Research Information System
Applied Soft Computing, vol.186, 2026 (SCI-Expanded, Scopus)
The global rise in food demand results from a complex interplay of factors, including population growth and evolving dietary preferences, which impact agricultural practices and supply chains. While technological advancements such as precision farming improve productivity, they also raise sustainability concerns. Agriculture faces significant challenges from external factors like weather disruptions, pests, and diseases, leading to considerable economic losses. Addressing these challenges requires precise plant diagnostics, achievable through artificial intelligence (AI) within the framework of precision agriculture. The presented study proposes a modified optimizer modeled after the particle swarm optimization (PSO) algorithm, tailored to optimize classification models for plant disease detection in Agriculture 4.0 using TinyML Internet of Things devices. The introduced two-layer approach leverages convolutional neural networks (CNNs) for extracting the features and boosting models (AdaBoost and CatBoost) for disease detection and identification from leaf images. Simulations on publicly available datasets demonstrate promising results. Additionally, an advanced interpretation method using Shapley additive explanations (SHAP) is proposed to interpret hybrid CNN-booster models, providing insights into the leaf features that influence classification decisions. Key contributions include enhanced accuracy and efficiency through the modified optimizer, a robust AI-based system for precise plant disease detection, a two-layer hybrid model framework, and an approach to feature impact interpretation in advanced hybrid models.