Optimizing Grape Leaf Disease Identification Through Transfer Learning and Hyperparameter Tuning

Abstract

Grapes are a globally cultivated fruit with significant economic and nutritional value, but they are susceptible to diseases that can harm crop quality and yield. Identifying grape leaf diseases accurately and promptly is vital for effective disease management and sustainable viticulture. To address this challenge, we employ a transfer learning approach, utilizing well-established pre-trained models such as ResNet50V2, ResNet152V2, MobileNetV2, Xception, and In-ceptionV3, renowned for their exceptional performance across various tasks. Our primary objective is to identify the most suitable network architecture for the classification of grape leaf diseases. This is achieved through a rigorous evaluation process that considers key metrics such as accuracy, F1 score, precision, recall, and loss. By systematically assessing these models, we aim to select the one that demonstrates the best performance on our dataset. Following model selection, we proceed to the crucial phase of fine-tuning the model’s hyperparameters. This fine-tuning process is essential to enhance the model’s predictive capabilities and overall effectiveness in disease identification. To accomplish this, we conduct an extensive hyperparameter search using the Hyperband strategy. Hyperparameters play a pivotal role in shaping the behavior and performance of deep learning models, and by systematically exploring a wide range of hyperparameter combinations, our goal is to identify the most optimal configuration that maximizes the model’s performance on the given dataset. Additionally, the study’s results were compared with those of numerous relevant studies.