A Comparative Study and Optimization of Deep Learning Models for Grape Leaf Disease Identification

Abstract

Background: In the domain of grape farming, the enduring difficulties of preventing germs and viruses pose significant risks to economic viability. Modern advances in artificial intelligence, machine learning, deep learning (AI, ML, and DL), and computer vision have produced efficient methods for identifying and classifying grape viral infections. Methods: The development and improvement of deep learning algorithms designed particularly for identifying and categorizing grape leaf infections is the focus of this study. Utilizing deep intelligence techniques, five important predetermined Deep learning algorithms were used: DenseNet121, VGG19, VGG16, InceptionV3 and ResNet50V2. Result: Comparing the training accuracy, validation accuracy, training loss and validation loss of these five deep learning models, Densenet121 model has shown best performance. Densenet121 model achieved a recall and accuracy score of 99.86%. These findings demonstrate the immense scope of our method for actual use in the production of grape leaf, offering a cheaper and more feasible method for preventing disease and minimizing monetary harm.