Damage area identification of underwater structures based on wavelength difference of light source

Traditional optical detection of underwater structures suffers from imaging blur affecting underwater object detection tasks. Hence, this work proposes a damaged area recognition method for underwater structures that exploits the differential characteristics of various light source wavelengths, solving the automatic recognition problem of fuzzy images representing underwater structures. Specifically, this paper constructs the spectral variance, local, and global feature representations of an optical image and presents an underwater structure by exploiting the various characteristics of the component images corresponding to different wavelengths. Multiple image features are organically fused through multivariate data fusion, enhancing the optical image of underwater structures. At the same time, we also developed a quantitative characterization method appropriate for underwater structure damage region estimation in optical images that effectively realizes the feature quantization process of the damaged region. Finally, we demonstrate the feasibility and effectiveness of the proposed method on an actual engineering test project. The results highlight that our method improves the image quality and recognition ability of the damaged areas, affording their automatic recognition. Considering image quantization, the suggested technique effectively realizes the quantitative characterization of a damaged area and provides the data support for the health diagnosis of underwater structures.