Apple classification based on multi-information fusion and DA-DBN

Abstract

Objective: In order to improve the precision of apple grade judgment model, the method of apple grade judgment was established. Methods: A decision model of apple rank based on multi-information fusion and dragonfly algorithm was proposed. Firstly, the HSV color feature, LBP texture feature and HOG shape feature of apple image were extracted by pre-processing such as data enhancement, normalization, Gauss filter and grayscale. Secondly, the performance of DBN model was affected by the selection of parameters, the network parameters of DBN model were optimized by DA algorithm, and a multi-information fusion and DA-DBN model for determining apple rank wws proposed. Results: Compared with GA-DBN, PSO-DBN, GWO-DBN and DBN, the model based on DA-DBN had the highest precision. Conclusion: The DBN model is optimized by dragonfly algorithm which can effectively improve the accuracy of apple rank determination model, which provides a new method for apple rank determination.

Keywords

• deep belief network
• dragonfly algorithm
• texture feature
• color feature
• shape feature
• apple