Research On Underwater Aquaculture Image Sharpening Algorithm Based On GAN

Factory aquaculture requires monitoring of fish species,size,weight,and water quality parameters to achieve information-based and intelligent aquaculture process control,thereby improving aquaculture efficiency.After in-depth research,we have found that using image processing techniques or probability models based on machine vision can more accurately capture the activities of fish populations,thereby better evaluating their health,growth,and quantity,and more accurately predicting their total mass.By adopting this technology,not only is the equipment simple,but the control process also very convenient,with extremely high estimation accuracy,which can provide reliable data support for the breeding industry and achieve intelligent management.However,due to the lack of sufficient sunlight and scattering,the quality of underwater imaging images has been greatly compromised,and measures must be taken to restore these images to meet the needs of practical applications.This article combines traditional image enhancement methods with deep learning techniques,utilizing pyramid attention mechanism and optical imaging models to extract features and restore underwater image quality.And an image enhancement system was designed and implemented.The main research content of the paper includes the following three parts:(1)An underwater image enhancement method based on pyramid attention mechanism and generative adversarial network.By utilizing the pyramid splitting attention mechanism and the generated adversarial network,we can create a brand new model that can effectively improve the quality of underwater images.This model combines multi-scale pyramid features with attention mechanisms to obtain more advanced features and detailed information in the image,thereby improving the generalization performance of the model.In addition,a new loss function is designed to optimize the GAN network structure.In order to improve the stability of GAN network,this paper combines the losses of CGAN and WGAN-GP,and adds the fusion loss as the total loss function of this method.The new loss function can improve the stability of network training,strengthen the constraint of network model,and enable the network to generate high-quality and clear images.(2)A method for underwater image enhancement based on dual adversarial neural networks has been proposed.Due to the complex underwater environment,this article fully considers the imaging laws of underwater images and combines channel and pixel attention mechanisms to restore underwater images.Firstly,preprocess the image: use the Optical Attenuation Prior(ULAP)algorithm and gamma correction to process the original image;Then the model framework of the improved algorithm was specifically introduced.Secondly,underwater images were input into two generators with the same structure for model training,and channel attention mechanism and pixel attention mechanism were added to the generator to further improve the feature extraction ability of the learning model.Finally,the image is preprocessed by ULAP and GC,and used as the penalty items of the two generators respectively.The pixel values of the two images are fused,and the training process of the generator is optimized by using multiple loss function.Through experiments,it is known that this algorithm can effectively solve the problems of color shift and insufficient illumination in underwater images,and through comparative experiments on synthetic datasets,it can be concluded that this method has good generalization ability.(3)A system has been developed based on the underwater image enhancement algorithm proposed in the first two chapters.The system consists of two parts: the first part uses an improved preprocessing algorithm to process underwater images.The second part uses an improved enhancement algorithm to enhance underwater images,significantly improving the quality of the enhanced image...