A Study on Hyperspectral Apple Bruise Area Prediction Based on Spectral Imaging

Achieving fast and accurate prediction of the fruit mechanical damage area is important to improve accuracy efficiency apple quality grading. In this paper, spectral data all samples in wavelength range from 376 1011 nm were collected, sample set was divided by physicochemical coeval distance method, preprocessing methods evaluated establishing a full-wavelength artificial neural network model. The selection performed competitive adaptive reweighted sampling, L1 parameter Pearson correlation coefficient partial least squares, network, support vector machine (Gaussian kernel) models established predict bruise size. surface fitting using actual area, regression equation time height established. results showed that (1) method first-order difference + SG smoothing can make model more accurate; (2) CARS-ANN has better performance higher operation efficiency, with root mean square error R-value 0.1150 0.8675, respectively; (3) sparrow search algorithm used optimize model, which improved reached 0.0743 0.9739. (4) relationship between information, time, degree obtained combining establishment fitted This study application extension value for rapid nondestructive area.