Abstract:

CWT - GRNN model was constructed to predict sugar content in cherry fruit by combining continuous wavelet transform (CWT) with generalized regression neural networks (GRNN). CWT was used to extract the key features, which were related to the sugar content of cherry. Three scales in the CWT domain were selected to efficiently extract the features of cherry fruit, and four feature spaces were divided according to the features of visible-near infrared (VIS-NIR) diffuse reflectance spectroscopy. Thus a feature vector, which contains twelve parameters, was input to the GRNN and the smooth factor of the GRNN was set as 0.0001. Twenty cherry samples were used to verify the performance of the CWT - GRNN model. Experimental results showed that the relative error of predicted samples was below 2%. CWT - GRNN model could be used to quickly, accurately and non-destructively predict the sugar content in cherry fruit. Also, the proposed method could be applied in control and  evaluation in fruit and vegetable industry.

Key words: visible-near infrared diffuse reflectance spectroscopy, internal quality index, nondestructive measurement, sugar content, cherry