YOLO photothermal image combined with external prestressed duct grouting cavity detection method

In the realm of intelligent detection for external prestressed duct grouting cavity, misidentifying the background of an infrared thermal image as part of the duct leads to erroneous cavity identification. This study introduces a novel method for detecting grout cavities by integrating visible light and infrared thermal images. By capturing visible light images and infrared thermal images of specimens at similar angles, the YOLOv5 model was employed to detect duct targets in the visible light images, extracting coordinates of the duct region to generate masks. Subsequently, the mask aligns with the corresponding area of the infrared thermal image using an image alignment method for further processing. Following that, the YOLOv5-seg model accurately segments the grout cavities within the duct region of the processed infrared thermal image. The findings illustrate a significant enhancement in the mean Average Precision at intersection-over-union thresholds from 0.5 to 0.95 ([email protected]:0.95) index of the YOLOv5s-seg model within this approach, increasing from 76.1 % to 85.2 %, thereby boosting recognition accuracy and mitigating external background interference in grout cavity segmentation. This successful resolution addresses recognition inaccuracies stemming from blurred edges, low contrast, and background interference when solely relying on infrared thermal images. The problem of recognition inaccuracy caused by blurred edge, low contrast and background interference in single use of infrared thermal image is solved successfully.