2 research outputs found
Machine Learning Approaches for Detecting Vine Diseases: A Comparative Analysis
No full textThis study investigates the classification of vine leaf diseases using convolutional neural networks (CNNs), focusing on three major diseases: powdery mildew, caused by fungus Uncinula necator, Red Blotches associated with pathogens such as Phomopsis viticola, Grapevine Leafroll Disease and leafroll associated Grape -linked virus (GLRaV). Accurate diagnosis of these high-risk diseases is critical to vine health and yields. We evaluated the performance of three CNN algorithms—MobileNetV2, ResNet50, and VGG16—by comparing their training and validation accuracies, as well as loss over ten seasons. MobileNetV2 emerged as the most robust model, exhibiting high accuracy and low loss, indicating strong generalizability. ResNet50 showed a steady increase in accuracy, but with high variability, indicating that probabilities with complex models or extended training requirements VGG16 showed notable improvements in training accuracy but encountered difficulties itinvolves consistency during validation, which means overfitting. Although MobileNetV2 proved to be the most efficient for this task, our analysis suggests that replicating ResNet50 and VGG16 can improve their performance. Future research will explore longer training times, larger datasets, and other methods to further improve the generalizability and robustness of this model This work highlights the ability of CNN to detect vine leaves emphasize early diseases andprovide a strategy for sustainable viticultural practices
Machine Learning Approaches for Detecting Vine Diseases: A Comparative Analysis
No full textThis study investigates the classification of vine leaf diseases using convolutional neural networks (CNNs), focusing on three major diseases: powdery mildew, caused by fungus Uncinula necator, Red Blotches associated with pathogens such as Phomopsis viticola, Grapevine Leafroll Disease and leafroll associated Grape -linked virus (GLRaV). Accurate diagnosis of these high-risk diseases is critical to vine health and yields. We evaluated the performance of three CNN algorithms—MobileNetV2, ResNet50, and VGG16 —by comparing their training and validation accuracies, as well as loss over ten seasons. MobileNetV2 emerged as the most robust model, exhibiting high accuracy and low loss, indicating strong generalizability. ResNet50 showed a steady increase in accuracy, but with high variability, indicating that probabilities with complex models or extended training requirements VGG16 showed notable improvements in training accuracy but encountered difficulties it involves consistency during validation, which means overfitting. Although MobileNetV2 proved to be the most efficient for this task, our analysis suggests that replicating ResNet50 and VGG16 can improve their performance. Future research will explore longer training times, larger data sets, and other methods to further improve the generalizability and robustness of this model This work highlights the ability of CNN to detect vine leaves emphasize early diseases and provide a strategy for sustainable viticultural practices
