Pomegranate Disease Classification Using Advanced Deep Learning Models

Early detection and classification of pomegranate diseases are crucial for maximizing crop yield and quality. This research study attempts to address the challenge of accurately identifying and classifying pomegranate diseases by leveraging advanced machine learning models. For this task, we selected three vision models with different architecture, complexity, and disk size requirements. The selected models where: DaViT-Base, a vision transformer model; EfficientNetV2M, a convolutional neural network; and MobileOne-S4, a lightweight model optimized for mobile devices. The methodology used involved training these models on a dataset of annotated pomegranate images, which was split into four disease categories and one healthy category. This was followed by a thorough evaluation of each model to determine the accuracy and potential deployment on mobile devices or drones. All three models showed exceptional results: DaViT-Base - 99.28%, EfficientNetV2M - 99.54%, MobileOne-S4- <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 9. 1 5 \%}$</tex>. The results show that lightweight models such as the MobileOne-S4 are viable options for the task of pomegranate disease classification, and implementing similar models has the potential to significantly enhance agricultural disease management. Future work will focus on exploring realtime classification.