Authors

Abstract

Millets are one of China's primary traditional food crops, and drought can adversely impact their yield and quality. To quickly detect the degree of drought stress in cereal grains, this study establishes a nondestructive classification model based on hyperspectral imaging technology. The raw spectral data underwent preprocessing using six pretreatment methods and various combinations of these methods. Subsequently, three distinct algorithms were employed for feature wavelength selection. To assess the severity of drought stress on millet, classification models were developed by integrating texture and color features, utilizing Support Vector Machine (SVM), Partial Least Squares Discriminant Analysis (PLS-DA), and Multilayer Perceptron (MLP) algorithms. The results indicate that the D1st-SVM model, based on CARS wavelength selection, exhibits the highest modeling performance when feature wavelengths are fused with significant texture and color variables, achieving an accuracy rate of 93%. These findings suggest that drought identification in millet can be performed quickly and nondestructively by integrating image features through hyperspectral imaging technology.

Abstract in Chinese

谷子是我国传统主要粮作之一，干旱对其的产量和品质都会产生不利影响。为了快速检测谷子受干旱胁迫程度，本研究基于高光谱成像技术建立一种无损的分类模型。本研究对原始光谱数据进行六种预处理方法，以及这些方法的不同组合，对光谱数据中的噪声进行处理。采用3种不同算法进行特征波长的选取。融合纹理特征和颜色特征，基于支持向量机（support vector machine，SVM）、最小二乘判别分析（partial least-squares discriminant analysis，PLS-DA）和多层感知机算法（Multilayer Perceptron，MLP）建立分类模型，来识别谷子受干旱胁迫程度。结果表明特征波段融合重要纹理特征、重要颜色特征变量时，基于CARS波长选择的D1st-SVM模型的建模性能最高，预测集的分类准确度为93%。研究结果表明，利用高光谱成像技术融合图像特征可以快速、无损地识别谷子是否受到干旱胁迫。