CN-121981915-A - Method for generating underwater image enhancement of countermeasure network based on U-Net structure

CN121981915ACN 121981915 ACN121981915 ACN 121981915ACN-121981915-A

Abstract

The invention discloses a method for generating an countermeasure network underwater image enhancement based on a U-Net structure, which comprises the steps of obtaining a fuzzy underwater image, constructing an underwater image enhancement model, inputting the fuzzy underwater image into the underwater image enhancement model, and obtaining a clear underwater image, wherein the underwater image enhancement model is constructed by adopting a generated countermeasure network and is obtained through training of a training set, and the generated countermeasure network comprises a generator of the U-Net network structure and a PatchGAN discriminator. According to the invention, the antagonism network is improved and generated on the basis of U-Net, and the attention mechanism is added to the generator part, so that the generator is more focused on an important part in the task of generating the image, and the discriminator uses the discriminator network in PAtchGAN to obtain a larger receptive field, thereby effectively enhancing the blurred underwater image.

Inventors

LI GANG
Cheng Chenyu
LI JINGLI
MA YONGJIE

Assignees

白城师范学院

Dates

Publication Date: 20260505
Application Date: 20231222

Claims (7)

1. The method for generating the underwater image enhancement of the countermeasure network based on the U-Net structure is characterized by comprising the following steps: acquiring a blurred underwater image; And constructing an underwater image enhancement model, inputting the blurred underwater image into the underwater image enhancement model, and obtaining a clear underwater image, wherein the underwater image enhancement model is constructed by adopting a generated countermeasure network and is obtained through training of a training set, and the generated countermeasure network comprises a generator of a U-Net network structure and a PatchGAN discriminator.
2. The method for generating an anti-network underwater image enhancement based on a U-Net structure according to claim 1, wherein the generator of the U-Net network structure comprises a SE attention mechanism module and a U-Net network, the U-Net network comprising an encoder and a decoder, the SE attention mechanism module being disposed between the encoder and the decoder.
3. The U-Net structure based generation countermeasure network underwater image enhancement method of claim 2, wherein inputting the blurred underwater image into the underwater image enhancement model, obtaining a clear underwater image comprises: inputting the blurred underwater image into an encoder of the U-Net network, and adopting convolution Leaky-ReLU activation and batch normalization processing to output a feature map; the feature map passes through an SE attention mechanism module to generate an enhanced feature map; And inputting the characteristic diagram and the enhancement characteristic diagram into a decoder of the U-Net network, and finally outputting the clear underwater image by deconvolution, dropout and batch normalization.
4. The method for generating an underwater image enhancement of an countermeasure network based on a U-Net structure according to claim 1, wherein said PatchGAN discriminator comprises a plurality of convolution layers, each convolution layer containing a leak-ReLU and BN.
5. The U-Net structure based generation countermeasure network underwater image enhancement method of claim 1, wherein training the underwater image enhancement model by a training set comprises: inputting the fuzzy underwater image in the training set into a generator of the U-Net network structure to obtain a clear underwater image; Inputting the clear underwater image and the real underwater image into the PatchGAN discriminator, discriminating the difference of each Patch, training the PatchGAN discriminator according to the clear underwater image and the real underwater image, and obtaining a discrimination result; inputting the discrimination result into the generator of the U-Net network structure, and training the generator of the U-Net network structure.
6. The method for generating an anti-network underwater image enhancement based on a U-Net structure according to claim 5, wherein training said underwater image enhancement model by a training set further comprises weighting and fusing L 1 loss, content loss and histogram matching loss to obtain an objective function.
7. The method for generating an opposing network underwater image enhancement based on a U-Net structure according to claim 6, wherein said objective function is Wherein, G t is an objective function, L cGAN (G, D) is an countermeasure loss, G, D are a generator and a discriminator, respectively, L 1 、L C and L his are divided into L 1 loss, content loss and histogram matching loss, and λ 1 ,λ 2 is a weight.

Description

Method for generating underwater image enhancement of countermeasure network based on U-Net structure Technical Field The invention belongs to the technical field of image processing and deep learning, and particularly relates to a method for enhancing an underwater image of a generated countermeasure network based on a U-Net structure. Background The image is used as an important information carrier, has an irreplaceable role in exploring a deep sea environment, and the clear underwater image has important significance for researching deep sea resources. However, underwater images often exhibit fading, blurring, distortion, and the like due to their special physical characteristics, so obtaining an undistorted underwater image with high sharpness is a challenging task. At present, the meaning of the restoration and enhancement algorithm of the underwater degraded image is to improve the phenomena of low contrast, low distortion and the like of the underwater image so as to obtain a clear underwater image. The enhancement of the underwater image has long-term significance to underwater detection work, such as the research of underwater detection vehicles, underwater biology, archaeology, the inspection and maintenance of underwater facilities and the like. The conventional image optimization algorithms of the underwater image enhancement and restoration technology include histogram equalization, wavelet transformation, retinex algorithm, image fusion algorithm and the like. In addition, in recent years, image optimization algorithms based on deep learning, which are developed at high speed, can be divided into two types, namely algorithms based on convolutional neural networks and algorithms based on generation countermeasure networks. The prior art has the disadvantage of recognizing that the image is not authentic and cannot be adapted to complex underwater environments. Due to factors such as water quality, illumination and the like in an underwater environment, color in an image can be distorted and offset, and the color distortion cannot be effectively corrected, so that the image quality is reduced. Underwater images may be affected by noise, and existing techniques may not accurately distinguish between noise and valid information in the image, resulting in lost detail or poor image quality. The existing underwater image enhancement technology generally lacks self-adaptive capability to different underwater environments, and cannot flexibly cope with different illumination conditions, water quality differences and scene changes. Disclosure of Invention In order to solve the technical problems, the invention provides a method for enhancing the underwater image of the generation countermeasure network based on the U-Net structure, which can effectively enhance the blurred underwater image and is convenient for the subsequent utilization of the underwater image. In order to achieve the above object, the present invention provides a method for generating an underwater image enhancement of an countermeasure network based on a U-Net structure, comprising: acquiring a blurred underwater image; And constructing an underwater image enhancement model, inputting the blurred underwater image into the underwater image enhancement model, and obtaining a clear underwater image, wherein the underwater image enhancement model is constructed by adopting a generated countermeasure network and is obtained through training of a training set, and the generated countermeasure network comprises a generator of a U-Net network structure and a PatchGAN discriminator. Optionally, the generator of the U-Net network structure includes an SE attention mechanism module and a U-Net network, the U-Net network includes an encoder and a decoder, and the SE attention mechanism module is disposed between the encoder and the decoder. Optionally, inputting the blurred underwater image into the underwater image enhancement model, and obtaining the clear underwater image includes: inputting the blurred underwater image into an encoder of the U-Net network, and adopting convolution Leaky-ReLU activation and batch normalization processing to output a feature map; the feature map passes through an SE attention mechanism module to generate an enhanced feature map; And inputting the characteristic diagram and the enhancement characteristic diagram into a decoder of the U-Net network, and finally outputting the clear underwater image by deconvolution, dropout and batch normalization. Optionally, the PatchGAN discriminator includes a number of convolutional layers, each layer comprising a leak-ReLU and BN. Optionally, training the underwater image enhancement model by a training set includes: inputting the fuzzy underwater image in the training set into a generator of the U-Net network structure to obtain a clear underwater image; Inputting the clear underwater image and the real underwater image into the PatchGAN discriminator, discriminating the