SmartAgro-ViT: A Self-Supervised Explainable AI Framework for Plant Disease Analysis
TL;DR
SmartAgro-ViT demonstrated superior classification accuracy and robustness compared to supervised transformer and convolutional neural network models and is useful for precision agriculture and smart farming systems since they are both effective and easy to understand.
Abstract
In order to avoid a global food shortage and maximise agricultural production, rapid and accurate detection of plant diseases is essential. Although image-based plant disease recognition has been enhanced by deep learning, the majority of these methods rely on massive annotated datasets and employ black-box models, rendering them ill-suited for usage in agricultural contexts. The Future of SmartAgro-ViT An AI system that uses transformers to analyse plant diseases can be self-supervised and explained. To lessen the need for human annotation, the suggested method employs self-supervised pretraining to learn detailed visual representations of plants from large datasets of unlabelled images. A fine-tuning process is necessary for supervised disease classification across different crop species. An explainability module makes the model more accessible by providing visual attention maps that emphasise areas of plant leaves that are related to diseases. This helps in making accurate predictions. In comprehensive studies conducted on benchmark plant disease datasets, SmartAgro-ViT demonstrated superior classification accuracy and robustness compared to supervised transformer and convolutional neural network models. Various lighting and backgrounds complement the structure. The intelligent plant disease diagnosis capabilities of SmartAgro-ViT are useful for precision agriculture and smart farming systems since they are both effective and easy to understand.