Research On Underwater Image Enhancement Method Based On Deep Learning

Recently,in order to build a maritime power and build a community of marine destiny,China’s technology,information and culture,which are the carrier and link of the ocean,have greatly increased in the proportion of the Marine field.Among them,the observation and research of Marine resources,intelligent fishing,is the focus of the development of the Marine industry.However,underwater image optimization algorithm has become an important technical support for the Marine industry.Marine resources research and technology occupies a place.In reality,due to the extremely complex underwater environment,which is often affected by illumination,wavelength,impurities in the water and many other factors,the images generated by the underwater imaging system appear blurred,degraded,color bias and other phenomena,which bring huge obstacles to the identification and positioning of Marine resources.In order to solve the problems of blur and color bias,three image enhancement algorithms are proposed in this paper.Firstly,an underwater image denoising algorithm based on non-local filter is proposed to enhance the image by removing noise.Secondly,an underwater image restoration method based on deep convolutional neural network is proposed,which takes removing color polarization as the core to enhance image.Finally,an underwater image restoration and enhancement algorithm based on Dual Attention Residual Network is proposed to enhance the details of the image.The work completed in this paper includes the following parts:(1)In order to solve the problem of Gaussian,pepper and salt mixed noise in the process of image acquisition and transmission,an underwater image denoising algorithm based on non-local filter is proposed.Firstly,Gaussian and salt-and-pepper noises are identified.Then,statistical knowledge is used to estimate the intensity of the mixed noise and remove the noise.Next,a group of discrete full-variational multi-directional contour templates are used to mine the texture information in the image.Finally,the missing details of the image are recovered by non-local filter.(2)A deep convolutional neural network-based algorithm is proposed to enhance the color of underwater images caused by special underwater environment.The algorithm adopts the powerful U-Net network as the foundation,to build a kind of based on partial color image convolution neural network,learning of input images and output images color deviation,and set the structure similarity measure as the loss function,make the enhanced(3)Underwater image with the input of underwater image details remain highly similar in content structure,on the basis of underwater image color enhancement,and retain the original image texture of structure.(4)An underwater image restoration and enhancement algorithm based on dual attention residuals network is proposed by combining the first and second parts.This algorithm adopts a novel denoising method,which combines the traditional denoising algorithm with the deep learning algorithm to remove color bias.By embedding the non-local attention module and the channel attention module in the network,the two attention mechanisms are used to jointly explore and enhance the potential of more image optimization.The three algorithms proposed in this paper have a good performance in improving the recognition degree of underwater images,accurate positioning,removing color bias,degradation,blur and other problems.The enhanced images are not only conducive to the identification and analysis of Marine organisms and seabed resources,but also of great significance to the seabed exploration in China.