Ensemble of the Deep Convolutional Network for Multiclass of Plant Disease Classification Using Leaf Images

Abstract

Plant diseases are a major threat to agricultural production. Reduced yield due to plant diseases can lead to immeasurable economic losses. Therefore, the detection and classification of crop diseases are of great significance. Current existing classification methods based on the single convolutional neural network (CNN) are not satisfactory for plant disease classification performance in a large number of classes. In this case, a CNN-based approach named multiclass plant EnsembleNet (MCPE) is proposed to address these problems. MCFN firstly adopts a data augmentation strategy-based AutoAugment enhance dataset. Next, an EnsembleNet including four CNNs is employed to classify plant species, with a new activation function, concatenated dynamic ReLU, which has better performance than conventional ReLU in the multiclass plant disease dataset. Then, training a diseases classifier for each plant, which is used to identify the types and severity of plant diseases. Experimental results on 61 plant diseases from 10 different plant species, over 40 000 images, show that MCFN outperforms the state-of-the-art methods in multiclass plant disease recognition and achieves a good identification accuracy of 97.5%. We believe that the method described in this paper can further improve the identification efficiency of plant diseases, thus providing a basis for the identification of other plant leaf diseases.