PLANT DISEASE DIAGNOSIS USING XCEPTION NET AND VISION TRANSFORMER

Abstract

Food security and economic stability can be impacted by plant diseases, which can result in large yield losses in agricultural production. Automated diagnosis of plant diseases can assist farmers and plant specialists in early disease detection, outbreak prevention, and treatment optimization. In this research, we investigate how to diagnose plant diseases using the Plant Village dataset with XceptionNet and Vision Transformer. We use XceptionNet for classification and Vision Transformer for featurization in a single pipeline, which allows us to leverage the complementary strengths of both models. While Vision Transformer is a transformer-based model that can capture long-range relationships and contextual data, XceptionNet is a deep convolutional neural network that can learn discriminative features from images. We prepare and divide the dataset into training, test, and validation sets before training our pipeline on the labelled photos. We test several hyper parameters and evaluate how well our pipeline performs using multiple metrics, including accuracy, precision, recall, loss, auc, and F1 score. Our findings demonstrate that our pipeline has a 99.2% accuracy rate.