A fine-grained dataset for sewage outfalls objective detection in natural environments

Area of Science:

• Environmental Science
• Computer Science
• Remote Sensing

Background:

• Sewage outfalls (SOs) release pollutants into water bodies, necessitating effective monitoring.
• Manual interpretation of high-resolution images for SO detection is labor-intensive and requires expertise.
• Existing datasets are insufficient for training robust automated SO detection models.

Purpose of the Study:

• To introduce a high-quality, annotated image dataset for sewage outfall detection.
• To facilitate the development of automated detection tools using deep learning and UAVs.
• To enhance intelligent river basin management through improved SO identification.

Main Methods:

• Collected 10,481 images using UAVs and handheld cameras in Chinese river basins.
• Annotated the dataset for accuracy and consistency in sewage outfall identification.
• Validated the dataset using the YOLOv10 object detection model.

Main Results:

• The iSOOD dataset provides a valuable resource for training deep learning models.
• Technical validation demonstrated the dataset's suitability for object detection tasks.
• The dataset supports the advancement of automated SO detection systems.

Conclusions:

• The iSOOD dataset is crucial for developing advanced deep learning models for SO detection.
• Integrating UAVs and AI can lead to efficient and intelligent river basin management.
• This work addresses the need for high-quality data in environmental monitoring applications.