Water surface garbage detection based on lightweight YOLOv5

Area of Science:

• Environmental Science
• Computer Science
• Robotics

Background:

• Increasing plastic waste in rivers due to economic growth since the 1980s poses environmental challenges.
• Manual garbage retrieval is inefficient and high-risk due to large volumes and operational dangers.
• Deep learning, particularly the YOLO algorithm, shows promise for object detection in environmental monitoring.

Purpose of the Study:

• To develop an efficient and lightweight deep learning algorithm for detecting surface garbage in rivers.
• To adapt the YOLOv5 algorithm for deployment on unmanned ships for automated garbage salvage operations.

Main Methods:

• Implementation of a lightweight version of the YOLOv5 algorithm tailored for water surface garbage detection.
• Validation of the algorithm using the Orca dataset to assess performance metrics.
• Focus on optimizing the algorithm for reduced parameter quantity and enhanced detection speed.

Main Results:

• The improved YOLOv5 algorithm demonstrated a 4.3% increase in detection speed.
• Achieved a mean Average Precision (mAP) of 84.9% and a precision of 88.7%.
• Reduced parameter quantity to only 12% of the original data, indicating significant lightweighting.

Conclusions:

• The lightweight YOLOv5 algorithm offers superior accuracy and speed for water surface garbage detection compared to the original.
• The algorithm's reduced size makes it suitable for deployment on resource-constrained hardware, such as unmanned ships.
• This advancement facilitates more efficient and widespread automated solutions for riverine plastic pollution management.