본문 바로가기
  • Home

RGB Image Based U-Net Learning Representation for Efficient Flame Segmentation

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2025, 30(4), pp.33~40
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : February 17, 2025
  • Accepted : April 14, 2025
  • Published : April 30, 2025

Jong-Hyun Kim 1

1인하대학교

Accredited

ABSTRACT

This paper proposes a method for efficiently detecting flame regions by extracting color-based features from RGB images and applying segmentation training using a U-Net architecture. The goal of the proposed approach is to accurately identify flame regions commonly observed in fire scenes. To achieve this, the fire images are preprocessed through smoke removal and color correction, followed by a reflection removal step to eliminate surrounding reflections caused by light. The segmented flame regions are then used to train a U-Net model, enabling stable flame segmentation in other fire images as well. Since the proposed method relies solely on RGB color features, it is lightweight in computation, allowing for efficient and reliable detection of flame regions. This makes it highly applicable across various device environments and market settings.

Citation status

* References for papers published after 2023 are currently being built.

This paper was written with support from the National Research Foundation of Korea.