Emerging Developments in Real-Time Edge AIoT for Agricultural Image Classification

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 au- tonomous 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 prac- tical 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 accu- racy 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.

@Article{PCM25a,
  author       = {Pintus, M. and Colucci, F. and Maggio, F.},
  title        = {Emerging Developments in Real-Time Edge AIoT for Agricultural Image Classification},
  journal      = {IoT},
  number       = {1},
  volume       = {6},
  month        = {february},
  year         = {2025},
  publisher    = {MDPI},
  keywords     = {precision agriculture, crop continuos monitoring, image classification, edge-based AIoT, real-time deep learning, hardware accelerators, synthetic datasets, wireless communication, KANs, XNets},
  doi          = {10.3390/iot6010013},
  url          = {https://publications.crs4.it/pubdocs/2025/PCM25a},
}