Abstract: Due to the time and energy consumption of traditional detection methods, a portable hyperspectral imager was used to rapidly nondestructive detection of chlorophyll content in tomato leaves under briny water irrigation. By collecting images of canopy leaves under different processing conditions, the spectral data of the leaves were extracted by the analysis software, and the outliers removal, preprocessing, characteristic wavelength extraction, model construction and visual distribution research were carried out. The results showed that the outlier elimination optimized the model and the original spectrum for data analysis. PLSR（Partial least squares regression） method was used to model the characteristic wavelength extracted by different methods and optimize the characteristic wavelength extracted by SPA. On this basis, the MLR（Multiple linear regression）, PCR（Principal component regression）, PLSR, SVR（Support vector machine regression）, and ANN（Artificial neural network） models were used to model the characteristic wavelengths extracted by SPA, and the different modeling effects were compared and analyzed. The MLR, PCR, and PLSR models were preferably selected. The optimal characteristic wavelengths were 392、465、686 and 760 nm, the prediction correlation coefficient of the optimal model was 0.896, and the prediction root-mean-square error was 1.111. Finally, the PLSR model was optimized to predict the chlorophyll content of tomato by visualizing three linear models. The optimized characteristic bands can be used to quantitatively and visually analyze the chlorophyll content of tomato leaves in the future.

Key words: hyperspectral imaging, tomato leaves, chlorophyll, visual distribution, brackish water