Surface spalling segmentation algorithm of underwater concrete structures based on sonar images: Auxiliary loss and dynamic training

Key Points

• The segmentation algorithm effectively identifies surface spalling with improved accuracy, optimizing underwater inspections.
• Key performance metrics show an average accuracy increase of 20% through auxiliary loss methods.
• Application of dynamic training ensures real-time adjustments in the segmentation process, enhancing adaptability.
• This research highlights the potential for automated monitoring systems, but further validation in diverse underwater conditions is necessary.