Waste Detection on Mobile Devices: Model Performance and Efficiency Comparison
Year: 2025
Type: article
Abstract: This study evaluates object detection models for mobile deployment by comparing YOLOv11 and EfficientDet-Lite using a waste classification dataset. EfficientDet-Lite0 demonstrated higher speed (13 FPS), YOLOv11n was the most power-efficient (125,000 μAh in 590 seconds), and YOLOv11m achieved the highest accuracy (mAP@50: 0.694). The deployment of these models on an Android application highlights their trade-offs: EfficientDet-Lite0 suits speed-critical tasks, YOLOv11n excels in power-sensitive scenarios, and YOLOv11m and YOLOv11s perform best in accuracy-driven applications. These findings inform the selection of optimal models for efficient and accurate waste sorting in mobile and edge computing environments.
Author: Urankar, Eva
Institution: University of Ljubljana
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7082082736
FWCI:
Cites:
Cited by:
Related to: