An annotated Dataset and Benchmark for Detecting Floating Debris in Inland Waters

Area of Science:

• Environmental Science
• Computer Vision
• Artificial Intelligence

Background:

• Marine litter poses a significant threat to aquatic ecosystems.
• Effective removal of floating waste from inland waters can prevent marine pollution.
• Accurate object detection is crucial for efficient removal of water surface debris.

Purpose of the Study:

• To introduce the first comprehensive dataset for water surface floater detection.
• To establish a benchmark for evaluating object detection algorithms in aquatic environments.
• To advance AI applications in water resource management and pollution control.

Main Methods:

• Development of the IWHR_AI_Lable_Floater_V1 dataset using shore-based filming equipment.
• Collection of 3000 annotated images from real-world water scenarios.
• Evaluation of mainstream object detection algorithms, including YOLOv9, on the new dataset.

Main Results:

• The proposed dataset contains 3000 images with detailed annotations for water surface floaters.
• Baseline experiments revealed low detection accuracies for current object detection models, including state-of-the-art ones.
• The results highlight the significant challenges in detecting floating objects in complex aquatic environments.

Conclusions:

• Floating object detection in water is a challenging task requiring further research.
• The IWHR_AI_Lable_Floater_V1 dataset provides a valuable resource for developing improved AI solutions.
• Advancements in AI-driven object detection are essential for effective marine litter management.