Research on underwater fish image denoising method based on deep learning

Underwater fish imagery is an important tool for the study of marine ecology, revealing biodiversity, ecosystem health and changes in fish distribution, helping to protect endangered species and promoting sustainable fisheries. When imaging underwater, image quality is seriously disturbed by a variety of noises, which can significantly affect the clarity and accuracy of the image. In this paper, an underwater fish image denoising method based on deep learning is proposed. The algorithm model consists of two parts: noise estimation subnetwork and non-blind subnetwork. In this case, the noise estimation subnetwork is used to estimate the noise level graph, and the non-blind denoising subnetwork is U-Net with residual for noise reduction. Compared with the traditional deep learning underwater image denoising algorithm, this paper introduces the asymmetric loss function as the basis of algorithm convergence judgement, which further enhances the denoising performance and generalisation ability of the algorithm. In addition, the algorithm is trained with real noisy image pairs to improve the denoising effect. The experimental results show that the method can significantly improve the clarity and recognition rate of underwater fish images while maintaining the image details. The research in this paper provides a new idea and method for underwater biological image processing, which is of great significance for underwater ecological monitoring, aquaculture and other fields.