CN-122023151-A - Underwater image enhancement method based on multichannel polarization information

CN122023151ACN 122023151 ACN122023151 ACN 122023151ACN-122023151-A

Abstract

The invention belongs to the technical field of image processing, and particularly discloses an underwater image enhancement method based on multichannel polarization information, which comprises the steps of constructing a U-shaped convolutional neural network formed by a multi-stage encoder and a decoder, and introducing space and channel double-branch attention based on polarization degree at a jump joint; the method comprises the steps of collecting paired color images of a sample in a scattering-free environment and a scattering-free environment in a plurality of polarization directions, carrying out cross recombination of a polarization channel and a color channel according to the scattering-free environment color images to generate a plurality of pseudo-color polarization images, calculating Stokes components from the plurality of polarization directions to obtain a single-channel polarization degree image, constructing a training set, taking the scattering-free image as a supervision tag training network, inputting the underwater polarization image to be restored into the training network, and outputting a restored image.

Inventors

LI LIJING
CAO YIN
FAN RONG

Assignees

北京航空航天大学

Dates

Publication Date: 20260512
Application Date: 20251203

Claims (10)

1. The underwater image enhancement method based on the multichannel polarization information is characterized by comprising the following steps of: constructing a convolutional neural network comprising a multi-stage encoder and a decoder, and setting a double-branch attention mechanism at jump joints between the encoder and the decoder at each stage according to the polarization degree; acquiring a first color image of the sample in a non-scattering environment and a second color image of the sample in at least one scattering environment, respectively, in a plurality of polarization directions; According to the sub-images of the second color image in different polarization directions, generating multiple pseudo-color polarization images through cross recombination of a polarization channel and a color channel, and generating a single-channel polarization degree image through Stokes components calculated in the multiple polarization directions; Constructing a training set through the pseudo-color polarization image and the polarization degree image, inputting the training set into the convolutional neural network, and training the convolutional neural network by taking the first color image as a supervision tag; And inputting the underwater polarized image to be restored into the trained convolutional neural network to output an underwater restored image.
2. The underwater image enhancement method according to claim 1, wherein the architecture of the convolutional neural network is a U-shaped structure, and the step of constructing the convolutional neural network including a multi-stage encoder and decoder comprises: a plurality of encoders are sequentially arranged at the input end from top to bottom, and each encoder compresses the input characteristics into deep characteristics step by step through a downsampling structure; a bottleneck layer is arranged between the bottom positions of the input end and the output end and used for bearing the deepest layer characteristics and serving as a connecting node between the coding path and the decoding path; And a plurality of decoders are sequentially arranged at the output end from bottom to top, and the deep layer characteristics of the decoders are recovered step by step through an up-sampling structure.
3. The underwater image enhancement method as claimed in claim 2, wherein the step of setting the dual branch attention mechanism according to the degree of polarization at the jump junction between the encoders and decoders of each stage comprises: At jump junctions between the encoder and the decoder of each stage, respectively generating a spatial attention weight and a channel attention weight according to the polarization degree image; applying the spatial attention weights to features from the encoder to enhance the weighted spatial regions according to the degree of performance of the spatial response in the polarization degree distribution; the channel attention weights are applied to features from the encoder to attenuate the weighted channel components according to the contribution of each channel to the polarization degree variation.
4. The underwater image enhancement method of claim 1, wherein the step of acquiring a first color image of the sample in a non-scattering environment and a second color image of the sample in at least one scattering environment in a plurality of polarization directions, respectively, comprises: Acquiring a first color image of the sample in a plurality of polarization directions in the non-scattering environment; Acquiring a color polarization image corresponding to the sample in at least one scattering environment having different simulated scattering characteristics; Repeating the acquisition process according to the scattering intensity change of the scattering medium, and forming all the color polarization images into a second color image corresponding to the first color image.
5. The underwater image enhancement method of claim 4, wherein the step of generating a plurality of pseudo-color polarization images by cross-recombining polarization channels and color channels based on the sub-images of the second color image in different polarization directions comprises: Extracting m different color channels from the obtained sub-image, and selecting m polarization channels from all kinds of polarization directions; and establishing a corresponding relationship which is not repeated between the color channels and the polarization channels so as to form a pseudo-color polarization image which is obtained by combining m color channels with m polarization channels which are different from each other.
6. The underwater image enhancement method of claim 5, wherein the pseudo-color polarization image is recombined with a polarization channel through a color channel to characterize a different polarization response than the second color image and expand the sample number of the training set.
7. The underwater image enhancement method of claim 6, wherein the step of generating the polarization degree image by stokes components calculated by a plurality of polarization directions comprises: Based on a plurality of color polariton images acquired by m polarization channels, respectively carrying out light intensity sampling on the color channels under each polarization channel to determine Stokes components used for representing the polarization states; calculating a polarization degree value for representing the proportion of the polarized component to the total light intensity at a pixel level according to the Stokes component; and carrying out fusion processing on the polarization degrees corresponding to the different color channels to generate the polarization degree image.
8. A system for implementing the underwater image enhancement method as claimed in any of claims 1 to 7, comprising: An image acquisition module for acquiring a first color image and a second color image in a plurality of polarization directions; a pseudo-color recombination module for generating a pseudo-color polarized image based on cross recombination of the color channel and the polarization channel of the second color image; The polarization degree generation module is used for calculating Stokes components according to a plurality of polarization directions and generating a single-channel polarization degree image; The training module is used for inputting the training set and the first color image into a convolutional neural network to execute network training; The image restoration module is used for inputting the underwater polarized image to be restored into the trained convolutional neural network so as to output a restored image.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and capable of running on the processor, characterized in that the processor implements the steps of the underwater image enhancement method as claimed in any of claims 1 to 8 when the computer program is executed.
10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the underwater image enhancement method as claimed in any of claims 1 to 8.

Description

Underwater image enhancement method based on multichannel polarization information Technical Field The invention relates to the technical field of image processing, in particular to an underwater image enhancement method based on multichannel polarization information. Background In the underwater imaging process, the suspended particles have strong back scattering and absorption effects on light rays, so that the problems of blurring, contrast reduction, color distortion and the like often occur in the image. The existing underwater polarization imaging method based on the physical model is characterized in that the model containing parameters such as transmissivity, scattering proportion, background light intensity and the like is constructed, and images with different polarization angles are utilized for inversion to inhibit scattering, but the method is highly dependent on priori assumption and parameter estimation, so that the method is difficult to adapt to complex and changeable actual scenes in turbid water bodies, and has weak generalization capability. With the development of deep learning, the data-driven polarized image restoration method can learn the mapping relation between the degraded image and the clear image, but generally only adopts gray polarized images or simply stacks polarized sub-channels, fails to fully utilize the synergistic characteristics of RGB color channels and polarized information, is limited by scarcity of real underwater data, single data enhancement mode and the like, and has insufficient stability and generalization capability under complex scattering conditions. Therefore, an underwater image enhancement method based on multichannel polarization information is provided to solve the problems. Disclosure of Invention The invention aims to provide an underwater image enhancement method based on multichannel polarization information, which aims to solve or improve the problems that the existing underwater polarization image restoration method has obvious limitations in the aspects of strong parameter acquisition dependence, weak generalization capability caused by insufficient polarization and color information and insufficient data, and the like, and is difficult to obtain a stable and reliable clear image in a complex scattering environment. In view of the above, a first aspect of the present invention is to provide a method for enhancing an underwater image based on multi-channel polarization information. A second aspect of the invention is to provide a system A third aspect of the present invention is to provide an electronic device. A fourth aspect of the present invention is to provide a computer-readable storage medium. The invention provides an underwater image enhancement method based on multichannel polarization information, which comprises the steps of constructing a convolution neural network comprising a multistage encoder and a multistage decoder, setting a double-branch attention mechanism according to polarization degree at jump joints between the encoder and the decoder at each stage, respectively acquiring a first color image of a sample in a non-scattering environment and a second color image in at least one scattering environment in a plurality of polarization directions, generating a plurality of pseudo-color polarization images through cross recombination of polarization channels and color channels according to sub-images of the second color image in different polarization directions, generating a single-channel polarization degree image through Stokes components calculated in the plurality of polarization directions, constructing a training set through the pseudo-color polarization images and the polarization degree image, inputting the training set into the convolution neural network, training the convolution neural network by taking the first color image as a supervision tag, and inputting the underwater polarization image to be restored into the trained convolution neural network to output an underwater restoration image. The second aspect of the invention provides a system which comprises an image acquisition module, a pseudo-color recombination module, a polarization degree generation module, a training module and an image restoration module, wherein the image acquisition module is used for acquiring a first color image and a second color image in a plurality of polarization directions, the pseudo-color recombination module is used for generating a pseudo-color polarization image based on cross recombination of a color channel and a polarization channel of the second color image, the polarization degree generation module is used for calculating Stokes components according to the plurality of polarization directions and generating a single-channel polarization degree image, the training module is used for inputting the training set and the first color image into a convolutional neural network to execute network training, and the image restoratio