<![CDATA[Accurate waste object detection is important for enabling efficient automated recycling and environmental management. While lightweight YOLO models are often managed in Python, integration and evaluation of the models in MATLAB remains a technical challenge due to limited support and documentation. This study intends to fill that gap by evaluating the performance of YOLOv5s, YOLOv7-Tiny, and YOLOv8n within the MATLAB environment for waste object detection using the TrashNet dataset. A semi-automatic labeling approach was employed, combining manual annotation with pseudo-labeling using a pretrained YOLOv8n model. The models were trained and exported to the ONNX format for MATLAB-based inference and analysis. Experimental results show that YOLOv8n achieved the highest mAP@0.5 of 0.954, while YOLOv5s demonstrated the most stable inference performance in MATLAB, consistently producing confidence scores above 90% and real-time speeds of up to 15.9 fps. In contrast, YOLOv7-Tiny achieved the fastest inference speed (up to 24.4 fps) but exhibited reduced classification consistency. Notably, YOLOv8n experienced confidence score degradation during MATLAB inference, suggesting post-processing discrepancies between native Python and ONNX-imported workflows. This research highlights MATLAB’s capability to serve as a functional evaluation platform for modern lightweight detectors and emphasizes its potential for expanding accessible AI applications in waste management systems.]]>
Copyrights © 2026
