AUTOMATIC CLASSIFICATION OF APPLE LEAF DISEASES BASED ON TRANSFER LEARNING

Mingming Lai∗,∗∗ and Lutao Gao∗,∗∗

Keywords

Transfer learning, apple leaf disease, VGG-16, SVM, hybrid transfer learning model, sparse automatic encoder

Abstract

To explore which type of transfer learning approaches and which layer of neural network to transfer from is more suitable for the classification of apple leaf diseases; we studied the effectiveness of the transfer learning, the appropriate transfer networks with the best feature extraction layers. The proposed model took Visual Geometry Group 16 (VGG-16) as the research object. Based on two common transfer strategies, we proposed three transfer methods and the compared performance showed that the hybrid transfer models have a good generalized ability on the apple leaf disease classification data set. Moreover, the support vector machine models with 5-fold cross-validation were adopted to the optimal network extraction layer of VGG-16 which was determined to be fc1. The average accuracy of the proposed model for apple leaf disease was 98.6%, which was 1.31% higher than that without using the best extraction layer. The model has a high accuracy that can provide certain suggestion for plant disease classification.

Important Links:

Go Back