Artificial intelligence methods for morphological analysis and diagnosis of bacterial and fungal leaf diseases

Bacterial and fungal leaf diseases frequently produce overlapping visual symptoms, complicating automated diagnosis and limiting the transferability of models across crops and field conditions. This review systematically evaluates artificial intelligence methods for the morphology-based detection and differentiation of these diseases. Following a PRISMA-based selection process, 95 records published between 2017 and 2026 were identified, of which 59 eligible studies were included. The survey covers the morphological characteristics that distinguish bacterial and fungal diseases, traditional image processing and machine learning approaches, deep learning and generative AI techniques, and publicly available datasets. It also critically compares the performance, advantages, and limitations of existing diagnostic approaches. The findings indicate that deep learning dominates the literature, particularly in fungal disease detection, whereas research on bacterial diseases and generative AI applications remains comparatively limited. Although convolutional and transfer-learning models generally demonstrated strong classification performance, comparisons across studies were constrained by differences in datasets, evaluation protocols, and performance metrics. The principal barriers to practical adoption include morphological overlap between infections, class imbalance, and scarcity of expert-annotated images of bacterial and early-stage diseases, reliance on controlled-background datasets, limited external validation, model complexity, and insufficient interpretability. Future research should prioritize pathogen-balanced multicrop field datasets, standardized reporting and cross-dataset evaluation, multimodal fusion of RGB images with spectral and environmental data, domain adaptation, validated generative augmentation, explainable models, and lightweight edge deployment. By connecting lesion morphology with methodological and dataset evidence, this review provides a structured roadmap for developing robust, interpretable, and field-ready plant disease diagnosis systems.