Hyperspectral characteristic wavelength selection method for moldy maize based on continuous projection algorithm fusion information entropy

Due to the number of spectral bands, large amount of data and high information redundancy, it is more difficult to identify the moldy maize samples using hyperspectral technology. In order to reduce the amount of data and obtain the most useful feature wavelengths for identifying the moldy maize samples using hyperspectral information, in this paper, a method of feature wavelength selection is proposed with the help of the continuous projections algorithm (SPA) coupled with the information entropy. Firstly, the hyperspectral data of moldy maize samples were subjected to pre-preprocess using multiplicative scatter correction (MSC) so as to eliminate signal noise. Then the continuous spectral algorithm was used to initially select a few wavelengths on the processed spectra to obtain eight primary feature wavelengths. And then the image information corresponding to the eight primary feature wavelengths was processed through the information entropy principle to obtain the best feature wavelength. The results show that the optimal wavelength for the identification of moldy corn is 819 nm by using the continuous projection algorithm fusion information entropy method. After extracting the texture features of the moldy maize images at the wavelength, Fisher Discriminant Analysis(FDA)was used to identify these moldy maize samples, and the correct discrimination rate of the 6 grades of moldy maize was up to 98.6%. This feature wavelength selection method can provide guidance for better use of hyperspectral techniques to identify the grades of moldy maize. (authors)