Detection of moisture content of tomato leaves based on dielectric properties and IRIV-GWO-SVR algorithm

Abstract: Tomato is one of the important crops in the world. Tomato contains a lot of vitamins and a variety of nutritional elements, which is popular among people. Water is indispensable to plants and the shortage of such critical resource leads to a decline in yield and quality of crops. Leaf moisture content is an important factor which can show water scarcity in crop. In order to realize more reasonable irrigation management during the growth of tomato, a new method for accurately, rapidly and effectively detecting tomato leave moisture based on dielectric properties was proposed in this study. Firstly, the dielectric properties (relative dielectric constant and dielectric loss factor) of 300 pieces of tomato leaves with different moisture contents were measured with an LCR measuring instrument at 37 discrete frequencies over the frequency range of 0.05-200 kHz, and the moisture contents of the tomato leaves were measured by dry weight method. Secondly, the iteration retaining informative variables (IRIV) algorithm was used to extract the characteristic variables of dielectric properties of tomato leaf samples, and simultaneously the effect of IRIV was compared with that of the successive projections algorithm (SPA) in order to determine the optimal method for characteristic variable selection. Finally, the support vector regression (SVR) machine was adopted to establish the relationship models between full variables, 2 kinds of characteristic variables and moisture content of tomato leaf samples, respectively. And the performances of all the models were evaluated by the determination coefficient for calibration set (RC2), root mean square error for calibration set (RMSEC), determination coefficient for prediction set (RP2) and root mean square error for prediction set (RMSEP). The research results showed that the measurement frequency and moisture content had a significant effect on the dielectric properties of tomato leaves. Between 0.05 and 200 kHz, the relative dielectric constant and dielectric loss factor decreased with the increase of the test frequency. When the frequency was less than 20 kHz, the decline was obvious. When the frequency was more than 20 kHz, the decline was not obvious. Between 0.05 and 200 kHz, the relative dielectric constant and dielectric loss factor increased with the increase of the moisture content of tomato leaves. And the results showed that IRIV-SVR model performed better than the other models with full-SVR and SPA-SVR, achieving the highest accuracy with RP2 =0.8721 and RMSEP=0.0390. Whereas, for the prediction accuracy of IRIV-SVR model, the desired effect was not achieved. For improving the prediction accuracy of SVR model, the grey wolf optimizer (GWO) algorithm was further introduced to intelligently optimize the parameters in the SVR model to find the optimum values. Consequently, the optimized model, IRIV-GWO-SVR, achieved the RP2 of 0.9638 and RMSEP of 0.0207, which proved that the method of selecting characteristic variables by IRIV algorithm combined with optimizing the parameters in SVR model by GWO algorithm can extremely raise the performance of prediction model for the moisture content of leaves. Hence, the method of dielectric properties technology combined with the IRIV-GWO-SVR model is feasible for detecting the moisture content of tomato leaves, also hopefully providing a new method and thought for water detection of other crops.