货架期是影响果蔬品质和供应安全的重要因素, 快速准确预测果蔬货架期已成为消费者、 生产者和管理者共同关注的问题。 猕猴桃含有多种有机物和氨基酸, 具有丰富的营养价值, 深受广大消费者的喜爱。 但由于猕猴桃表面颜色变化不明显, 人们仅凭感官难以准确判断猕猴桃的货架期和质量等级。 采用高光谱成像结合化学计量学方法对不同储存条件下的保鲜猕猴桃进行了货架期预测。 首先采集了4 ℃和(18±2) ℃下保鲜时间为0, 2, 4天各120个猕猴桃样本在400～1 000 nm的高光谱数据, 测定其硬度值和可溶性固形物含量(SSC), 获取猕猴桃切片高光谱图像。 对猕猴桃平均光谱提取并进行Savitzky-Golay卷积平滑预处理后, 通过光谱数据主成分分析(PCA), 发现不同货架期和储存温度的猕猴桃样本在前2个主成分空间形成一定的聚类, 4 ℃下猕猴桃样本出现少量重叠。 为了减少波长变量, 提高运算速度, 使用载荷系数法(XL)与连续投影算法(SPA)选择特征波长。 其中, 4 ℃猕猴桃样本的XL和SPA特征波长分别7个(481, 501, 547, 665, 723, 839, 912 nm)和10个(406, 428, 520, 617, 665, 682, 723, 818, 878和983 nm); (18±2) ℃猕猴桃样本XL特征波长为508, 545, 665, 672, 720, 839和909 nm, SPA特征波长为575, 622, 731, 756, 779, 800, 828, 865, 920和983 nm。 基于3∶1的光谱数据集划分三个货架期虚拟等级值1, 2和3, 以全光谱数据、 特征波长为输入, 建立非线性最小二乘支持向量机(LS-SVM)预测模型。 结果表明, 对于4 ℃下3种货架期猕猴桃样本, 在全光谱和XL, SPA特征波长上的预测集准确度分别达到92.2%, 92.2%, 91.1%; (18±2) ℃时预测准确度均为100%。 猕猴桃切片图像PCA分析显示, 除PC5中有部分噪声影响外, 其他主成分图像均能完整反映猕猴桃切片信息, PC2图像可以明显呈现出猕猴桃切片在不同货架期的变化程度。 进一步分析猕猴桃硬度和可溶性固形物含量发现, 随着货架期延长, 猕猴桃可溶性固形物含量逐渐增加, (18±2) ℃时二者存在正相关性, 相关系数为0.557 6。 硬度则随货架期延长逐渐减小, 4和(18±2) ℃下硬度值和货架期之间存在负相关性, 相关系数分别为-0.335 6和-0.562 0。 结合猕猴桃光谱信息, 可以发现猕猴桃光谱反射率与其单个理化指标不成线性关系, 而是多个指标的综合反映。 因此, 采用高光谱成像技术可以全面、 准确、 快速的预测猕猴桃货架期, 为猕猴桃的生产、 销售提供技术指导。

The shelf-life of fruits and vegetables is an important factor that affects the quality, which is concerned by the consumers, farmers and producers. Kiwifruit contains a variety of organic substances and amino acids, which has rich nutritional value and is deeply loved by consumers. However, due to its own characteristics such as the color characteristics of kiwifruit, it is difficult for consumers to make an accurate judgment on the edible degree of kiwifruit in the shelf-life by sensory evaluation. Therefore, non-destructive testing of the shelf life of fruits and vegetables is vital for agricultural products. In this research, hyperspectral imaging technology with chemometric methods was employed to estimate the shelf-life of kiwifruits which were stored in 4 ℃ and (18±2) ℃ among 3 periods (0, 2, 4 d). The spectral data covering the range of 400～1 000 nm were collected from 720 kiwifruit samples of 3 periods at 4 ℃ and (18±2) ℃. Meanwhile, the firmness and solid soluble content (SSC) of kiwifruits were measured, and the spectral data of kiwifruit slices were collected. The mean spectra (90 kiwifruits in the training set and 30 kiwifruits in prediction set) were extracted from each kiwifruit. Then, principal component analysis (PCA) was implemented for samples stored at different temperatures. Cluster analysis was performed based on PC1, while some overlap phenomenon showed in kiwi samples at 4 ℃. X-loadings of principal component analysis (PCA) and successive projection algorithm (SPA) method were applied to select the effective wavelengths, which are helpful for enhancing computer velocity. Based on X-loadings, 7 wavelengths (481, 501, 547, 665, 723, 839, 912 nm) were selected for samples stored at 4 ℃ and 7 wavelengths (508, 545, 665, 672, 720, 839, 909 nm) were selected for samples stored at (18±2) ℃, respectively. Similarly, for the SPA method, 10 wavelengths (406, 428, 520, 617, 665, 682, 723, 818, 878, 983nm) were selected for samples stored at 4 ℃ and 10 wavelengths (575, 622, 731, 756, 779, 800, 828, 865, 920, 983 nm) were selected for samples stored at (18±2) ℃, respectively. Thereafter, virtual levels (1, 2, 3) were assigned to the samples of 3 periods at 4 ℃ and (18±2) ℃, respectively. Least square-support vector machine (LS-SVM) was used to build classification models on full spectral data, effective wavelengths selected based on PCA and SPA, respectively. The results showed that the accuracy of the predictions reached to 92.2%, 92.2% and 91.1% among 3 periods at 4 ℃ and the accuracy of the predictions reached to 100% among 3 periods at (18±2) ℃, respectively. Also, the firmness and SSC of kiwifruits were measured and analyzed by one-way analysis of variance (ANOVA), the results showed that there was a negative correlation between firmness and shelf-life and the correlation coefficient was -0.335 6 and -0.562 0 at 4 ℃ and (18±2) ℃, respectively. There was a positive correlation between SSC and shelf-life and the correlation coefficient was 0.557 6 at (18±2) ℃. The shelf-life of kiwifruits can be estimated by the firmness index of samples stored atboth 4 ℃ and (18±2) ℃. While the SSC of samples stored at (18±2) ℃ was a significant estimation index. Further, the images of PC1-PC7 can preserve the integrity of the kiwifruit slice surface information, PC2 image can clearly show the degree of kiwifruit slices with different shelf-life. The results of this study indicate that it is feasible to use the hyperspectral imaging technique combined with the chemometric methods to classify the shelf-life of kiwifruits. Meanwhile, this research realized the rapid prediction of the shelf-life of kiwifruits and provided theoretical support for the quality and classification of fruit and vegetable shelf-life. Further, this study help forproviding technical supports for the developed instruments used for real time estimating the shelf-life of fruits and vegetables in further study.