Nondestructive detection for SSC and firmness of plums by hyperspectral imaging and artificial neural network

Abstract

Hyperspectral imaging technique and artificial neural network were used to investigate the feasibility of the nondestructive prediction for firmness and soluble solids content (SSC) of “Red” and “Green” plums. And the standard normal variation (SNV) was adopted to preprocess original spectral reflectance of region of interests. Then 5 and 28 characteristic wavelengths were selected from 256 full wavelengths by the methods of successive projection algorithm (SPA) and competitive adaptive reweighted sampling (CARS), respectively. An error back propagation (BP) network model was proposed based on selected characteristic variables to predict firmness and SSC of plums. The SSC prediction accuracy of CARS-BP model in calibration set (<i>r</i>_c= 0.989, RMESC = 0.451 &deg;Brix) was slightly higher than SPA-BP model (<i>r</i>_c= 0.978, RMESC = 0.589 &deg;Brix), while the SSC prediction accuracy of SPA-BP model in prediction set (<i>r</i>_p = 0.964, RMESP = 0.778 &deg;Brix) was slightly higher than CARS-BP model (rp = 0.955, RMESP = 0.851 &deg;Brix).