参考值。数字折射仪是由日本ATAGO 公司生产的便携式数显折射计PAL-1,也称为数显可溶性固形物计,测量范围是0.0%~53.0%。

1.5 建模与检验数据

原始数据集包含134 个贵妃杧果样本和150 个台农1号杧果样本,通过KS算法分别划分为两个独立数据集:贵妃杧果校正集由101个样本组成,预测集覆盖剩余的33 个样本见表1;台农1 号杧果校正集由113 个样本组成,预测集覆盖剩余的37 个样本见表2。表1、表2分别显示不同品种杧果校正集和预测集的质量属性的最小值、最大值、平均值、标准差(s)以及变异系数(CV)。

表1 贵妃杧果可溶性固形物含量的测量结果统计
Table 1 Statistics of soluble solids content measurements of Guifei mango

样本Sample校正集Correction set预测集Prediction set样本总量Total sample size数量Number 101最小值Minimum value/%12.9最大值Maximum value/%18.7平均值Average value/%14.648标准差Standard deviation 1.538变异系数Coefficient of variation 0.105 33 11.0 19.5 15.282 1.968 0.129 134 11.0 19.5 14.831 1.674 0.113

表2 台农1 号杧果可溶性固形物含量的测量结果统计
Table 2 Statistics of soluble solids content measurements of Tainong No.1 mango

样本Sample校正集Correction set预测集Prediction set样本总量Total sample size数量Number 113最小值Minimum value/%11.8最大值Maximum value/%16.7平均值Average value/%13.903标准差Standard deviation 1.201变异系数Coefficient of variation 0.086 37 11.7 16.8 14.235 1.341 0.094 150 11.7 16.8 13.985 1.245 0.089

共获取了134 个贵妃杧果样本和150 个台农1号杧果样本的可溶性固形物含量值(分别在杧果的顶部、中部、底部3个位置取样),以及其对应区域的平均光谱反射率。采用KS 算法对不同种类的杧果可溶性固形物含量值及其光谱反射率作为建模数据,其余可溶性固形物含量值及对应的光谱反射率作为检验数据。笔者在本研究中采用PLS 模型[15],可以在不丢失主要光谱信息的前提下选择为数较少的新变量来代替原来较多的变量,解决了由于谱带的重叠而无法分析的问题。PLS回归揭示了光谱变量(X)与样本性质(Y[16]之间的线性关系,得到的模型可表示为:

式中,b和e分别为回归系数和预测误差。PLS建模效果由相关系数(R2)和均方根误差(RMSE)评估,R2和RMSE 分别表示实际值和可溶性固形物含量的预测值之间的相关性和偏差。通常,良好的模型具有高R2值和低RMSE值[17]

2 结果与分析

2.1 贵妃杧果不同可溶性固形物含量的光谱分析

贵妃杧果原始光谱数据在400~1000 nm范围的平均光谱反射率曲线如图2 曲线(A)所示。结果表明,所有样品均表现出相似的光谱曲线趋势,不同可溶性固形物含量在540~630 nm 和750~900 nm 区间差异明显,540~630 nm的光谱反射率随着可溶性固形物含量升高呈降低的趋势,750~900 nm的光谱反射率随着可溶性固形物含量升高呈升高的趋势。贵妃杧果在550、857 nm附近出现小峰值,这是由贵妃杧果的有机分子中含O-H 基团振动的合频、各级倍频的吸收作用引起的;在509、680、963 nm左右处出现较宽的吸收带,509 nm处与类胡萝卜素的存在有关;680 nm 左右的低反射率表明该区域的高吸光度,吸收红色的色素,主要原因是叶绿素的存在使果实具有特有的绿色[18];在680 nm 处的峰值之后,反射率急剧上升,这与杧果含有番茄红素有关;963 nm可能与水和碳水化合物的变化或组织结构的变化引起的散射有关[19];在940~947 nm处为可溶性固形物含量的吸收峰,该波段为C-H 基团的三级倍频特征吸收峰[20];970~980 nm 出现的吸收峰主要与杧果的含水量有关,该波段为O-H 基团的二级倍频特征吸收峰[18]

图2 贵妃杧果(A)和台农1 号(B)杧果的近红外光谱图
Fig.2 Near-infrared spectrogram of the Guifei mango(A)and Tainong No.1 mango(B)

2.2 台农1号杧果不同可溶性固形物含量的光谱分析

台农1 号杧果原始光谱数据在400~1000 nm 范围的平均光谱反射率曲线如图2 曲线(B)所示。结果表明,所有样品均表现出相似的光谱曲线趋势,500~640 nm红黄光区域反射率呈直线上升,这与杧果表皮颜色有关。由曲线看出,不同可溶性固形物含量在600~750 nm 区间差异明显,600~700 nm 的光谱反射率随着可溶性固形物含量升高呈降低的趋势。680 nm周围的吸收带同贵妃杧果相似,反映了花青素和叶绿素引起的果实颜色的变化;光谱在720~960 nm之间无明显吸收峰,在857 nm处有较小吸收峰,963 nm 处有明显吸收峰,这是由贵妃杧果的碳水化合物和水中含O-H基团振动的合频、各级倍频的吸收作用引起的[21]

2.3 可溶性固形物含量相近时不同品种杧果的光谱曲线分析

杧果品种不同,但其光谱曲线吸收峰位置大致相近,在509、680、857、963 nm等4个波长附近有4个吸收峰,如图3 所示。两个品种杧果在400~500 nm范围内,光谱曲线无明显变化,但在500~750 nm 范围内,光谱曲线差异显著,是由于台农1号杧果果实呈黄色,叶绿素含量较少,对红色色素的吸收率相对较少,因此在680 nm 左右光谱反射率高于贵妃杧果,反映了花青素和叶绿素引起的果实颜色的变化;在500~640 nm 范围内台农1 号杧果光谱反射率直线上升,形成陡坡,而贵妃杧果光谱反射率趋于平稳;在750~1000 nm范围内,两个品种的杧果光谱曲线变化趋势相似。

图3 可溶性固形物含量相近不同杧果的光谱曲线
Fig.3 Spectral profiles of different mangoes with similar soluble solids content

2.4 水果可溶性固形物含量与光谱反射率的相关性分析及模型构建

杧果光谱反射率对可溶性固形物含量的响应差异较大,如图4所示。对台农1号杧果而言,在400~920 nm范围内,其光谱反射率与可溶性固形物含量呈正相关;700 nm左右光谱反射率急剧下降,在725~920 nm平缓下降至0,其中920 nm处为零界点,即与可溶性固形物含量相关性为0;在920~1000 nm范围内,其光谱反射率与可溶性固形物含量呈负相关。在509、675、963 nm 附近,台农1 号杧果光谱反射率与可溶性固形物含量形成3个峰值,因此,509、675、963 nm可作为台农1号杧果对可溶性固形物含量的3 个特征响应波段。从图4 可知,对于贵妃杧果而言,光谱反射率与可溶性固形物含量在400~736 nm范围内呈正相关,在455 nm 附近出现1 个谷值,该谷值为0.2,即相关系数为0.2;700 nm左右光谱反射率急剧下降,在736 nm 处为零界点,即与可溶性固形物含量相关性为0;在550~675 nm范围内,贵妃杧果的光谱反射率与可溶性固形物含量相关性较为稳定,相关系数保持在0.7 左右;贵妃杧果在736~1000 nm 范围内,其光谱反射率与可溶性固形物含量呈负相关。图中显示台农1号杧果与贵妃杧果的可溶性固形物含量与各波段的光谱反射率相关系数较高的集中在550~700 nm,其中相关系数最高的是670 nm左右处,其相关系数为0.837。并且在700 nm左右处,台农1 号杧果与贵妃杧果反射率都呈现急剧下降至0趋势。

图4 不同杧果的可溶性固形物含量与各波段反射率的相关系数
Fig.4 Correlation coefficients between soluble solids content of different mangoes and the reflectance of each waveband

2.4.1 基于CARS 特征波长选择的PLS 模型 用CARS 进行特征波长选择,选择最小RMSECV 的运行最佳。特征光谱变量的数量随着采样次数的增加先迅速下降然后平缓减少(图5-A)。说明CARS 有“粗选”和“精选”2 个选择方式,极大地提高了变量选择效率。随着采样次数的增加,RMSECV呈先缓慢减小后陡然增大的趋势(图5-B)。这是因为消除了无信息变量,然后有效变量的数量迅速增加。特征光谱变量随着采样次数变化的回归系数路径如图5-C。当图5-B 中RMSECV 值达到最小值时,各特征光谱变量的回归系数位于图5-C 中的“*”所在的垂直线位置。RMSECV=0.636 6 为最低时,提取出16个特征光谱变量,占全波段的4.9%。提取的特征光谱为:431、509、596、698、709、807、809、824、851、857、863、900、952、963、989、991 nm。

图5 杧果可溶性固形物含量的CARS 特征波长选取
Fig.5 CARS characteristic wavelength selection for soluble solids content of mango

A.采样变量数;B.均方根误差;C.回归系数路径。
A.Number of sampling variables;B.RMSECV;C.Regression coefficient path.

2.4.2 基于GA提取特征波长选择的PLS模型 在GA 运算过程中,设定初始群体为30,交叉率为50%,变异率为1%,迭代次数为100。以最小的RMSECV 值为标准,RMSECV 变化图如图6-A 所示。筛选出波长点在迭代过程中出现频次较多的波长点为最优波长点,最终选定特征波长点为38 个,如图6-B所示,占原始光谱的11.6%。提出的特征波长如下:507、509、510、512、514、588、590、592、597、599、601、603、605、607、620、788、807、809、818、857、859、861、863、864、866、868、903、905、907、909、911、913、963、985、987、989、997、992 nm。

图6 杧果可溶性固形物含量的GA 特征波长选取
Fig.6 GA characteristic wavelength selection of mango soluble solids content

A.RMSECV 变化图;B.GA 筛选图。
A.RMSECV change diagram;B.GA screening diagram.

2.4.3 基于UVE 提取特征波长选择的PLS 模型在UVE算法中,以噪声矩阵处最大稳定性绝对值的99%作为剔除阈值。如图7 所示,左侧曲线代表光谱变量的稳定性值,右侧曲线代表噪声变量的稳定性值,两水平虚线为变量的选择阈值(±28.74)。虚线内部为无用信息,外部为有用信息。最终选取了36 个特征波长,占原始光谱的11.0%。提出的特征波长为:416、421、433、460、492、531、533、592、594、597、618、622、625、627、629、635、646、648、649、657、661、664、666、668、670、679、685、794、826、837、857、950、956、963、966、992 nm。

图7 杧果可溶性固形物含量的UVE 特征波长选取
Fig.7 Selection of UVE features of mango soluble solids content

由3种特征波长提取方法提取的特征波长可以看出,其中包括可溶性固形物含量与各波段的光谱反射率相关系数较高的波长(509、550、680、857、963 nm)。由此得出由叶绿素、花青素、碳水化合物和O-H 对应特征波段建模,模型的准确率更高,进而区分不同品种杧果以及对杧果可溶性固形物含量进行预测。其中用准确性和可靠性最高的是CARS-PLS 预测模型,可视化可溶性固形物含量实测值和预测值分布的散点图如图8 所示,实线为实测值与预测值之间理想相关性对应的回归线y=0.851 5x+12.208,决定系数R2为0.880 6,斜率Slope为0.851 5、截距为12.208,RMSECV 为0.636 6。结果表明,由可溶性固形物含量与各波段的光谱反射率相关系数较高的波长构建的PLS模型对杧果可溶性固形物含量的预测具有较好的效果,无论是校正集还是预测集,预测值都最接近相应的实际值。

图8 特征波长光谱反射率与可溶性固形物含量的模型检验
Fig.8 Model test of spectral reflectance of characteristic wavelengths and soluble solids

3 讨 论

在这项研究中,笔者深入研究了不同品种杧果可溶性固形物含量与杧果光谱反射率的对应关系。随着生活水平的提高,消费者对准确分级和加工的杧果的需求持续激增,开发一种简单、快速和非破坏性的技术来评估杧果糖度水平变得势在必行。

笔者的方法涉及高光谱成像的利用,这有助于获取覆盖400~1000 nm 光谱范围内327 个波段的详细光谱数据。杧果样品中可溶性固形物含量的参考值是通过便携式数字折光仪(ATAGO,日本)经过三次单独测量获得的,以平均值作为可溶性固形物含量的参考值。使用ENVⅠ软件打开原始光谱图像后,笔者手动从10×10 像素的正方形区域中选择并提取平均光谱数据。随后采用MATLAB R2018b软件进行光谱数据预处理,具体采用多重散射校正(MSC)算法,有效降低随机噪声。选择K 均值聚类(KS)算法来模拟杧果可溶性固形物含量值及其光谱反射率。剩余的可溶性固形物含量值和相应的光谱反射率作为测试数据。此外,笔者利用偏最小二乘(PLS)建模的力量,使笔者能够提取必要的光谱信息,同时降低原始数据的维度。

经过光谱分析,笔者发现杧果皮细胞表现出更致密的结构,导致对可见光谱的吸收更强。值得注意的是,509 nm和680 nm附近的吸收峰分别与花青素和叶绿素有关,这使得它们对评估果实成熟度很有价值。笔者的发现与Cen等[22]的研究一致,他们也确定了与花青素和叶绿素相关的525 nm 和675 nm左右的吸收水平,为评估水果成熟度提供了可行性。随着杧果成熟,其绿色果皮颜色逐渐褪色,导致525 nm处的吸收系数增大,675 nm处的吸收系数减小。虽然笔者的研究没有深入研究杧果果肉组织的吸收和散射系数,但笔者利用了CARS 特征波长提取方法。这使笔者能够构建基于特征波长带的PLS模型,从而为杧果可溶性固形物含量预测带来有希望的结果。此外,笔者的结果与Wilson 等[23]的结果相似,他们证明近红外区域比可见光区域包含更多与CH和OH基团相关的共振信息。在笔者的研究中,在963 nm处观察到明显的吸收峰,这是由于碳水化合物和水中OH基团振动的组合频率吸收。

通过分析不同杧果品种的光谱曲线,笔者发现这些曲线表现出相似的总体趋势。值得注意的是,509、680、857 和963 nm 附近的吸收峰始终存在,其中680~750 nm范围显示反射率呈上升趋势,形成明显的斜率。因此,680~750 nm 波段范围以及509、550、680、857和963 nm五个波段被确定为杧果果肉的特征波段。特别重要的是,与贵妃杧果相比,台农1号杧果在500~750 nm范围内表现出明显更高的光谱反射率。此外,尽管波长范围不同,但这两个杧果品种都显示出陡峭的斜坡结构。在600~700 nm 红光范围内所有品种的光谱反射率与可溶性固形物含量呈显著相关。

然而,精确确定杧果中的化学信息相对应的特征波长仍然是一个挑战,可能会影响模型的准确性。未来的研究工作应该致力于解决这个问题,从而进一步提高杧果可溶性固形物含量预测的精度。总之,笔者的研究强调了杧果果实可溶性固形物含量及其光谱特征。通过高光谱成像获得的研究结果有望更准确地评估杧果质量和成熟度,并在杧果分级和加工方面具有潜在应用价值。

4 结 论

综合不同品种的杧果光谱反射率,分析其可溶性固形物含量的响应波段范围。研究表明,波段响应最高的是670 nm左右处,其相关系数为0.837,利用可溶性固形物含量与各波段光谱反射率的相关系数高低对CARS-PLS 建模进行检验,其建模R2 为0.880 6、斜率Slope为0.851 5、截距为12.208、RMSE为0.636 6,取得较好的效果。研究结果表明,应用高光谱成像技术检测杧果可溶性固形物含量具有可行性。

参考文献References:

[1] GAO Q,WANG P,NⅠU T,HE D J,WANG M L,YANG H J,ZHAO X Q.Soluble solid content and firmness index assessment and maturity discrimination of Malus micromalus Makino based on near-infrared hyperspectral imaging[J].Food Chemistry,2022,370:131013.

[2] 曹霞,周学成,范品良.基于近红外漫反射光谱技术的杧果糖度无损检测方法研究[J].农机化研究,2013,35(1):177-180.CAO Xia,ZHOU Xuecheng,FAN Pinliang.Nondestructive detection on predicting sugar content of mango by near-infrared diffuse reflectance[J].Journal of Agricultural Mechanization Research,2013,35(1):177-180.

[3] WEⅠNGERL V,UNUK T.Chemical and fruit skin colour markers for simple quality control of tomato fruits[J].Croatian Journal of Food Science and Technology,2015,7(2):76-85.

[4] 张保华,李江波,樊书祥,黄文倩,张驰,王庆艳,肖广东.高光谱成像技术在果蔬品质与安全无损检测中的原理及应用[J].光谱学与光谱分析,2014,34(10):2743-2751.ZHANG Baohua,LⅠJiangbo,FAN Shuxiang,HUANG Wenqian,ZHANG Chi,WANG Qingyan,XⅠAO Guangdong.Principles and applications of hyperspectral imaging technique in quality and safety inspection of fruits and vegetables[J].Spectroscopy and Spectral Analysis,2014,34(10):2743-2751.

[5] WALSH K B,MCGLONE VA,HAN D H.The uses of near infrared spectroscopy in postharvest decision support:A review[J].Postharvest Biology and Technology,2020,163:111139.

[6] 何馥娴,蒙庆华,唐柳,黄新,卢旭恒,王瑞扬,张克智,李钰.高光谱成像技术在水果品质检测中的研究进展[J].果树学报,2021,38(9):1590-1599.HE Fuxian,MENG Qinghua,TANG Liu,HUANG Xin,LU Xuheng,WANG Ruiyang,ZHANG Kezhi,LⅠYu.Research progress in hyperspectral imaging technology for fruit quality detection[J].Journal of Fruit Science,2021,38(9):1590-1599.

[7] 高升,王巧华,付丹丹,李庆旭.红提糖度和硬度的高光谱成像无损检测[J].光学学报,2019,39(10):355-364.GAO Sheng,WANG Qiaohua,FU Dandan,LⅠQingxu.Nondestructive detection of sugar content and firmness of red globe grape by hyperspectral imaging[J].Acta Optica Sinica,2019,39(10):355-364.

[8] YUAN L M,YOU L F,YANG X F,CHEN X J,HUANG G Z,CHEN X,SHⅠW,SUN Y Y.Consensual regression of soluble solids content in peach by near infrared spectrocopy[J].Foods,2022,11(8):1095.

[9] SEKⅠH,MA T,MURAKAMⅠH,TSUCHⅠKAWA S,ⅠNAGAKⅠT.Visualization of sugar content distribution of white strawberry by near-infrared hyperspectral imaging[J].Foods,2023,12(5):931.

[10] GAO Q,WANG P,NⅠU T,HE D J,WANG M L,YANG H J,ZHAO X Q.Soluble solid content and firmness index assessment and maturity discrimination of Malus micromalus Makino based on near-infrared hyperspectral imaging[J].Food Chemistry,2022,370:131013.

[11] RⅠCCⅠOLⅠC,PÉREZ-MARÍN D,GARRⅠDO-VARO A.Optimizing spatial data reduction in hyperspectral imaging for the prediction of quality parameters in intact oranges[J].Postharvest Biology and Technology,2021,176:111504.

[12] SAEYS W,DO TRONG N N,VAN BEERS R,NⅠCOLAÏ B M.Multivariate calibration of spectroscopic sensors for postharvest quality evaluation:A review[J].Postharvest Biology and Technology,2019,158:110981.

[13] MⅠSHRA P,KARAMⅠA,NORDON A,RUTLEDGE D N,ROGER J M.Automatic de-noising of close-range hyperspectral images with a wavelength-specific shearlet-based image noise reduction method[J].Sensors and Actuators B:Chemical,2019,281:1034-1044.

[14] HELLAND ⅠS,NAES T,ⅠSAKSSON T.Related versions of the multiplicative scatter correction method for preprocessing spectroscopic data[J].Chemometrics and Ⅰntelligent Laboratory Systems,1995,29(2):233-241.

[15] GELADⅠP,KOWALSKⅠB R.Partial least-squares regression:A tutorial[J].Analytica Chimica Acta,1986,185:1-17.

[16] LⅠU D,SUN D W,ZENG X A.Recent advances in wavelength selection techniques for hyperspectral image processing in the food industry[J].Food and Bioprocess Technology,2014,7(2):307-323.

[17] RYCKEWAERT M,HÉRAN D,FEⅠLHES C,PREZMAN F,SERRANO E,COURAND A,MAS-GARCⅠA S,METZ M,BENDOULA R.Dataset containing spectral data from hyperspectral imaging and sugar content measurements of grapes berries in various maturity stage[J].Data in Brief,2023,46:108822.

[18] PU H B,LⅠU D,WANG L,SUN D W.Soluble solids content and pH prediction and maturity discrimination of lychee fruits using visible and near infrared hyperspectral imaging[J].Food Analytical Methods,2016,9(1):235-244.

[19] FENG Y Z,SUN D W.Near-infrared hyperspectral imaging in tandem with partial least squares regression and genetic algorithm for non-destructive determination and visualization of Pseudomonas loads in chicken fillets[J].Talanta,2013,109:74-83.

[20] WU D,WANG S J,WANG N F,NⅠE P C,HE Y,SUN D W,YAO J S.Application of time series hyperspectral imaging (TSHSⅠ) for determining water distribution within beef and spectral kinetic analysis during dehydration[J].Food and Bioprocess Technology,2013,6(11):2943-2958.

[21] MⅠSHRA P,WOLTERⅠNG E,BROUWER B,ECHTELT E H V.Ⅰmproving moisture and soluble solids content prediction in pear fruit using near-infrared spectroscopy with variable selection and model updating approach[J].Postharvest Biology and Technology,2021,171:111348.

[22] CEN H Y,LU R F,MENDOZA F,BEAUDRY R M.Relationship of the optical absorption and scattering properties with mechanical and structural properties of apple tissue[J].Postharvest Biology and Technology,2013,85:30-38.

[23] WⅠLSON R H,NADEAU K P,JAWORSKⅠF B,TROMBERG B J,DURKⅠN A J.Review of short-wave infrared spectroscopy and imaging methods for biological tissue characterization[J].Journal of Biomedical Optics,2015,20(3):030901.

Fruit soluble solids content non-destructive detection based on visible/near infrared hyperspectral imaging in mango

LⅠN Jiaojiao1,2,MENG Qinghua1,2*,WU Zhefeng1,2,CHANG Hongjuan1,2,NⅠChunyu1,2,QⅠU Zouquan1,2,LⅠHuarong1,HUANG Yuqing3

(1School of Physics and Electronics, Nanning Normal University, Nanning 530001, Guangxi, China;2Guangxi Key Laboratory of Information Functional Materials and Intelligent Information Processing, Nanning Normal University, Nanning 530001, Guangxi, China;3Key Laboratory of Environmental Evolution and Resource Utilization of the Beibu Gulf,Ministry of Education&Guangxi/Key Laboratory of Earth Surface Processes and Intelligent Simulation/Nanning Normal University,Nanning 530001,Guangxi,China)

Abstract:【Objective】The city of Baise,located in Guangxi,China,exhibits a subtropical monsoon climate.The distinctive flavor of mangoes in this city is attributed to the unique combination of both climatic conditions and geographical environment.Baise's mango is characterized as a small core,high nutritional value and low fiber content,making it highly favored by consumers.Sugar content is an important indicator of the intrinsic quality of mangoes.With the increasing demand for mango grading and deep processing due to the improvement of people’s living standards,it is imperative to develop a simple, rapid and non-destructive technique for detecting mango brix content.However, most researchers have focused on developing detection models for single species or classes of fruits using spectrometers with low stability and weak universality that hinder the industrialization of scientific research outcomes.Therefore, this study aimed to explore the differences in brix spectra and characteristic response band ranges among different types of mangoes using NⅠR-HSⅠtechnology.The ultimate goal was to establish a high-precision detection model for sugar content in various fruits with Guifei mango and Tainong No.1 mango serving as research objects.【Methods】The hyperspectral image data were acquired using a hyperspectral imaging system.A total of 327 bands of hyperspectral images were obtained in the spectral range between 400-1000 nm for this experiment.The digital refractometer that we used was a portable digital refractometer PAL-1 from ATAGO, Japan.Measurements were taken three times independently,and the average value was calculated as the reference value for soluble solids in mango samples.After opening the original spectral image with ENVⅠsoftware and extracting the original spectral data within a pixel square 10×10, the average spectral data of each region were manually selected and extracted.Subsequently, MATLAB R2018b software was employed to perform spectral data modeling and original segmentation of the image data.The multiple scattering correction (MSC) algorithm was chosen to effectively reduce random noise in the spectral data, with its noise reduction effect being influenced by the number of smoothing points utilized.Therefore,MSC preprocessing was applied to process the spectral data accordingly.To model different types of mango brix values along with their corresponding spectral reflectance as training data, we employed the KS algorithm.The remaining brix values and their corresponding spectral reflectance were treated as test data.The PLS model can be utilized to select a smaller set of new variables that replaced a larger set without losing crucial spectral information.This addressed challenges posed by overlapping bands in spectroscopy analysis.【Results】The analysis of the spectral curves of different mango varieties showed that there were consistent overall trends among them.Notably, absorption peaks occurred at approximately 509, 680, 857 and 963 nm wavelengths.Ⅰn the red light region (680-750 nm), reflectance showed a distinct increasing trend with a steep slope formation.Thus,the characteristic wavebands for mango pulp can be identified as the range of 680-750 nm and specific bands at 509,550,680,857 and 963 nm.Within the range of 500-750 nm,Tainong No.1 mango exhibited significantly higher spectral reflectance compared to Guifei mango.Moreover, both fruits displayed steep slope formations in their spectral curves when sugar levels were similar;however,these slopes occurred at different positions.Specifically,Tainong No.1 mango's steep slope was observed around wavelengths of 500-640 nm while Guifei mango’s occurred around wavelengths of 680-750 nm.Both varieties exhibited absorption peaks near wavelengths of approximately 680 and 857 nm, while similar trends were displayed in spectral reflectance within the range of 750-1000 nm.The response of spectral reflectance to sugar content varied widely among different mango varieties; nevertheless, a strong correlation existed within the red light range (600-700 nm) for all varieties.Ⅰt was found that precise determination of characteristic wavelengths corresponding to chemical information in mangos remained challenging, which may impact model accuracy.Therefore, this issue needs to be addressed in future studies to enhance accurate prediction models for determining mango saccharinity.Combined with the spectral reflectance data of different mango varieties, we can analyze the effect of their respective band ranges on sugar content.The peak response was observed at about 670 nm with a correlation coefficient of 0.837, indicating the highest spectral sensitivity.Notably, the CARS-PLS prediction model exhibited superior accuracy and reliability in predicting mango brix levels.The regression analysis revealed an ideal correlation between measured and predicted values,represented by the equation y=0.851 5x+12.208 (R2=0.880 6).This relationship was further supported by a slope of 0.851 5, an intercept of 12.208, and RMSECV=0.636 6.The PLS model constructed using wavelengths with high correlation coefficients between brix and spectral reflectance in each band gave better results in predicting mango brix.【Conclusion】Both the calibration set and the prediction set showed that the predicted values were very close to the corresponding actual values.The results showed that it was feasible to apply hyperspectral imaging technology to detect mango brix.This study successfully employed NⅠR-HSⅠtechnology to analyze the differences in spectral and characteristic response bands of mangoes with varying sugar contents.The developed high-precision detection model demonstrated promising results in predicting mango brix.These findings have validated the feasibility of employing hyperspectral imaging technology for mango brix detection,with great potential applications in mango grading and processing.Further research is warranted to enhance accurate saccharinity prediction by precisely identifying characteristic wavelengths associated with chemical information in mangoes.

Key words:Mango;Near-infrared(NⅠR);Hyperspectral imaging(HSⅠ);Soluble solids content;Nondestructive testing;Spectral difference

中图分类号:S667.7

文献标志码:A

文章编号:1009-9980(2024)01-0122-11

DOⅠ:10.13925/j.cnki.gsxb.20230269

收稿日期2023-07-19

接受日期:2023-11-12

基金项目广西科技基地和人才专项(桂科AD20238059);广西学位与研究生教育改革项目(JGY2022220);广西普通本科高校示范性现代产业学院-南宁师范大学智慧物流产业学院建设项目示范性现代产业学院(6020303891823)

作者简介林娇娇,女,在读硕士研究生,主要研究方向为近红外高光谱成像。E-mail:615912553@qq.com

*通信作者Author for correspondence.E-mail:mqhgx@163.com