|
By using wavelength selection methods to eliminate irrelevant and redundant information from near-infrared spectroscopy data before modeling, the performance of the apple sugar degree prediction model can be improved. However, there are many shortcomings in commonly used single-wavelength selection methods. This paper combines the competitive reweighted sampling (CARS) algorithm, the successive projection algorithm (SPA), and the genetic algorithm (GA), which are three primary single-wavelength selection methods, to merge the wavelength subsets obtained from them. These subsets are then combined through intersection and union fusion, resulting in 8 Venetian wavelength subsets. A fructose prediction model for apples is established by combining them with the Ridge Regression model. The results show that the number of wavelength subsets obtained by the preferred CARS∪SPA method is 74, accounting for 14.4% of the total wavelengths. Its R2 p is maximum at 0.952, and the minimum RMSEp is 0.036. This indicates that the wavelength selection method of CARS∪SPA reduces the multicollinearity of spectral data, effectively utilizes complementary information in the wavelength variables obtained by CARS and SPA methods, and enhances the robustness and accuracy of the model. The research work can provide theoretical basis for the rapid grading of apple quality and the development of related detection devices.
|
