Leaf Disease Identification Using Model Hybrid Based on Convolutional Neuronal Networks and K-Means Algorithms

Abstract

Plant leaf diseases usually affect agriculture a lot, which is one of the important sources of income for people, so diseases must be detected and recognized quickly and effectively. The research aims to identify these diseases automatically using a model based on deep learning known as convolutional neural networks and the K-means algorithm. The methodology applied for the detection, three previously trained networks, VGG16, VGG19, and ResNet50, were used for the extraction of characteristics, the principal component analysis algorithm was also used to reduce dimensionality, and finally, the K-means algorithm classification. The training of the models was carried out with the use of a Kaggle open database of 7771 images which contain 38 types of diseases and healthy leaves. VGG16, VGG19, and ResNet50 were trained where the accuracy of 97.43%, 98.35%, and 98.38% was obtained. The precision obtained with the VGG16 hybrid model and the K-means algorithm was 96.26%. Therefore, the hybrid model is effective for the identification of plant diseases.