CartoMark: a benchmark dataset for map pattern recognition and map content retrieval with machine intelligence

Area of Science:

• Geographic Information Science
• Computer Vision
• Artificial Intelligence

Background:

• Maps are crucial for visualizing real-world data, with big data increasing map complexity.
• Many map datasets are underutilized due to a lack of standardized, labeled benchmark datasets.
• Deep learning methods require extensive labeled data for effective map content analysis.

Purpose of the Study:

• To develop a large-scale, well-labeled benchmark dataset for map analysis.
• To facilitate the application of machine intelligence techniques to diverse map-related tasks.
• To improve the discovery, acquisition, and utilization of map data.

Main Methods:

• Creation of a comprehensive benchmark dataset with detailed labels.
• Application of state-of-the-art machine intelligence for map text annotation recognition.
• Utilizing deep learning for map scene classification, super-resolution, and style transfer.

Main Results:

• A novel, large-scale benchmark dataset for map analysis is established.
• Demonstrated potential for machine intelligence in map text recognition and scene classification.
• Facilitated advancements in map feature detection, pattern recognition, and content retrieval.

Conclusions:

• The developed dataset addresses the critical need for labeled resources in map intelligence.
• This work empowers advanced machine learning applications for understanding complex map data.
• The dataset will accelerate research and development in automated map analysis and information discovery.