Plastic water bottle detection model using computer vision in aquatic environments

Area of Science:

• Environmental science
• Computer science
• Machine learning

Background:

• Measuring watershed macrotrash contamination is challenging due to manual, labor-intensive methods.
• Accurate quantification of plastic pollution in aquatic environments is crucial for effective management.

Purpose of the Study:

• To develop an automated system for counting plastic bottles in rivers and streams.
• To leverage computer vision and deep learning for efficient waste tracking.

Main Methods:

• Utilized YOLOv8 object detection model trained on diverse trash and plastic bottle image datasets.
• Integrated Norfair object tracking library for continuous monitoring.
• Developed a novel post-processing algorithm to minimize false positives.

Main Results:

• Achieved high performance in detecting and tracking plastic bottles.
• Demonstrated exceptional accuracy with only one false positive in test scenarios.
• Obtained a recall rate exceeding 0.947 for plastic bottle detection.

Conclusions:

• The automated approach offers a highly accurate and efficient solution for watershed macrotrash monitoring.
• This technology can significantly reduce the labor and time required for pollution assessment.
• The developed model provides a reliable tool for tracking plastic debris in aquatic ecosystems.