Multi-pattern Recognition Technology based Feature Selection and Processing of Apple Near Infrared Spectrum

In order to improve the accuracy of the prediction model of soluble solids content in Red Fuji apple, this paper processes the spectrum serially via Wavelet Packet Threshold Denoising method, Multivariate Scattering Correction(MSC) and Mahalanobis Distance(MD) Eliminating Anomaly Sample Method, and then extracts spectral features via Partial least squares (PLS) method and Wavelet packet transform (WPT) method. This paper proposes a feature selection method based on ant colony algorithm that uses the pheromone and heuristic information of ant colony to select the optimal feature combination. The dimension of feature combination selected in this paper is 20. We uses PLS to establish the prediction model, and gainthe correlation coefficient as 0.9435, which shows that the processed data improves the precision of the model. The prediction model of soluble solids in apples established can be applied as a substitution to full-band information.