Method for visual detection of the state of the surface of underwater structures in the presence of interfering water bodies

Abstract

This invention discloses a visual detection method for the surface condition of underwater structures in aquatic environments containing interfering objects. The method includes the following steps: First, a visualized optical flow map is obtained through self-supervised learning optical flow technology and preprocessed. A threshold is determined using the Otsu method to segment the background and foreground. A novel inter-frame difference method based on adjacent frame alignment is used to further search for small-sized underwater interfering objects. STTN is used to recover the regions occluded by underwater interfering objects in the current frame. Then, PVTv2 is used as the backbone to extract the feature pyramid of the input image, while AGFE and MFCA are integrated to improve the efficiency of underwater structure defect detection. Finally, a loss function is defined to supervise the defect detection network, allowing for better selection of optimal parameters based on the output results. This invention provides high accuracy and strong environmental adaptability for detecting the surface condition of underwater structures in aquatic environments containing interfering objects, improving both detection efficiency and accuracy.

Images