利用太赫兹光谱技术构建番茄水分胁迫状态检测模型

快速检测番茄水分胁迫状态，对于科学有效地进行番茄的水肥管理，保障和提高番茄的产量和品质具有重要意义。该研究利用太赫兹光谱对水分极为敏感的特性，提出了基于太赫兹光谱技术的番茄水分胁迫状态的快速检测方法。试验利用太赫兹光谱系统获取不同水分胁迫番茄叶片的功率谱、吸光度及透射率频谱数据。采用（Savitzky-Golay, SG）算法对数据进行降噪，利用稳定性竞争自适应重加权（Stability Competitive Adaptive Reweighted Sampling, SCARS）算法进行了多维特征频段的提取；在此基础上，建立了叶片含水率功率谱、吸光度及透射率等单一维度下的多元线性回归（Multiple Linear Regression, MLR）模型。结果表明，太赫兹功率谱和吸光度与叶片含水率之间呈负相关；而透射率则随水分胁迫程度的提高逐渐升高，呈正相关。为了进一步提高模型的精度，使用支持向量机（Support Vector Machines, SVM）在融合3种维度太赫兹特征的基础上，建立了番茄含水率的融合预测模型，结果表明，预测集R2达到0.951 4，RMSE为0.366 8，均高于单一维度检测模型，实现了番茄水分的快速检测。

Tomato water stress state detection model by using terahertz spectroscopy technology

Rapid detection of water stress is of great significance for scientific and effective management of water and fertilizer, further improving the yield and quality of tomatoes. In this study, a new detection model was proposed for water stress state in tomatoes using terahertz spectroscopy. "Hezuo 906" tomato was taken as the research object. A systematic experiment was performed in a Venlo-type greenhouse at the Key Laboratory of Modern Agricultural Equipment and Technology, Ministry of Education, Jiangsu University, Zhenjiang Province, China. The soilless culture was adopted, where the matrix was perlite. Kawasaki nutrient solution was used to provide the same nutritional environment for the samples. Artificial ventilation was adopted to ensure the temperature and humidity in the greenhouse in the appropriate range. Water and fertilizer were controlled precisely to ensure the balance of nutrient elements. Four water stresses were set at 20%, 40%, 60%, and 80% of the standard irrigation amount from 5 days after transplanting. Each gradient was repeated 10 times. The pinnate compound leaves of inverted 6 leaves were collected on the 65th day after the water stress treatment, particularly representing the growth state of tomatoes. 20 samples were collected for each water stress treatment in a total of 80 samples. Samples were dried for subsequent characterization. A terahertz spectral system was then utilized to acquire the power spectrum, absorbance, and transmittance spectrum of tomato leaves under different water stress. Savitzky-golay (SG) was used to reduce the noise of data. Stability competitive adaptive reweighted sampling (SCARS) was used to extract multi-dimensional characteristic frequency bands. Multiple linear regression (MLR) models were established between tomato moisture content and power spectrum, tomato moisture content and absorbance, tomato moisture content, and transmission. The results showed that the terahertz power spectrum and the absorbance were negatively correlated with the water content of blades in the frequency range of 0.5-1.5 THz. However, the transmittance gradually increased with the increase of water stress, showing a positive correlation. Among them, the model presented the best performance, when using the characteristics of the power spectrum in the frequency domain. Specifically, the determination coefficient of the prediction set was 0.900 7, and the root mean square error (RMSE) of the prediction set was 0.482 5. Furthermore, a fusion prediction model was established for tomato moisture content using support vector machines (SVM) on the basis of integrating three dimensions of terahertz features of absorbance, transmittance, and power spectrum, in order to further improve the accuracy of the model. It was found that R2 of the prediction set was 0.951 4, while RMSE of the prediction set was 0.3668, indicating higher than the single-dimensional detection model. The improved model can be applied to detect the moisture content of tomato leaves using terahertz time-domain spectroscopy. The finding can provide a sound foundation for the detection of crop water stress.