Potato Leaf Disease Detection: Evaluating the Efficacy of Convolutional Neural Network
DOI:
https://doi.org/10.7492/apc4kz76Abstract
To achieve healthy crop yields and guarantee food security in agriculture, prompt and precise plant disease diagnosis is crucial. However, due to the variety of leaf looks and the complexity of symptoms, it might be difficult to identify illnesses in potato leaves. This calls for the creation of a practical and efficient technique that can get past these obstacles and raise the precision of illness diagnosis. This research offers a thorough investigation of potato leaf disease detection utilizing a multi-architecture Convolutional Neural Networks (CNNs) technique, using the capabilities of computer vision and deep learning.
The classification capabilities of five distinct CNN architectures—VGG16, VGG19, MobileNetV2, ResNet50, and AlexNet—are evaluated. Dataset acquisition, data augmentation, model selection, hyperparameter tuning, and assessment were all included in the study, which resulted in a thorough examination of training efficiency, model convergence, and detection accuracy. With an astounding 97% testing accuracy and 98% specificity, ResNet50 was the best performer, according to our data. On the other hand, the least successful architecture was VGG19. Classifying healthy leaf types appropriately was a persistent problem for all models, suggesting a possible area for model improvement. The effectiveness of deep learning in diagnosing plant health is demonstrated in this work, along with the significance of specificity as a crucial measure in these kinds of jobs. The findings of this study offer a possible path for in-field real-time identification of potato leaf diseases.