Mobile Application for Plant Disease Detection Based On Combination Of Deep Learning And Image Segmentation

Abstract

It's important for sustainable agricultural practice to accurately detect plant disease in order to facilitate early intervention to limit crop loss. This project describes a deep learning–based pipeline of image segmentation and classification in order to improve the reliability of plant disease detection. The segmentation process sought to separate the plant from its background to limit noise and allow the classification model to focus on features that were relevant. We experimented with many segmentation models and settled on U-Net with EfficientNet and Inception backbones because U-Net produced clean and accurate leaf masks. We evaluated many pretrained CNN architecture methods in the classification stage, and identified MobileNetV2 as a model that embodied the balance between speed, lightweight, and an impressive level of accuracy with the segmented leaf images. We will demonstrate that when deep learning segmentation and classification is combined with classification models, the performance of disease identification improves significantly. This project indicates the use of deep learning for agriculture applications works and promotes the development of smart, automated plant disease monitoring systems.