<![CDATA[This study focuses on the implementation of the YOLOv4-Tiny algorithm on Raspberry Pi 5 for detecting plastic bottle waste in aquatic environments. The primary goal is to optimize the frame per second (FPS) while maintaining detection accuracy. A dataset consisting of 914 images was augmented using RoboFlow to enhance the robustness of the model under real-world conditions. Experiments were conducted in a controlled pool environment with an input resolution of 320x320 pixels. Results demonstrated an average FPS of 7-8, with detection accuracy ranging between 67% and 80%. Further evaluation reported a total loss of 0.3, mean Average Precision (mAP) of 97.94%, precision of 93%, recall of 96%, F1 score of 0.95, and an average Intersection over Union (IoU) of 76.47%, indicating effective bounding[1] box prediction capabilities. These results highlight the potential of YOLOv4-Tiny as a lightweight and real-time detection solution, particularly for low-computational devices such as Raspberry Pi. The findings provide a solid foundation for developing efficient plastic waste detection systems, which can be deployed across various aquatic locations, supporting environmental monitoring and waste management initiatives.]]>
Copyrights © 2025
