Research On Underwater Image Enhancement Algorithm Based On Deep Learning

China is rich in marine resources,but it has not been fully exploited.Compared with rivers,lakes and other waters,the underwater environment of the ocean is more complex and dangerous,and the risk coefficient of artificial exploration and development is too high.Underwater vehicle(AUV)has become an important tool for human beings to explore the ocean,and visual images play an important role in exploring and perceiving the surrounding environment of AUV.Due to the absorption and scattering of the water body,the underwater image has some problems,such as low contrast,image blur,color deviation and so on,which affect the follow-up vision task of the underwater vehicle.Therefore,the acquisition of highquality underwater images is of great significance for human exploration and development of the ocean.According to the principle of underwater image imaging,this paper studies the underwater image enhancement technology,and mainly does the following research work:(1)First of all,in order to solve the problem that the traditional algorithm is not suitable for the changeable underwater environment,this paper proposes to optimize the parameters of underwater dark channel prior algorithm.Secondly,in order to solve the problem of large texture and long computing time in the image enhanced by the underwater dark channel prior algorithm,this paper combines the parameter optimization with the underwater dark channel prior loss,and obtains additional regularization through network training.The enhanced image color and contrast are better,and the texture and block problems are reduced.At the same time,through the training model,the overall amount of calculation is greatly reduced,and the calculation time is also greatly reduced.Finally,in order to solve the problem that traditional algorithms can not process video images in real time,this paper proposes a strategy that only calculates in stable frames.In this paper,several groups of experiments are compared,and through subjective evaluation and objective evaluation indicators.It is verified that the enhanced image color and contrast are better,and can adapt to different underwater environments.Through the analysis of the real-time performance of the algorithm,the feasibility of the application of this algorithm in engineering is verified.(2)Aiming at the problem that the generating countermeasure network lacks the underwater truth image during the underwater image enhancement,which leads to the low contrast of the generated image.First of all,this paper proposes to add the NIQE index to the loss of the generator to make the generated image have a higher contrast and more in line with the human eye perception,and at the same time make the generated image have a better effect than the truth image set by the existing data set.Secondly,this paper proposes to add the NIQE index to the discriminator structure to diversify the discriminant factors of the discriminator in FUn IE-GAN.Finally,this paper proposes a new GAN structure,which trains 10 generators with different weights according to the generator loss,and selects the generated image which is suitable for this paper.In this paper,several groups of experiments are compared,and through subjective evaluation and objective evaluation indicators.It is verified that the enhanced image of this algorithm is better than the truth image set by the existing data set.Through the analysis of the real-time performance of the algorithm,the feasibility of the application of this algorithm in engineering is verified.