A novel image multitasking enhancement model for underwater crack detection

Remotely Operated Vehicles (ROVs) carrying vision systems provide an efficient solution for the underwater crack search. However, the degradation of underwater images severely limits the prognosis of cracks. For the problem of ROV image multiple degradation in complex underwater environments, a robust and accurate multitask enhancement method for underwater crack images is proposed, which can simultaneously enhance the color, brightness. and deblurring of images. In the model, we propose a depth-residual encoder-decoder and feature calibration module to address low-level feature loss. Meanwhile, we propose a simulation method to construct paired training data. Experiments show that our model outperforms existing methods in image enhancement and provides significant enhancements for downstream tasks. The model has been successfully applied to practical engineering and shows good adaptability, which can well assist ROVs for underwater crack detection. In future work, we will continue to improve the robustness of the ROV crack detection system under more complex noise scenarios.