Multi-label Classification of Plant Diseases Using the Binary Relevance Approach: An Application for Tomato
- Innovations and Interdisciplinary Solutions for Underserved Areas 8th International Conference, InterSol 2025, Ouagadougou, Burkina Faso, July 3–4, 2025, Proceedings : 18-31
Résumé
Burkinabe agriculture is a fundamental pillar of the national economy. It plays a crucial role in the economic, social, and food security sectors, contributing to the stability and development of the country. However, this sector faces several challenges, particularly plant diseases, which lead to yield losses, inflated production costs, and food insecurity. As part of the agropastoral offensive aimed at improving agricultural production to achieve food self-sufficiency, combating these plant diseases is of paramount importance. Today, artificial intelligence (AI) offers significant advancements in the fight against these agricultural problems. In this article, we propose an approach using machine learning, focusing on multi-label classification for the simultaneous detection of multiple diseases on a plant. We used the EfficientNetB3 [1] neural network as a feature extractor for plant images and applied a technique based on the Binary Relevance approach with different base algorithms including Multi-Layer Perceptron (MLP), Support Vector Machine (SVM) Random Forest, Perceptron and Decision Tree. Our experiments showed that the MLP algorithm provided satisfactory performance, with an F1-Score of 86.06%, a Hamming Loss of 5.34%, and an accuracy of 80.02%. This approach represents an important step towards the rapid and accurate diagnosis of plant diseases in Burkina Faso.
Mots-clés
Agriculture, Agronomy, Pattern Recognition Receptors in Plants, Plant Pathology, Plant Biotechnology, Pomology