• Deep learning models outperform traditional ML in complex image-based crop diagnoses. • Hybrid frameworks and edge computing enable faster, real-time field-level detection. • Dataset diversity is essential for improving model generalization and adaptability. • Persistent challenges include overfitting, data imbalance, and limited rare disease representation. • Future systems require IoT-integrated, multimodal, and scalable AI-driven detection pipelines. The rapid development of machine learning (ML) and deep learning (DL) technologies has brought a significant change in disease detection of crops. This review emphasizes the analysis of different ML and DL models applied on agricultural crops, focusing on detection accuracy, model efficiency, and adaptability across various disease types. A critical examination is conducted on the primary challenges of achieving real-time, scalable, and robust models in dynamic agricultural environments. The key challenges such as data imbalance, lack of rare disease data, technological barriers, model scalability and generalization challenges are explored in depth. Furthermore, the review investigates how integrating advanced technologies such as the Internet of Things (IoT), big data analytics, and edge computing can support real-time deployment and enhance the responsiveness of AI-driven disease detection systems in real-world agricultural settings. While current DL models demonstrate strong potential, limitations in generalization and field-level scalability highlight the need for further innovation. Future directions are proposed for developing holistic, AI-powered frameworks that are efficient, adaptive, and suitable for varied agro ecological conditions. Overall, the review offers a comprehensive synthesis of the current landscape, technological enablers, and persistent gaps in the field of AI-based crop disease management.
Saini et al. (Sun,) studied this question.