Research On Enhancement Algorithm Of Underwater Biological Image Based On Feature Fusion

Living underwater resources are a critical component of the marine ecosystem and are essential for maintaining the health of balance of nature and marine ecosystems.As an important means of exploitation of marine resources and detection of the environment,the underwater image has important application value.The underwater biological image can be used in the investigation of marine ecosystem and biodiversity,and in the research of underwater ecology,marine biomedicine research,underwater cultural heritage protection and other fields.However,the images obtained by the camera equipment are often of low quality,mainly due to the complexity of the underwater environment,light attenuation and other factors,water molecules and suspended particles in the underwater environment will scatter and absorb light,causes the light to weaken and distort gradually.High-quality underwater images are of great significance for future development.It can provide more detailed and accurate data support for the development and protection of marine resources,help mankind to better understand the state of marine ecology and resources,and achieve the sustainable development and utilization of the oceans.In order to improve the visual perception of underwater biological images,many methods are proposed according to different types.The traditional image enhancement method based on physical model can directly explain the meaning of the parameters in the model,which is easier to understand and analyze,difficult to generalize to other environments.Based on the non-physical model method,it is easy to improve the image quality by directly modifying the image parameters,but it can not deal with the noise and blur in the image.At present,the depth-based learning method is a hot research topic,which has strong generalization and can deal with image texture and detail better.In order to improve the quality of underwater biological image effectively,two different models of underwater biological image enhancement are designed by using convolutional neural network to improve the visual quality of underwater biological image.The research contents of this paper are as follows:(1)In order to solve the quality problems such as serious color deviation and low contrast of underwater image caused by the phenomenon of light absorption and scattering,in this paper,a method of underwater image enhancement based on lightweight feature fusion network and multi-color model correction is proposed.Firstly,a feature fusion network using self-building blocks instead of the structure of codec and encoder in convolutional layer is used to correct the color bias of underwater image,the improved feature fusion module reduces the damage of the full connection layer to the image spatial structure,protects the spatial features and reduces the parameters of the module.At the same time,the improved attention module is used to extract the texture details and protect the background information.Then the multi-color model correction module is used to correct according to the relationship between pixels to further reduce color bias and improve the contrast and brightness.Finally,the experimental results show that,compared with the latest image enhancement methods,the average values of NRMSE,PSNR and SSIM are 9.3%,3.7% and 2.3% higher than those of the second place,respectively.On the no-reference image data set,UCIQE,IE,and NIQE scores improved on average by 6.0%,2.9%,and 4.5% over second place.Combining subjective perception and objective evaluation,this method can correct the color deviation of underwater image,improve the contrast and brightness,and improve the image quality.(2)In order to improve the quality of underwater biological image,eliminate color bias and blur,and enhance the texture and detail of the image,a method of underwater biological image enhancement based on multi-scale feature and inverse medium transfer map fusion is proposed,which aims to improve the quality of underwater biological image and enhance the understanding of underwater ecosystem,protection of underwater biological resources.The method of this paper has two innovations.Firstly,the attention mechanism based on global soft pool channel is proposed and applied to multi-scale feature fusion module.The module can adaptively control the weights of different scale feature images,and then fuse the information according to the weights to improve the image detail and texture.Secondly,the medium transfer map proposed in the physical imaging model is used to enhance the contrast of the image and to solve the problem of blurring the important information of the biological image.In order to verify the method proposed in this paper,a simulated real underwater biological image data set is presented,which uses real lake water and wheat seedling powder solution to simulate turbid and color-biased underwater environment,the Chinese mitten crab underwater biological image data sets(CMCDS)were obtained in this environment.On the open data set EUVP,the proposed method has higher quantization score and better generalization ability than the existing underwater image enhancement methods,the UCIQE,IE,and NIQE scores improved by an average of4.2%,2.1%,and 1.6% over second place.On the CMCDS data set,UCIQE,IE,and NIQE scores improved by an average of 7.9%,6.1%,and 2.0%,respectively,over second place.Based on subjective perception and objective evaluation,The method proposed in this paper has significant enhancement effect on underwater biological image,which can correct color deviation to a large extent,eliminate blur,restore image details,and has strong generalization.