Plant leaf disease detection and classification using modified transfer learning models
Meenakshi Srivastava, Jasraj Meena
Abstract
Agriculture is a dominating field that plays an essential role in the economic development of any country. In India, agriculture contributes about 17% of the total GDP. But, decreasing land under agriculture has become a prominent problem harming both, the economy and the interest of farmers. Alongside, every year farmers face many difficulties; one such is leaf disease problems. The farmers use unscientific approaches to identify leaf disease, which is a time-consuming process. This can be tackled by using plant disease detection and classification system based on ML and DL. Many researchers have used ML techniques to detect plant diseases, but most of them do not solve overfitting and testing problem. This paper initially proposes plant leaf disease detector and classifier framework using DL. Then these five Deep Convolutional Neural Network models (Vgg16, MobileNetV2, Xception, InceptionV3, and DenseNet121) are used to detect and classify plant diseases. Two datasets have been considered; First from Mendeley (plant leaf dataset), having 4590 leaf images split into twenty-two classes. Second from PlantVillage (Cherry Dataset), having 2052 leaf images split into two classes. Further, apply pre-processing techniques such as data augmentation, resizing, and rescaling to remove the problem of overfitting. Then testing and training of the proposed models have been on leaf images. To evaluate the quality and quantity of proposed models, quality matrix has been used, and the result shows that for dataset1, MobileNetV2 outperforms other existing models with 98.9% accuracy. For the Cherry Dataset, DenseNet121 outperforms other existing models with 99.9% accuracy.