Semantic Enhanced Segmentation Based on Thermal Images with Superpixel

Semantic segmentation is receiving increasing attention from researchers. Many emerging applications require accurate and efficient segmentation mechanisms, and this need coincides with the rise of almost all fields related to computer vision, especially in the construction industry. Building a semantic segmentation model for accurate and rich semantic information to identify, classify and segment building/structural components in construction site scenarios can greatly improve the productivity and informatization of the construction industry and has great potential for automated construction and visual monitoring. However, the accuracy of traditional image recognition technology is low, and it is difficult to adapt to the complex environment, and the illumination and Angle factors will affect the reliability of the final model. To address these challenges, this paper proposes a method to improve the segmentation performance of semantics using thermal images. By using temperature as a characteristic to distinguish different materials, the proposed method improves the accuracy of the segmentation model and has a high potential for automated construction and visual monitoring.