Mobile application for plant diseases detection using vision transformer architecture

Abstract

Every year, a significant portion of crops is lost to plant diseases worldwide. Farmers often struggle to accurately identify these diseases manually due to limited knowledge and resources. This project aims to address this issue by developing a mobile application that uses a deep learning model to detect and classify plant diseases from leaf images. Using the PlantVillage dataset, a comparative study evaluated several transfer learning models, including VGG16, DenseNet201, and ResNet50, against the proposed MobileViT-XXS model for vision transformers. The MobileViT-XXS model demonstrated superior performance, achieving a test accuracy of 97%, outperforming the other convolutional neural network models.As a result, a mobile android application was developed using the MobileViT-XXS model for plant disease detection. This application aims to improve crop productivity and the overall economy by enabling farmers to make prompt and precise decisions regarding crop diseases.