ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

Benchmarking High-Speed and Low-Power AI Models on Low-Cost Edge Devices for Agriculture Applications

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600566.(doi:10.13031/aim.202600566)
Authors:   Mohamed Abdallah Salem, Ahmed Harb Rabia
Keywords:   Edge Computing, Deep leaning, Machine Learning, Crop Monitoring, Raspberry Pi, Inference Acceleration, AI Accelerator.

Abstract. Edge artificial intelligence (AI) has emerged as a practical solution for real-time decision-making in precision agriculture; however, deploying deep learning models on low-cost, resource-constrained hardware remains a key challenge. This study presents a systematic evaluation of edge computing platforms for real-time plant disease detection, focusing on the deployment of YOLO-based object detection models on Raspberry Pi 5 with and without hardware acceleration. Using the PlantDoc dataset, different YOLO model variants were trained, exported into multiple deployment formats (including ONNX, MNN, NCNN, and TorchScript), and optimized for edge execution. Performance was evaluated in terms of inference latency, power consumption, CPU and memory utilization, and detection accuracy. Results indicate that lightweight formats such as MNN and NCNN achieve substantially lower inference times and system load compared to full-framework formats when deployed on the Raspberry Pi 5 without acceleration. The integration of the Hailo-8L AI accelerator further reduced inference latency to the microsecond level while maintaining power consumption below 10 W, enabling true real-time operation. Comparative analysis revealed that YOLOv8 achieved higher detection accuracy, whereas YOLOv5 offered improved efficiency and lower computational demand, highlighting application-dependent trade-offs. Overall, the findings demonstrate that hardware–software co-optimization enables scalable edge AI deployment for precision agriculture, supporting real-time plant disease monitoring and autonomous agricultural systems.

(Download PDF)    (Export to EndNotes)