Implementation of a Plant Disease Identification System using the CNN Algorithm and the Web-Based Django Framework

Abstract

This study addresses the need for efficient and accessible plant disease identification systems in the era of Agriculture 4.0, where advances in artificial intelligence (AI) and machine learning (ML) support data-driven agricultural practices. The increasing popularity of home gardening highlights challenges faced by users in identifying plant diseases due to limited knowledge and diagnostic tools. Therefore, this research aims to develop a web plant disease detection system using the Django framework and convolutional neural networks (CNNs). The model was trained on a controlled dataset consisting of 57,320 leaf images collected from the PlantVillage and Turmeric Plant Disease datasets. Image preprocessing was applied, including resizing, normalization, and data augmentation such as image rotation, zooming, image inversion and brightness adjustmen. Class imbalance during training was handled using class weighting. The dataset is divided into a training set and a validation set for model development and evaluation. The CNN model achieved an accuracy of 92% on the labeled validation dataset, with a mean F1 score of 0.79 and a weighted mean F1 score of 0.92. For generalization testing, an uncontrolled (wild) dataset consisting of 223 images collected from online sources was used, resulting in an accuracy of 11%, indicating limited real-world generalization due to domain differences. Despite this limitation, the proposed system demonstrates the feasibility of CNN-based plant disease classification in a web application.

Code and data availability

The paper's CNN plant-disease model was trained primarily on the public Kaggle 'New Plant Diseases Dataset' (PlantVillage-derived, 54,528 images), which is a paper-specific, publicly available image dataset directly used for the study's phenotyping measurements. The Turmeric Plant Disease Dataset (Mendeley DOI 10.17632