A comprehensive review of machine learning and deep learning approaches for fungal classification in plants

Abstract

Plant diseases are harmful and common nowadays. These abnormal conditions caused by fungi, bacteria, viruses, nematodes or environmental stress like nutrient deficiencies and drought can damage plant health and reduce productivity. They may cause crop yield losses, threaten food security and adversely impact agricultural economies. This review provides a comprehensive examination of recent machine learning and deep learning technologies in fungal classification for plant disease diagnosis. Instead of focusing on individual models, the survey highlights broader advancement trends across macroscopic field imaging, microscopic analysis, hybrid approaches and attention-enhanced architectures. The analysis reveals that the deep learning algorithms consistently outperform traditional techniques. Particularly when deep learning algorithms are supported by transfer learning and attention mechanisms which improve feature extraction for complex fungal patterns. However, challenges such as limited annotated datasets, high inter-class similarity and variations in environmental imaging conditions continue to hinder model generalisation. Studies addressing these limitations and demonstrating the combination of hybrid models, improved optimisation strategies and domain-specific datasets can significantly enhance classification accuracy. Overall, this review synthesises current progress, identifying persistent gaps and outlining future directions for building more robust, scalable and real-world-ready fungal disease detection systems to support sustainable agricultural management.