UGV-NBWASTE: An oriented dataset for non-biodegradable waste in Bangladesh

Area of Science:

• Environmental Science
• Computer Vision
• Data Science

Background:

• Non-biodegradable waste poses significant environmental challenges, particularly in landfills and aquatic ecosystems.
• Effective management strategies require accurate identification and classification of diverse waste materials.
• Existing datasets may lack the diversity needed to address real-world waste scenarios.

Purpose of the Study:

• To introduce the UGV-NBWASTE dataset for non-biodegradable waste management.
• To provide a diverse collection of waste images captured in challenging environmental conditions.
• To facilitate the development of robust waste detection and classification models.

Main Methods:

• Collected 3600 images of eight distinct non-biodegradable waste types from Bangladesh.
• Captured images using smartphones in varied indoor and outdoor settings, including water and mud.
• Applied image pre-processing techniques: quality assurance (verification, cleaning) and enhancement (brightness normalization, resizing).
• Annotated images using oriented bounding boxes (OBB) for angle-invariant detection.

Main Results:

• The UGV-NBWASTE dataset comprises 3600 images of eight common, hard-to-manage non-biodegradable waste items.
• Images represent diverse conditions, including floating in water and partially buried in mud.
• Oriented bounding box annotations enable reliable waste identification regardless of orientation or condition (flat, crumpled, obscured).

Conclusions:

• The UGV-NBWASTE dataset offers a valuable resource for advancing non-biodegradable waste management research.
• The dataset's diversity and annotation format support the development of more accurate and resilient waste detection systems.
• This resource can aid in tackling environmental pollution caused by persistent waste materials.