Dataset for flood area recognition with semantic segmentation

Area of Science:

• Computer Science
• Environmental Science
• Data Science

Background:

• Floods are recurrent natural disasters in Indonesia, causing significant damage and loss of life.
• Social media data, particularly from platforms like Instagram Reels, is increasingly utilized for disaster-related information analysis.
• Existing datasets may not fully capture the dynamic and diverse visual aspects of flood events.

Purpose of the Study:

• To introduce a novel, large-scale dataset of flood events captured from Instagram Reels in Parepare, Indonesia.
• To provide a valuable resource for computer vision research, specifically for semantic segmentation and object recognition in flood-affected areas.
• To facilitate the development of advanced algorithms for automated flood area identification and damage assessment.

Main Methods:

• Collected 7248 video clips depicting flood events in Parepare from Instagram Reels.
• Converted video data into image format, ensuring diverse conditions (area, time, viewpoint).
• Preprocessed images for clarity and applied object annotations with distinct color labels for computer vision tasks.

Main Results:

• A comprehensive dataset of 7248 preprocessed and annotated images of flood events.
• The dataset captures diverse visual characteristics of floods from various perspectives.
• Annotations facilitate the training and evaluation of computer vision models.

Conclusions:

• The presented flood dataset is a significant contribution to computer vision research for disaster management.
• It enables advancements in semantic segmentation and object recognition for flood-related applications.
• This resource supports the development of more effective flood monitoring and response systems.