Severity Classification of Brown spot Disease in Passion Fruits Using Multi-Task Deep Learning

Abstract

Experts in plant pathology especially in passion fruit plants are few across the country. From this scenario arises the need for a tool that will enable the farmers to carry out brownspot disease diagnosis and crop health monitoring for the farmers. This report presents the development of a multi-task deep learning model which can be used to detect brown spot disease and also perform severity classification of the disease. Labeled data was provided for use in the project. The dataset contains leaves infected with brown spot disease with there corresponding severity levels, that is, Level 1, Level 2, Level 3 and Level 4. An object detection model and classification model were trained on the dataset provided. We proposed and developed a multitask deep learning model. The model was deployed on a smart phone application.