Key Wavelengths Screening Based on Causality Analysis in Predicting Apple Brix

This study proposed an improved variable selection method based on causality analysis(CA), which effectively mined the influence mechanism of sugar substances on the apple brix detection, to improve the accuracy of prediction model. The approach is: by selecting the feature wavelengths with strong causal relationship with weighted to screen the strongest feature set correlating with the target variable. The selected wavelengths set is used as input to construct the prediction model based on partial least squares. The experimental analysis on the "Red Fuji" apple spectral dataset showed that the improved model improved the R2 by 7%, decreased the RMSE by 24% and reduced the number of characteristic wavelengths from 161 to 88 compared to the traditional method. The accuracy of the prediction model was improved. The complexity of the model was also optimized. This study demonstrates the feasibility of using visible-near infrared spectroscopy to detect the internal quality of apples as a priori knowledge. In turn, it provides a new idea for the feature wavelength selection method based on causal analysis.