A novel wavelength selection algorithm based on permutation analysis and variable combination

As one of the most important preprocessing procedures in spectral detection, wavelength selection plays an irreplaceable role in reducing the model computation consumption and prediction error. However, many wavelength selection methods only use a mono-directional strategy to capture the synergetic influence between wavelength points, which causes information loss and limits their performance. Herein, a novel algorithm called bidirectional permutation analysis (BPA) is proposed. To completely capture the synergetic influence between each wavelength point, we first modify the permutation analysis to select wavelength points that have a high synergetic effect with any other wavelength points. Then, the variable combinational population analysis (VCPA) is used to select other wavelength points from another direction that whether they are suitable to combine with front selected wavelengths. Benefited from this bidirectional selection structure, BPA could more comprehensively evaluate the synergetic effect between wavelength points and avoid the loss of useful information. The performance of BPA was estimated by using the spectral common dataset of corn oil. The comparison results indicate that our algorithm performs better on both accuracy and stability than four state-of-the-art algorithms: CARS, VCPA, VCPA-GA, and VCPA-IRIV.