Abstract: In order to improve the accuracy and reliability of the prediction results of potato VC content, a method for constructing input variables of detection model based on Fisher discriminant analysis (FDA) separable data fusion was proposed. Firstly, hyperspectral information of 200 potato samples was collected by hyperspectral imaging technology; and by comparing the modeling results the spectral data before and after the preprocessing by six methods, the multiplicative scatter correction was determined as the optimal preprocessing method. Secondly, competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA) and CARS-SPA combination algorithm were used to extract the corresponding feature wavelengths, and 34 effective feature wavelengths were finally determined through comparative analysis. Thirdly, the effective feature wavelengths were fused by FDA to achieve the separability of the data, and the fused new variables was screened according to the discrimination capacity for different samples, so as to determine the input variables of the detection model to be constructed. Finally, the partial least squares model and BP neural network model were established for the variables screened before and after FDA fusion, respectively, and the detection results were compared and analyzed. The results show that the correlation coefficient of the BPNN model is increased from 0.9726 to 0.9990, and the root mean square error is also reduced from 0.7723 to 0.1727 when the first three fused variables of 34 feature wavelengths extracted by CARS algorithm after FDA fusion are used as the input variables for the detection model, which not only reduces data analysis dimension, but also improves its detection ability. Therefore, it is suitable and effective to construct input variables of detection model based on FDA separable data fusion to improve the accuracy of potato VC content detection results.

Key words: Hyperspectral imaging, Fisher discriminant analysis, Potato, VC content detection, Model