Suratanee Apichat, Chutimanukul Panita, Saelao Tanapon, Chadchawan Supachitra, Buaboocha Teerapong, Plaimas Kitiporn
Department of Mathematics, Faculty of Applied Science, King Mongkut's University of Technology North Bangkok, Bangkok, Thailand.
Intelligent and Nonlinear Dynamic Innovations Research Center, Science and Technology Research Institute, King Mongkut's University of Technology North Bangkok, Bangkok, Thailand.
PLoS One. 2024 Oct 2;19(10):e0309132. doi: 10.1371/journal.pone.0309132. eCollection 2024.
Hyperspectral imaging has emerged as a powerful tool for the non-destructive assessment of plant properties, including the quantification of phytochemical contents. Traditional methods for antioxidant analysis in holy basil (Ocimum tenuiflorum L.) are time-consuming, while hyperspectral imaging has the potential to rapidly observe holy basil. In this study, we employed hyperspectral imaging combined with machine learning techniques to determine the levels of total phenolic contents in Thai holy basil. Spectral data were acquired from 26 holy basil cultivars at different growth stages, and the total phenolic contents of the samples were measured. To extract the characteristics of the spectral data, we used 22 statistical features in both time and frequency domains. Relevant features were selected and combined with the corresponding total phenolic content values to develop a neural network model for classifying the phenolic content levels into 'low' and 'normal-to-high' categories. The neural network model demonstrated high performance, achieving an area under the receiver operating characteristic curve of 0.8113, highlighting its effectiveness in predicting phenolic content levels based on the spectral data. Comparative analysis with other machine learning techniques confirmed the superior performance of the neural network approach. Further investigation revealed that the model exhibited increased confidence in predicting the phenolic content levels of older holy basil samples. This study exhibits the potential of integrating hyperspectral imaging, feature extraction, and machine learning techniques for the rapid and non-destructive assessment of phenolic content levels in holy basil. The demonstrated effectiveness of this approach opens new possibilities for screening antioxidant properties in plants, facilitating efficient decision-making processes for researchers based on comprehensive spectral data.
高光谱成像已成为一种用于植物特性无损评估的强大工具，包括对植物化学成分含量的定量分析。传统的圣罗勒（Ocimum tenuiflorum L.）抗氧化剂分析方法耗时较长，而高光谱成像有潜力快速检测圣罗勒。在本研究中，我们采用高光谱成像结合机器学习技术来测定泰国圣罗勒中总酚含量水平。从26个处于不同生长阶段的圣罗勒品种获取光谱数据，并测量样本的总酚含量。为提取光谱数据的特征，我们在时域和频域中使用了22个统计特征。选择相关特征并将其与相应的总酚含量值相结合，以建立一个神经网络模型，将酚含量水平分为“低”和“正常至高”两类。该神经网络模型表现出高性能，接收器操作特征曲线下的面积达到0.8113，突出了其基于光谱数据预测酚含量水平的有效性。与其他机器学习技术的对比分析证实了神经网络方法的优越性能。进一步研究表明，该模型在预测较老圣罗勒样本的酚含量水平时具有更高的置信度。本研究展示了将高光谱成像、特征提取和机器学习技术相结合用于快速无损评估圣罗勒中酚含量水平的潜力。这种方法所展示的有效性为筛选植物中的抗氧化特性开辟了新的可能性，有助于研究人员基于全面的光谱数据进行高效决策。
