A novel deep learning-based floating garbage detection approach and its effectiveness evaluation in environmentally...

Area of Science:

• Environmental Science and Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Floating garbage poses a significant threat to water ecosystems and sustainability.
• Complex environmental factors like water flow and wind complicate garbage detection and collection.
• Automated detection and collection systems are crucial for effective water pollution management.

Purpose of the Study:

• To develop an efficient and economical deep learning solution for detecting aggregated floating garbage on water surfaces.
• To improve the accuracy and efficiency of floating garbage detection systems.
• To create a lightweight model suitable for outdoor environmental monitoring applications.

Main Methods:

• Utilized a deep learning approach based on YOLOv8 (You Only Look Once v8).
• Enhanced the model by improving the backbone, incorporating Wise-Powerful IoU loss, and adding an AuxHead detection head.
• Evaluated the model's performance in terms of detection accuracy (mAP, P) and efficiency (model size, parameters).

Main Results:

• Achieved a mean Average Precision (mAP) of 89.4% and Precision (P) of 95.8% for detecting aggregated floating garbage.
• Reduced model size to 18.8 MB and decreased parameters by 32.2% compared to the original model.
• Demonstrated a 61.5% improvement in collection rate compared to Faster R-CNN, with significant energy savings and pollution reduction.

Conclusions:

• The proposed lightweight YOLOv8-based model effectively addresses the challenge of detecting aggregated floating garbage in complex environments.
• The model's superior detection performance and efficiency contribute to improved water ecosystem protection and restoration.
• The solution offers substantial environmental benefits, including reduced water pollution, energy conservation, and carbon emission reduction.