The authors suggest a novel detection method for ceiling damage using image differencing and Grad-CAM. The proposed method requires the latest image and the past image of the facility to compare. It adjusts the difference due to camera position, sunlight condition, etc. The authors show validity of this method showing four pairs of input images. The proposed method can appropriately detect damaged areas with better accuracy for facility inspection by narrowing the areas of inspection using image differencing and Grad-CAM procedures.
