CN-122023159-A - Underwater image enhancement method and system based on three-channel stochastic resonance

CN122023159ACN 122023159 ACN122023159 ACN 122023159ACN-122023159-A

Abstract

The invention discloses an underwater image enhancement method and system based on three-channel stochastic resonance, and belongs to the technical field of underwater image processing. Aiming at the problems of low contrast, serious noise interference and the like caused by strong backscattering of a turbid underwater image, the invention constructs a stochastic resonance enhancement frame matched with the physical characteristics of an RGB channel. And regarding the target information as a weak signal, realizing detail amplification through noise energy transfer, designing a multi-target parameter self-adaptive module for resolution perception, and dynamically adjusting the parameters of the SR system. The method can effectively inhibit noise, retain texture details, improve the overall perceived quality of the underwater image, and is suitable for the enhancement task of the low-quality and strong-scattering underwater image.

Inventors

SONG AIGUO
HU YUANBO
Ji Qinjie

Assignees

东南大学深圳研究院

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (9)

1. The three-channel stochastic resonance-based underwater image enhancement method is characterized by comprising the following steps of: S1, inputting an underwater image to be enhanced, and extracting a red channel signal, a green channel signal and a blue channel signal of the underwater image in an RGB color space; S2, respectively establishing a bistable Langmuir equation with delay feedback as a stochastic resonance system model of the channel for each channel signal extracted in the step S1, wherein the bistable Langmuir equation is used for the channel The built model is expressed as: , As the original input signal for the channel, In order for the injected ambient noise to be present, And (3) with As a parameter of the potential barrier, Is a channel Is used to determine the delay feedback intensity coefficient of the (c), In order to delay the time constant, Is the state variable of the system at the time t; s3, calculating statistic of each channel signal extracted in the step S1, and dynamically determining initial barrier parameters for the stochastic resonance system model of the corresponding channel in the step S2 based on the statistic And ; S4, adopting a multi-objective optimization algorithm, taking the overall perceived quality index UIQM and the noise suppression index ENL of the enhanced image as combined optimization targets, and adaptively optimizing a group of control parameters for the stochastic resonance system model of the three channels established in the step S2, wherein the control parameters comprise a control parameter for controlling the injection noise Noise parameter of characteristic and delay parameter for controlling delay feedback characteristic , ]; S5, utilizing the control parameters obtained by the optimization in the step S4 and the barrier parameters determined in the step S3 to run the stochastic resonance system model of each channel established in the step S2, and respectively enhancing the signals of each channel to obtain enhanced channel signals; And S6, combining the enhanced three-channel signals obtained in the step S5, and outputting a final underwater enhanced image.
2. The method according to claim 1, wherein the injected ambient noise in step S2 Generated by a Ornstein-Uhlenbeck random process whose dynamic characteristics are determined by the noise parameters to be optimized in step S4 , Control of ] In order to return the rate parameter to the original value, Is a noise intensity parameter.
3. The method of claim 1, wherein the statistics in step S3 include at least a variance and a dominant frequency of the corresponding channel signal.
4. The method according to claim 1, wherein the adaptive optimization using the multi-objective optimization algorithm in step S4 comprises the following sub-steps: S4.1 dividing the control parameters to be optimized into noise parameter sets G noise = [ , Sum delay parameter set G delay = [ , ]; S4.2, initializing a parameter population comprising a plurality of individual solutions, wherein each individual solution comprises a value combination of a noise parameter set and a delay parameter set; s4.3, using the parameter combination to run the enhancement flow of the steps S2 to S5 for each individual solution in the current population, obtaining an enhanced image and calculating UIQM values and ENL values of the enhanced image as two objective function values of the individual solution; s4.4, based on the objective function values of all individual solutions, performing non-dominant sorting to identify pareto fronts; S4.5, selecting individuals from the current front edge and the population according to a reference vector allocation strategy and an ecological niche selection strategy to construct a next generation population; S4.6, performing grouping genetic operation on parameters in the next generation population respectively in the noise parameter group G noise and the delay parameter group G delay to generate new offspring individuals; and S4.7, iteratively executing the steps S4.3 to S4.6 until a preset termination condition is met, and outputting the finally obtained pareto optimal solution set.
5. The method according to claim 4, wherein the grouping genetic operation in step S4.6 is performed, in particular, by performing an analog binary crossover operation on parameters within the same parameter set.
6. The method according to claim 1 or 4, wherein the running stochastic resonance system model in step S5 is implemented by iteratively solving the bistable langerhans equation by a fourth-order langerhans-kutta numerical integration method.
7. An underwater image enhancement system based on three-channel stochastic resonance, comprising: a channel separation module for performing the operation of step S1 of claim 1, extracting RGB three channel signals of the underwater image; a resonance modeling module for performing the operations of step S2 of claim 1, establishing a bistable langerhans' equation model with delay feedback for each channel; A parameter initialization module for performing the operations of step S3 of claim 1, calculating channel statistics and dynamically initializing barrier parameters; a parameter optimization module for performing the operation of step S4 in claim 1, and adaptively optimizing the control parameters of the model by using a multi-objective optimization algorithm; a signal enhancement module, configured to perform the operation described in step S5 of claim 1, and operate each channel model using the optimized and initialized parameters to obtain an enhanced channel signal; An image synthesis module for performing the operation of step S6 of claim 1, combining the enhanced channel signals to output a final image.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 6 when the program is executed by the processor.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Description

Underwater image enhancement method and system based on three-channel stochastic resonance Technical Field The invention relates to the technical field of image enhancement and underwater vision processing, in particular to an underwater image enhancement method and an underwater image enhancement system based on a stochastic resonance theory. Background The underwater optical imaging has important application in the fields of ocean scientific research, underwater robot detection and the like, but the problems of low contrast, color distortion, fuzzy details, serious noise interference and the like of the underwater image are generally caused by the selective absorption of a water body to an optical signal and the scattering effect of suspended particles. The current underwater image enhancement method is mainly divided into a deep learning method and a traditional method. Although the deep learning method has better performance in terms of color correction and detail retention, the method is highly dependent on training data and has limited generalization capability, the traditional method has limited promotion of weak structure signals in complex turbid scenes and is easy to amplify noise, and the method based on a physical model is dependent on accurate estimation of parameters such as depth, attenuation coefficient and the like, so that the problems of artifact and color overcompensation are easy to occur when scattering is serious and depth estimation is not right. In addition, the existing underwater image enhancement method based on stochastic resonance theory generally processes images as a whole or a single channel, and cannot independently model the differences of RGB three-channel signal attenuation and scattering noise according to a physical model (such as Akkaynak model) of underwater light attenuation. Meanwhile, the stochastic resonance model cannot be effectively integrated into a delay feedback mechanism capable of simulating the underwater backscattering space-time correlation, and the adjustment of system parameters is dependent on experience or single-target optimization, so that the self-adaptive balance between the promotion of weak signals (such as image details) and the inhibition of strong noise is difficult to realize, and the enhancement effect in complex turbid underwater scenes is limited. Disclosure of Invention In order to solve the problems, the invention provides an underwater image enhancement method and an underwater image enhancement system based on three-channel stochastic resonance, which realize cooperative optimization of weak signal amplification and noise suppression through multi-channel stochastic resonance modeling, noise energy transfer and multi-target parameter self-adaptive adjustment. The technical scheme is as follows: An underwater image enhancement system based on three-channel stochastic resonance, comprising: 1. The sub-channel stochastic resonance enhancement module, underwater imaging, follows the Akkaynak model (equation 1), which reveals the physical facts that red light decays most rapidly, signals are weakest, but scattering noise is relatively small, and blue-green light decays more slowly, accompanied by stronger scattering noise. Based on the method, a bistable Langmuir equation (formula 2) with delay feedback is established for each channel so as to simulate the space-time correlation of underwater back scattering and control the transfer efficiency of noise energy to signal energy through barrier height; (1) (2) In the multichannel stochastic resonance enhancement model, a parameter of [ ,The Ornstein-Uhlenbeck process of the control simulates underwater environmental noise. 2. The parameter self-adaptive adjusting module dynamically estimates barrier parameters of each channel according to statistic such as channel variance, dynamic range, internal noise variance, main frequency and the like, takes UIQM (perceived quality) and ENL (equivalent vision number, measured noise suppression) of an image as combined optimization targets, adopts a multi-target evolutionary algorithm based on reference vectors to carry out grouping collaborative optimization on a noise parameter set and a delay parameter set, and automatically searches for a pareto optimal solution set, thereby achieving optimal balance among detail enhancement, noise suppression and overall visual quality. 3. And the image reconstruction module is used for iteratively solving a system equation through a fourth-order Dragon-Kutta method to obtain enhanced channel signals, recombining the enhanced channel signals and outputting the enhanced underwater image. In the parameter self-adaptive adjusting module, the detailed operation is as follows: Step 1, parameter grouping and initialization The underwater image and statistics (such as variance, dominant frequency, etc.) of each channel are input. The parameters to be optimized are divided into two groups of physical meaning association: