番茄果实营养丰富备受人们喜爱。 番茄生长周期长, 需水量大, 水分含量是影响番茄植株生长发育的主要因素; 快速发现番茄植株水分亏缺状态, 对于科学有效地进行番茄的灌溉管理, 保障和提高番茄的产量和品质具有重要意义。 利用高光谱成像技术, 实时识别番茄叶片干旱胁迫程度, 提出了一种基于高光谱成像技术的番茄叶片干旱胁迫的识别方法。 首先, 选取红樱桃番茄为实验品种, 在室内培养12盆番茄幼苗。 在保证其他管理措施相同的基础上, 通过控制施水量来控制番茄的胁迫状态, 干旱胁迫程度设计3个处理(适宜水分、 中度和重度胁迫)。 分批次采集不同干旱程度番茄幼苗嫩叶在400~1 000 nm范围的高光谱图像, 并提取了每个样本的光谱和纹理特征。 使用标准化(Norm)、 多元散射校正(MSC)、 一阶导数(1st)和标准正态变量变换(SNV)四种预处理方法对光谱数据进行预处理去除光谱中的噪声。 使用连续投影算法(SPA)、 竞争性自适应重加权算法(CARS)以及竞争性自适应重加权算法结合连续投影算法(CARS-SPA)选取光谱重要特征波段, 用灰度梯度共生矩阵(GLGCM)提取番茄叶片的纹理特征, 用SPA选择纹理特征的重要变量。 融合重要光谱特征与重要纹理特征结合支持向量机(SVM)构建识别番茄干旱胁迫模型, 同时选用自适应增强算法(AdaBoost)与K-近邻(KNN)与SVM模型对比。 结果表明, 融合重要光谱特征与重要纹理特征后, 基于CARS-SPA波长选择的SNV-SVM模型具有最好的分类效果, 训练集的分类准确度(ACCT)为94.5%, 预测集的分类准确度(ACCP)为95%, AdaBoost模型分类效果次之ACCT为86.5%, ACCP为87%, KNN模型分类效果最差ACCT为81.5%, ACCP为79%。 因此, 该方法对番茄叶片干旱胁迫程度实时识别有较好的效果, 可为构建智能化的干旱胁迫分析技术提供参考。

Tomato is rich in nutrition, which most people love. It has a long growth cycle and requires a lot of water, and water content is the main factor influencing the tomato plant’s growth and development. It is of great significance to find out the water deficit state of tomato plants quickly for scientific and effective irrigation management of tomatoes, guaranteeing and improving the yield and quality of tomatoes. In this study, hyperspectral imaging technology was used to identify the degree of drought stress on tomato leaves in real-time, and a recognition method of drought stress on tomato leaves based on hyperspectral imaging technology was proposed. Firstly, a red cherry tomato was selected as the experimental variety, and 12 POTS of tomato seedlings were cultured in the laboratory. On the basis of ensuring the same as other management measures, the stress state of the tomato was controlled by controlling the amount of water applied. Then, three treatments (suitablewater, moderate and severe stress) were designed for the degree of drought stress. The hyperspectral images in the 400~1000nm range of young leaves of tomato seedlings with different drought degrees were collected in batches, and each sample’s spectral, texture characteristics were extracted. The spectral features were pre-processed using four methods, namely normalization (Norm), multiple scattering correction (MSC), first derivative (1st) and standard normalized variate (SNV) to remove noise from spectral data. The important feature bands of the spectral features were selected using the successive projections algorithm (SPA), competitive adaptive reweighting algorithm (CARS) and the competitive adaptive reweighting algorithm combined with the continuous projection algorithm (CARS-SPA). The texture features of tomato leaves were extracted by the Gray Level-Gradient Co-occurrence Matrix (GLGCM), and the important variables of texture features were selected by SPA. Finally, a support vector machine (SVM) was applied to fuse the above-mentioned various features to build a tomato drought stress recognition model, and Adaptive boosting (AdaBoost), and K-Nearest Neighbor (KNN) were used to compare and analyze with SVM model. The results showed that the SNV-SVM model based on CARS-SPA wavelength selection has the best classification effect after the fusion of important spectral features and texture features. The classification accuracy of the training set (ACCT) is 94.5%, and the classification accuracy of the prediction set (ACCP) is 95%. The adaBoost model had the second highest classification effect, ACCT 86.5%, and ACCP 87%. KNN model had the worst classification effect, with ACCT 81.5%, and ACCP 79%. Therefore, the method presented in this paper has a good effect on the real-time recognition of the drought stress degree of tomato leaves and can provide a reference for the construction of intelligent drought stress analysis technology.