Key points are not available for this paper at this time.
Advances in deep learning (DL) models and next-generation edge devices enable real-time image classification, driving a transition from the traditional, purely cloud-centric IoT approach to edge-based AIoT, with cloud resources reserved for long-term data storage and in-depth analysis. This innovation is transformative for agriculture, enabling autonomous monitoring, localized decision making, early emergency detection, and precise chemical application, thereby reducing costs and minimizing environmental and health impacts. The workflow of an edge-based AIoT system for agricultural monitoring involves two main steps: optimal training and tuning of DL models through extensive experiments on high-performance AI-specialized computers, followed by effective customization for deployment on advanced edge devices. This review highlights key challenges in practical applications, including: (i) the limited availability of agricultural data, particularly due to seasonality, addressed through public datasets and synthetic image generation; (ii) the selection of state-of-the-art computer vision algorithms that balance high accuracy with compatibility for resource-constrained devices; (iii) the deployment of models through algorithm optimization and integration of next-generation hardware accelerators for DL inference; and (iv) recent advancements in AI models for image classification that, while not yet fully deployable, offer promising near-term improvements in performance and functionality.
Pintus et al. (Mon,) studied this question.