CylinDeRS: A Benchmark Visual Dataset for Robust Gas Cylinder Detection and Attribute Classification in Real-World...

Area of Science:

• Computer Vision
• Machine Learning
• Environmental Monitoring

Background:

• Gas cylinder detection and attribute identification are crucial for industrial safety and combating environmental crimes.
• Existing deep learning applications are limited by the lack of specialized datasets for gas cylinder analysis.
• Online trade growth increases the need for automated systems to monitor hazardous materials.

Purpose of the Study:

• Introduce CylinDeRS, a novel dataset for gas cylinder detection and attribute classification in real-world settings.
• Provide a benchmark for evaluating deep learning models in gas cylinder analysis.
• Facilitate advancements in safety and environmental crime detection.

Main Methods:

• Developed CylinDeRS, a dataset comprising 7060 RGB images with over 25,250 annotated gas cylinder instances.
• Annotated instances for object detection and attribute classification (material, size, orientation).
• Conducted experiments using state-of-the-art models to establish performance baselines.

Main Results:

• Achieved a maximum mean Average Precision (mAP) of 91% for gas cylinder detection.
• Attained a maximum accuracy of 71.6% for attribute classification.
• Demonstrated the dataset's utility and the challenges of real-world gas cylinder analysis.

Conclusions:

• CylinDeRS is a valuable resource for advancing gas cylinder detection and attribute classification research.
• The dataset highlights the complexities of real-world applications and provides a foundation for future work.
• This resource is critical for improving safety and supporting efforts against illegal hazardous substance trade.