Sousa et al. (2025) Interoperable IoT/WSN Sensing Station with Edge AI-Enabled Multi-Sensor Integration for Precision Agriculture

Identification

• Journal: Agriculture
• Year: 2025
• Date: 2025-12-28
• Authors: Matilde Lopes Sousa, Ana Marques Alves, Rodrigo Antunes, Martim Aguiar, Pedro Gaspar, Nuno Pereira
• DOI: 10.3390/agriculture16010069

Research Groups

• C-MAST—Centre for Mechanical and Aerospace Science and Technologies, Covilhã, Portugal
• Department of Electromechanical Engineering, University of Beira Interior, Covilhã, Portugal

Short Summary

This study develops and evaluates LITecS, a modular, solar-powered IoT/WSN sensing station with edge AI for precision agriculture and biodiversity monitoring, demonstrating its adaptability and sustained operation in two distinct field deployments.

Objective

• To determine if a single modular station with flexible hardware architecture can simultaneously support precision agriculture (PA) and biodiversity monitoring without redesign.
• To assess if dual-level Edge AI implementation can ensure continuous autonomous operation in agricultural environments.
• To investigate how the energy autonomy of a modular system can be ensured under variable computational loads throughout the year.

Study Configuration

• Spatial Scale: Field deployments in agricultural (vineyard) and wild-flora environments (Serra da Gardunha mountain slopes) in Portugal, utilizing three stations per deployment.
• Temporal Scale: Approximately one year (365 days) of continuous monitoring for each field deployment.

Methodology and Data

• Models used:
  • K-Means clustering (for vegetation greenness level analysis)
  • BatDetect2 CNN-based model (for bat call detection and species classification)
  • YOLOv8 model (for initial plant detection)
  • EfficientNetB5 model (for precise species and phenological state classification)
• Data sources:
  • Sensors: Thermometer, pluviometer, anemometer, wind direction sensor, barometer, hygrometer, soil moisture, leaf wetness, ultrasonic microphones (10–160 kHz).
  • Imaging: Raspberry Pi Camera Module 3 Wide (120° FoV) and Raspberry Pi Camera Module 3 (76° FoV), capturing images at 4608 × 2592 pixels resolution.
  • Edge Computing: ESP32 microcontroller (low-power continuous acquisition), Raspberry Pi Zero 2 W single-board computer (higher-power tasks, AI inference).
  • Communication: LTE and/or LoRaWAN for data transmission to cloud.
  • Data Storage & Management: MySQL (version 8.4.x) database, cloud storage, Fast API (version 0.120.0) backend, React-based interactive dashboard.

Main Results

• The LITecS station demonstrated high modularity, successfully adapting its core architecture to two distinct use cases (precision viticulture and wild-flora monitoring) without hardware redesign.
• A two-tier ESP32-Raspberry Pi architecture enabled energy autonomy, with an estimated average daily consumption of 80 Wh/day. A 12 V/30 Ah battery and 30 W solar panel supported seasonal off-grid operation, though occasional winter interruptions occurred during prolonged low irradiance.
• In the BioD’Agro viticulture deployment, automated vegetation "greenness" analysis provided a quantitative green-to-brown/yellow ratio for crop-condition tracking. The acoustic module achieved 0.94 Average Precision (AP) for bat vs. non-bat detection and 0.85 mean Average Precision (mAP) for species classification using BatDetect2.
• In the Montanha Viva wild-flora deployment, three stations operated for 365 days, achieving image-based operational availabilities of 93%, 66%, and 97%, including continued data acquisition during a wildfire event.
• Computer vision pipelines (YOLOv8 and EfficientNetB5) for flora monitoring achieved mAP@50 > 0.96 for well-represented classes (Armeria transmontana, Umbilicus rupestris), with lower values for under-represented classes (Echinospartum ibericum), highlighting the importance of dataset balance.

Contributions

• Introduction of a novel, modular IoT/WSN sensing station architecture capable of supporting diverse agro-ecological monitoring objectives (precision agriculture and biodiversity) without requiring hardware redesign.
• Implementation of a dual-level Edge AI strategy (ESP32 for low-power tasks, Raspberry Pi for high-power AI inference) to ensure continuous autonomous operation and energy efficiency in remote agricultural environments.
• Development of a comprehensive, solar-powered system with robust data processing and visualization pipelines, enabling real-time decision support for farmers and researchers, and demonstrating resilience in challenging field conditions, including wildfire events.

Funding

• Project BioDAgro–Sistema operacional inteligente de informação e suporte á decisão em AgroBiodiversidade (project PD20-00011)
• Project Montanha Viva—An intelligent prediction system for decision support in sustainability (project PD21-00009)
• PROMOVE program funded by Fundação La Caixa and supported by Fundação para a Ciência e a Tecnologia and BPI.
• Fundação para a Ciência e a Tecnologia (FCT), I.P., through project UIDB/0151/2025, Centre for Mechanical and Aerospace Sciences and Technologies (C-MAST). DOI: https://doi.org/10.54499/UID/00151/2025.

Citation

@article{Sousa2025Interoperable,
  author = {Sousa, Matilde Lopes and Alves, Ana Marques and Antunes, Rodrigo and Aguiar, Martim and Gaspar, Pedro and Pereira, Nuno},
  title = {Interoperable IoT/WSN Sensing Station with Edge AI-Enabled Multi-Sensor Integration for Precision Agriculture},
  journal = {Agriculture},
  year = {2025},
  doi = {10.3390/agriculture16010069},
  url = {https://doi.org/10.3390/agriculture16010069}
}