Research on Machine Learning Based Fusion Method for Underwater Oil and Gas Pipeline Image Restoration

To address the issues of insufficient brightness, color distortion, and blurred details in offshore oil and gas pipeline images—problems primarily induced by light absorption, scattering, and suspended particles in the underwater environment—this study proposes a two-stage image enhancement approach that integrates traditional physical enhancement techniques with a deep Transformer-based network. First, gamma correction and wavelet fusion are used to achieve brightness equalization and detail enhancement. Subsequently, the Uformer network is introduced to perform global feature modeling and color consistency reconstruction. The results show that the proposed method not only outperforms the existing algorithms in terms of indicators such as PSNR, SSIM, UIQM, UCIQE and NIQE, but also shows better clarity and detection adaptability in the recovery of texture and corrosion characteristics in the oil and gas pipeline. This work provides an efficient and stable image enhancement scheme for underwater engineering detection and intelligent inspect.