Hybrid Deep Learning Framework for Grape Plant Disease Identification and Meta-Agnostic Visualization

In India, grapes play a significant role in the agricultural sector. Grapes and their derivatives are a key component of India’s exports. The susceptibility of grape leaves to various diseases poses a threat to large-scale grape production. The objective of this study is to identify and detect leaf diseases using deep learning techniques to improve the detection rate before diseases spread widely. The research employs a pre-trained deep Convolutional Neural Network (CNN), specifically EfficientNet-B0, which is a light weighted one, for transfer learning. A total of 9027 images of grapevine leaves are classified into four categories: Isariopsis Leaf Spot, Black Rot, Black Measles, and Healthy. The main focus was on distinguishing between healthy leaves and infected ones, followed by identifying the specific disease type. Through training these images on the transfer learning models, an impressive image classification accuracy rate of 99.74% was achieved with EfficientNet-B0.In order to train and validate our model, we make use of the extensive plant village dataset, which is considered the largest standard dataset available. To assess the effectiveness of our proposed system, we calculate deep learning metrics along with ROC and AUC. The outcomes clearly indicate that our system attains an exceptional maximum validation accuracy and an AUC of 1, establishing it as the most optimal and top-performing approach when compared to existing state-of-the-art deep learning techniques for automated identification of plant diseases.