Ma Zhiling, Wei Changbin, Wang Wenhui, Lin Wenqiu, Nie Heng, Duan Zhe, Liu Ke, Xiao Xi Ou
South Subtropical Crop Research Institution, Chinese Academy of Tropical Agricultural Sciences, Key Laboratory of Tropical Fruit Biology, Ministry of Agriculture & Rural Affairs, Key Laboratory of Hainan Province for Postharvest Physiology and Technology of Tropical Horticultural Products, Academy of Tropical Agricultural Sciences, Zhanjiang Key Laboratory of Tropical Crop Genetic Improvement, Zhanjiang, Guangdong, China.
Yunnan Agricultural University, Puer, Yunnan, China.
PeerJ. 2024 May 14;12:e17379. doi: 10.7717/peerj.17379. eCollection 2024.
Accurately detecting the anthocyanin content in eggplant peel is essential for effective eggplant breeding. The present study aims to present a method that combines hyperspectral imaging with advanced computational analysis to rapidly, non-destructively, and precisely measure anthocyanin content in eggplant fruit. For this purpose, hyperspectral images of the fruits of 20 varieties with diverse colors were collected, and the content of the anthocyanin were detected using high performance liquid chromatography (HPLC) methods. In order to minimize background noise in the hyperspectral images, five preprocessing algorithms were utilized on average reflectance spectra: standard normalized variate (SNV), autoscales (AUT), normalization (NOR), Savitzky-Golay convolutional smoothing (SG), and mean centering (MC). Additionally, the competitive adaptive reweighted sampling (CARS) method was employed to reduce the dimensionality of the high-dimensional hyperspectral data. In order to predict the cyanidin, petunidin, delphinidin, and total anthocyanin content of eggplant fruit, two models were constructed: partial least squares regression (PLSR) and least squares support vector machine (LS-SVM). The HPLC results showed that eggplant peel primarily contains three types of anthocyanins. Furthermore, there were significant differences in the average reflectance rates between 400-750 nm wavelength ranges for different colors of eggplant peel. The prediction model results indicated that the model based on NOR CARS LS-SVM achieved the best performance, with a squared coefficient of determination (R) greater than 0.98, RMSEP and RMSEC less than 0.03 for cyanidin, petunidin, delphinidin, and total anthocyanin predication. These results suggest that hyperspectral imaging is a rapid and non-destructive technique for assessing the anthocyanin content of eggplant peel. This approach holds promise for facilitating the more effective eggplant breeding.
准确检测茄子果皮中的花青素含量对于有效的茄子育种至关重要。本研究旨在提出一种将高光谱成像与先进的计算分析相结合的方法，以快速、无损且精确地测量茄子果实中的花青素含量。为此，收集了20个不同颜色品种果实的高光谱图像，并使用高效液相色谱（HPLC）方法检测花青素含量。为了最小化高光谱图像中的背景噪声，对平均反射光谱采用了五种预处理算法：标准归一化变量变换（SNV）、自动标度（AUT）、归一化（NOR）、Savitzky-Golay卷积平滑（SG）和均值中心化（MC）。此外，采用竞争性自适应重加权采样（CARS）方法来降低高维高光谱数据的维度。为了预测茄子果实中的矢车菊素、矮牵牛素、飞燕草素和总花青素含量，构建了两个模型：偏最小二乘回归（PLSR）和最小二乘支持向量机（LS-SVM）。HPLC结果表明茄子果皮主要含有三种类型的花青素。此外，不同颜色茄子果皮在400 - 750 nm波长范围内的平均反射率存在显著差异。预测模型结果表明，基于NOR CARS LS-SVM的模型性能最佳，矢车菊素、矮牵牛素、飞燕草素和总花青素预测的决定系数平方（R）大于0.98，RMSEP和RMSEC小于0.03。这些结果表明高光谱成像是一种快速无损的技术，可用于评估茄子果皮中的花青素含量。这种方法有望促进更有效的茄子育种。
