Optimization of Characteristic Wavelengths in Prediction of Lycopene in Tomatoes Using Near‐Infrared Spectroscopy

Abstract To improve the predictive ability and robustness of the near‐infrared correction model and to simplify the model, the backward interval partial least squares, the synergy interval partial least squares, the uninformative variable elimination partial least squares and the genetic algorithm partial least square ( GA‐PLS ) methods were used to select the characteristic wavelengths in the prediction of lycopene in tomatoes. The optimal characteristic variables and regression model were determined by the model evaluation parameters. The best model was set up using the GA‐PLS method. Compared with the model based on the full spectra, the variables used were reduced from 1,816 to 142, the correlation coefficient increased from 0.7104 to 0.9072, the root mean square errors of cross‐validation and prediction decreased from 21.58 to 8.76 and from 22.03 to 8.93, respectively. The experimental results showed that the use of GA‐PLS method, in the selection of characteristic variables of tomato lycopene, could effectively reduce the number of variables, decrease model complexity and improve the predictive precision. Practical Applications Using near‐infrared spectroscopy could quickly detect the lycopene contents of tomatoes. By extracting spectra characteristic wavelengths, it could reduce the irrelevant information variables and improve the accuracy of measurement of lycopene in tomatoes.