Research On Underwater Image Enhancement Algorithm Based On Deep Learning

Underwater images occupy a pivotal position in the fields of intelligent fishing,underwater navigation and water engineering detection of underwater robots.Clear underwater images can provide powerful underwater project detection and classification,underwater mineral exploration and other large underwater projects.Help and support.However,in actual underwater engineering,it is difficult to shoot underwater scenes,and the underwater imaging environment and lighting conditions are more complicated,resulting in very serious texture blurring,low contrast,and color imbalance in the underwater image,which seriously limits the underwater The application of images in the fields of marine geography and marine military.Therefore,in-depth research on the improvement of underwater image quality has very valuable theoretical significance and important practical application value.Therefore,this paper proposes an underwater image enhancement algorithm based on deep learning to solve the problems of local or overall blur and low color saturation in underwater images.First,a residual recursive adversarial network model is used to deblur underwater images.The model adopts a multi-scale architecture,and the network model is consistent at each scale.It consists of a recursive block structure containing four residual blocks and a convolutional long-term and short-term memory network unit.Due to the recursive structure used in the model,the running time of the model during the deblurring process is greatly reduced.This paper also builds a loss function that combines content loss and confrontation loss to reduce the difficulty of the model's convergence during training.Moreover,the model introduces convolutional long-term and short-term memory network elements in each scale network,and uses the jump connection method to realize the "end-to-end" information transmission of underwater images.Compared with other scene deblurring models,this model can effectively remove the blur phenomenon of underwater images and improve the color contrast of the image to a certain extent.Secondly,a super-resolution image reconstruction algorithm using the sparse features of the image subspace is used to enhance the details of the deblurred result image.In this process,the PCANet model with Gaussian function is first used to extract the subspace features of the image,and the sparse optimization of the subspace features is achieved by solving the best left and right projection matrix of its feature map.Then,the learned low-pass filter is used to decompose the sparse features of the subspace into multiple feature maps.After iteratively updating to obtain the optimal solution,the sparse features of the low-resolution image and the mapping function are combined to estimate the sparse features of the super-resolution image.Said.Finally,combined with the corresponding high-pass filter to perform convolution and summation to obtain the final reconstructed image.This method enhances the edge texture and detail information of the image,and solves the problems of color imbalance and low contrast of the underwater image.Finally,through simulation experiments,the model selection and parameter settings in the deblurring phase and the super-resolution image reconstruction process are clarified.The experimental results show that the underwater image enhancement algorithm proposed in this paper improves the subjective visual effect of the image and makes the underwater image more observable.At the same time,through comparison with other underwater image enhancement algorithms,it is found that the algorithm in this paper has also achieved relatively good results in objective evaluation indicators such as peak signal-to-noise ratio,structural similarity,and visual information fidelity.