Underwater robotic turbid water vision enhancement method and system

Abstract

This invention discloses a method and system for visual enhancement of underwater robots in turbid waters, relating to the fields of machine vision and image restoration technology. The method includes: acquiring a sequence of original color images collected in real-time by the underwater robot; mapping the pixel intensity of the original color image sequence to a complex space to construct a complex domain analytic signal; extracting the multi-directional spatial gradient tensor of the original color image, and calculating pseudo-polarization information and environmental characterization parameters accordingly; constructing a virtual phase compensation operator using the pseudo-polarization information and the environmental characterization parameters; wherein the pseudo-polarization information drives the compensation intensity of the virtual phase compensation operator, and the environmental characterization parameters are used to locally weight the compensation intensity. Its beneficial effects include: correcting edge blurring and texture obfuscation caused by wavefront phase disturbances, and significantly improving the edge steepness of structural contours in the image.

Description

Technical Field

[0001] This invention relates to the field of machine vision and image restoration technology, and in particular to a method and system for enhancing the vision of underwater robots in turbid waters. Background Technology

[0002] In tasks such as underwater inspection of ports, dam inspection, and marine ranch monitoring, the optical imaging system carried by underwater robots (ROVs / AUVs) is the core means of acquiring details of the surface of structures. However, due to the presence of a large number of suspended particles (such as silt and plankton) in water, light undergoes severe scattering and absorption effects during transmission. Among these effects, backscattered light caused by particles forms superimposed high-brightness background noise on the imaging plane, resulting in a significant decrease in image contrast. This makes it difficult for operators to identify key target information such as cracks and attached organisms, severely limiting the operational accuracy and safety of underwater robots.

[0003] Existing technologies typically employ dehazing algorithms based on dark channel priors or purely data-driven deep learning models for image restoration. However, such methods have significant drawbacks when dealing with extremely turbid environments: traditional algorithms mainly map and adjust pixel intensity in the real domain, ignoring the perturbation effect of water scattering on the phase of light waves. This results in their inability to repair edge blurring and texture fuzziness caused by wavefront distortion. Furthermore, they are prone to contrast overload or loss of detail when processing non-uniform background light. Therefore, it is necessary to design a visual enhancement method and system for underwater robots in turbid waters to solve the above problems. Summary of the Invention

[0004] In view of the above-mentioned prior art, this application is made. Embodiments of this application provide a method and system for visual enhancement of underwater robots in turbid waters, which corrects edge blurring and texture blurring caused by wavefront phase disturbance, and significantly improves the edge steepness of the outline of structures in the image.

[0005] According to one aspect of this application, a method for visual enhancement of underwater robots in turbid waters is provided, comprising:

[0006] The system acquires the original color image sequence collected in real time by the underwater robot, maps the pixel intensity of the original color image sequence to complex space, and constructs a complex domain analytic signal; it extracts the multi-directional spatial gradient tensor of the original color image, and calculates the pseudo polarization information and environmental characterization parameters accordingly.

[0007] A virtual phase compensation operator is constructed using the pseudo polarization degree information and the environmental characterization parameters; wherein, the compensation intensity of the virtual phase compensation operator is driven by the pseudo polarization degree information, and the compensation intensity is locally weighted and corrected using the environmental characterization parameters;

[0008] In the complex space, the virtual phase compensation operator is used to perform phase rotation compensation on the complex domain analytic signal, and the amplitude of the complex domain analytic signal is suppressed simultaneously according to the pseudo polarization degree information to obtain an enhanced complex domain analytic signal.

[0009] The modulus is extracted from the enhanced complex domain analytical signal. Based on the energy distribution characteristics of the blue and green channels in the original color image, the energy attenuation ratio of the red channel in the current water area is dynamically calculated. Based on this, a gain compensation coefficient is generated, and the enhanced image is output.

[0010] According to another aspect of this application, a visual enhancement system for underwater robots in turbid waters is provided, comprising:

[0011] The complex construction module is used to acquire the original color image sequence collected in real time by the underwater robot, map the pixel intensity of the original color image sequence to the complex space, construct the complex domain analytic signal, extract the multi-directional spatial gradient tensor of the original color image, and calculate the pseudo polarization information and environmental characterization parameters accordingly.

[0012] An operator generation module is used to construct a virtual phase compensation operator using the pseudo polarization degree information and the environmental characterization parameters; wherein, the pseudo polarization degree information is used to drive the compensation intensity of the virtual phase compensation operator, and the environmental characterization parameters are used to perform local weight correction on the compensation intensity;

[0013] The complex domain compensation module is used to perform phase rotation compensation on the complex domain analytic signal in the complex space using the virtual phase compensation operator, and simultaneously suppress the amplitude of the complex domain analytic signal according to the pseudo polarization information to obtain an enhanced complex domain analytic signal.

[0014] The energy gain output module is used to extract the modulus of the enhanced complex domain analytical signal. Based on the energy distribution characteristics of the blue and green channels in the original color image, it dynamically calculates the energy attenuation ratio of the red channel in the current water area, and generates a gain compensation coefficient accordingly to output the enhanced image.

[0015] According to another aspect of this application, an electronic device is provided, including a memory and a processor, the memory being used to store computer-executable instructions, and the processor being used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method described above.

[0016] According to another aspect of this application, a computer storage medium is provided that stores computer-executable instructions thereon, which, when executed by a processor, implement the steps of the method described above.

[0017] Compared with existing technologies, the underwater robot visual enhancement method and system for turbid waters according to the embodiments of this application utilizes a virtual phase compensation operator to perform phase rotation compensation on the analytical signal. This directly offsets the wavefront phase distortion caused by water scattering on light wave transmission, correcting edge blurring and texture blurring caused by wavefront phase disorder from a physical mechanism perspective, and significantly improving the edge steepness of the structure contours in the image. While suppressing the amplitude of the complex domain signal to eliminate backscattering components, the precise intervention of local weights effectively avoids the contrast overload problem that is prone to occur when processing non-uniform background light, ensuring that high-magnification denoising is achieved in strong scattering areas while the weak texture details in low scattering areas are completely preserved. It can scientifically restore the red light component, suppressing underwater "green bias" or "blue bias" phenomena while ensuring the naturalness and realism of color restoration, and avoiding image noise amplification caused by overcompensation. Attached Figure Description

[0018] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 This is a flowchart illustrating the underwater robot's visual enhancement method for turbid waters according to the present invention.

[0020] Figure 2 This is a schematic diagram of the overall logic of the underwater robot's visual enhancement method for turbid waters according to the present invention.

[0021] Figure 3 This is an extended schematic diagram of the underwater robot visual enhancement method for turbid waters according to the present invention. Detailed Implementation

[0022] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0023] Example 1:

[0024] In existing technologies, underwater visual enhancement techniques have evolved from traditional computer vision algorithms to physical imaging model compensation, but they still face significant challenges in scenarios with extremely high turbidity and non-uniform scattering. For raw images acquired by underwater robots, suspended particles in the water not only significantly attenuate light intensity, but also cause severe distortion of the wavefront phase due to random collisions of light waves during transmission. Traditional color balancing or contrast stretching methods based on the real domain often ignore the perturbation effect of water scattering on the light wave phase, making it impossible to repair edge blurring and texture blurring caused by wavefront distortion. Furthermore, they are prone to contrast overload or loss of detail when processing non-uniform background light, seriously affecting the accuracy of underwater target recognition and precision operations.

[0025] To address the aforementioned issues, this application leverages the ability of complex domain analytic signals to effectively characterize the amplitude and phase evolution of light waves. Based on this, it proposes a technical approach combining complex space mapping, virtual phase compensation operator construction, and cross-channel energy dynamic inversion. By mapping the original image sequence to complex space and extracting pseudo-polarization degree and environmental characterization parameters, the parameters drive the compensation operator to synchronously perform phase rotation correction and amplitude suppression in the complex domain. Furthermore, using the blue and green channels as a physical benchmark, it dynamically compensates for red light attenuation, thereby correcting edge blurring and restoring true colors through a physical mechanism. This provides a robust visual perception foundation for the automated operation of underwater robots.

[0026] Reference Figures 1-3 As an embodiment of the present invention, a method for enhancing the vision of underwater robots in turbid waters is provided, comprising: S1-S5.

[0027] To facilitate understanding, we will use a real-world scenario of an underwater robot (ROV) conducting underwater inspections of a port dam as an example. Assume the ROV's optical imaging system acquires a single frame of raw color image in real time. The image resolution is 640×480 pixels, containing three color channels: red (R), green (G), and blue (B), with pixel intensity values ​​ranging from 0 to 255 for each channel. Given the high turbidity of the current water body, which contains suspended particles such as silt, the acquired image exhibits an overall greenish tint, and the edges of the target structure (cracks on the dam surface) are blurred with low contrast.

[0028] Figure 1 The illustration shows a method for enhancing the vision of an underwater robot in turbid waters according to an embodiment of this application, specifically including:

[0029] like Figure 1 As shown, in step S1: the original color image sequence collected in real time by the underwater robot is acquired, the pixel intensity of the original color image sequence is mapped to the complex space, and a complex domain analytic signal is constructed; the multi-directional spatial gradient tensor of the original color image is extracted, and the pseudo polarization information and environmental characterization parameters are calculated accordingly.

[0030] In this step, acquiring the raw color image sequence collected in real time by the underwater robot refers to reading multiple frames of continuously acquired color images from the optical imaging system on the underwater robot. Each frame contains pixel intensity information for the red, green, and blue color channels. The raw color image sequence can be several frames acquired over a continuous period of time, or it can contain only a single frame at the current moment. In subsequent processing, the pixel intensity of each color channel is used as input data.

[0031] Mapping the pixel intensity of the original color image sequence to complex space to construct a complex-domain analytic signal involves mathematically transforming the pixel intensity values ​​of each color channel in the original image to extend them from the real domain to the complex domain, thus simultaneously incorporating amplitude and phase information. Specifically, a Hilbert transform can be applied along the spatial direction to the two-dimensional pixel intensity distribution of each color channel. The original real-valued pixel intensity is taken as the real part of the complex signal, and the result obtained after the Hilbert transform is taken as the imaginary part of the complex signal. The combination of the two constitutes the complex-domain analytic signal. The reason for using this complex-domain analytic signal representation is that water scattering not only attenuates the intensity of light waves (corresponding to amplitude information) but also causes random distortion of the wavefront phase (corresponding to phase information). Traditional image processing methods based on the real domain can only manipulate pixel intensity values ​​and cannot address phase-level distortions. However, by mapping the image to complex space, amplitude and phase are explicitly represented separately, thus providing a data foundation for subsequent simultaneous phase correction and amplitude adjustment in the complex domain. It should be noted that the Hilbert transform is a commonly used method for constructing analytic signals. Those skilled in the art can also use other equivalent methods for constructing analytic signals (such as multidimensional analytic signal construction methods based on the Riesz transform), and this application does not limit this.

[0032] Extracting the multi-directional spatial gradient tensor from the original color image refers to calculating the spatial rate of change of pixel intensity along multiple preset directions (such as horizontal, vertical, and two diagonal directions) in the spatial domain of the original color image, and organizing the gradient information of each direction into a tensor form. The purpose of the multi-directional spatial gradient tensor is to capture the variation pattern of pixel intensity in the image from different spatial orientations, so that the subsequent analysis of scattering degradation features is not limited to a single direction, but can cover structural information in multiple spatial orientations. It should be noted that the specific number and angle distribution of the "multi-directional" directions can be set according to the actual application scenario. For example, spatial gradients can be calculated along four directions: 0°, 45°, 90°, and 135°, and the number of directions can be increased or adjusted as needed.

[0033] The pseudo-polarization degree information and environmental characterization parameters are calculated based on this. This means that by using the gradient information in each direction carried by the multi-directional spatial gradient tensor, two types of characteristic parameters are obtained through a specific calculation method:

[0034] Among them, pseudo polarization information borrows the concept of optical polarization to describe the degree of signal degradation at each pixel location in the image after being scattered and degraded by water. In physical optics, the degree of polarization reflects the polarization state of light waves. The "pseudo-polarization degree" in this application is not obtained through actual measurement using optical devices such as polarizers, but is indirectly estimated based on the multi-directional spatial gradient information of the image. Therefore, it is called "pseudo-polarization degree". Specifically, at a certain pixel position in the image, if the position is in a region heavily affected by scattering (e.g., a region with strong backscattered light superposition), the gradient magnitudes in each direction tend to be consistent and generally low (because scattering makes the image structure in this region blurry and uniform). At this time, the calculated pseudo-polarization degree value is high, indicating that the signal degradation at this position is severe. Conversely, if the position is in a region less affected by scattering (e.g., at the clear edge of the target object), the gradient magnitudes in each direction have more obvious differences (because the edge of the structure has a stronger gradient response in a specific direction). At this time, the pseudo-polarization degree value is low. The pseudo-polarization degree information is used in subsequent steps to drive the compensation intensity of the virtual phase compensation operator, so that the region with more severe scattering degradation receives a greater compensation force.

[0035] Environmental characterization parameters are used to quantify the intensity of background noise (such as non-uniform background light and backscattered light) at each pixel location. These parameters are also calculated using a multi-directional spatial gradient tensor, and their physical meaning is to reflect the strength distribution of background noise interference experienced by different regions of the image. After normalization, the environmental characterization parameters are dimensionless; a larger value indicates stronger background noise interference in the corresponding region. In subsequent steps, these parameters are used to locally weight the compensation intensity of the virtual phase compensation operator, adapting the compensation process to the actual non-uniform scattering distribution in the image.

[0036] It should be noted that although both pseudo-polarization degree information and environmental characterization parameters are calculated from multi-directional spatial gradient tensors, they describe different physical dimensions: pseudo-polarization degree information focuses on describing the degree of signal degradation itself, while environmental characterization parameters focus on describing the intensity of environmental background interference. They each play different roles in subsequent steps.

[0037] Following the aforementioned underwater inspection example, the original color image (640×480 pixels, RGB three channels) acquired by the ROV is processed as follows: First, a Hilbert transform is applied to the pixel intensity of each color channel to obtain a complex domain analytic signal, at which point each pixel position has a corresponding amplitude value and phase value; then, the spatial gradient is calculated along the four directions of horizontal (0°), vertical (90°) and two diagonals (45°, 135°) to form a multi-directional spatial gradient tensor. Taking pixel A (coordinates (320, 240)) at the edge of a crack on the dam surface in the image as an example, the gradient response at the crack edge is stronger in the vertical direction and weaker in the horizontal direction, with clear differences in gradients in each direction. Therefore, the calculated pseudo-polarization degree value of this pixel is low (assumed to be 0.15), indicating that the signal degradation at this location is relatively mild. Taking pixel B (coordinates (100, 400)) in a murky background region far from the structure in the image as an example, this region is heavily affected by backscattering, and the gradient amplitudes in each direction tend to be consistent and generally low. Therefore, the calculated pseudo-polarization degree value of this pixel is high (assumed to be 0.72), indicating that the degradation at this location is severe. Meanwhile, the environmental characterization parameter value corresponding to pixel B (assumed to be 0.81 after normalization) is higher than that of pixel A (assumed to be 0.23 after normalization), reflecting that the background interference intensity in the region where pixel B is located is greater.

[0038] The above scheme maps the original image to complex space and extracts pseudo-polarization information and environmental characterization parameters, providing a data foundation for constructing compensation operators in the complex domain and performing differentiated phase correction and amplitude adjustment.

[0039] After constructing the complex-domain analytic signal and extracting the pseudo-polarization information and environmental characterization parameters, the method proceeds to the construction of a virtual phase compensation operator. The core of this step is to construct a compensation operator capable of performing phase correction on the analytic signal in the complex domain using the pseudo-polarization information and environmental characterization parameters obtained in S1. This allows different pixel locations to receive compensation levels adapted to their scattering degradation. The reason for constructing such a compensation operator is that wavefront phase distortion caused by water scattering is unevenly distributed across different regions of the image—regions with severe scattering exhibit large phase distortion, while regions with slight scattering exhibit small phase distortion. Therefore, an operator capable of adaptively adjusting the compensation intensity pixel-by-pixel is needed, rather than applying a uniform compensation amount to the entire image. Compared to traditional methods that only adjust pixel intensity in the real domain, this step establishes the compensation operation in the phase dimension of the complex domain, enabling correction from the physical mechanism of wavefront distortion. Alternative solutions include: uniformly compensating the entire image with a fixed phase offset, or using an adaptive contrast enhancement method based on the image's gray-level statistical features; however, the former cannot adapt to the differences in the degree of degradation in different regions in non-uniform scattering scenes, while the latter is still limited to the real domain and cannot reach the correction of phase distortion.

[0040] return Figure 1 In step S2: a virtual phase compensation operator is constructed using pseudo polarization degree information and environmental characterization parameters; wherein, the compensation intensity of the virtual phase compensation operator is driven by pseudo polarization degree information, and the compensation intensity is locally weighted and corrected using environmental characterization parameters.

[0041] Specifically, constructing a virtual phase compensation operator includes:

[0042] Based on the pseudo-polarization degree information, a phase compensation reference quantity that varies with the positive direction of the pseudo-polarization degree is generated;

[0043] In this step, the phase compensation reference amount refers to the reference phase offset initially determined based on the pseudo-polarization degree information, used to correct wavefront phase distortion at each pixel location, and its unit is radians (rad). "Positive variation with pseudo-polarization degree" means that for pixel locations with higher pseudo-polarization degree values ​​(i.e., more severe signal degradation), the corresponding phase compensation reference amount is larger, meaning a larger phase correction amplitude needs to be applied; conversely, for pixel locations with lower pseudo-polarization degree values, the phase compensation reference amount is correspondingly smaller. The positive variation relationship between the phase compensation reference amount and pseudo-polarization degree can be achieved using linear mapping, piecewise linear mapping, or a monotonically increasing function, etc. This application does not limit the specific form of the mapping function, as long as it satisfies the positive variation relationship. It should be noted that the phase compensation reference amount is only a preliminary estimate of the compensation amount; it does not yet consider the differences in environmental background interference in different regions of the image, therefore, it still needs to be locally corrected using a subsequent spatial weight distribution function.

[0044] Constructing a spatial weight distribution function using environmental characterization parameters includes:

[0045] Extract local gradient consistency features from the original color image sequence;

[0046] In this step, local gradient consistency refers to the degree of consistency in the distribution of gradient directions among pixels within a local spatial neighborhood of an image (e.g., a preset window centered on a certain pixel). Specifically, within the local neighborhood of a pixel, if the neighborhood is in a structurally clear region (such as an object edge), the gradient directions of the pixels within the neighborhood tend to be consistent or concentrated, resulting in a high local gradient consistency feature value. Conversely, if the neighborhood is in a blurred region affected by scattering interference or a region with messy textures, the gradient directions of the pixels within the neighborhood are scattered and inconsistent, resulting in a low local gradient consistency feature value. Local gradient consistency can be quantified by calculating the distribution entropy of gradient directions within the local neighborhood: the more concentrated the gradient direction distribution, the lower the distribution entropy, and the higher the consistency feature value; conversely, the more scattered the gradient direction distribution, the higher the distribution entropy, and the lower the consistency feature value. It should be noted that the window size of the local neighborhood can be set according to the image resolution and actual needs; for example, a 5×5 or 7×7 pixel window can be used.

[0047] Then, the local gradient consistency features are spatially mapped to the environmental characterization parameters to obtain the structural confidence factor of each pixel, as shown in the following formula:

[0048] ;

[0049] in, This represents the structural confidence factor at pixel (x, y), reflecting the degree to which this point represents valid structural information. This represents the local gradient consistency feature at pixel (x, y), extracted from the gradient direction distribution entropy of the original color image sequence in the spatiotemporal neighborhood, reflecting the stability of the image structure. This represents the environmental characterization parameter (a normalized dimensionless parameter) corresponding to the pixel (x,y), quantifying the intensity of interference from environmental background noise (such as non-uniform background light) in this region.

[0050] In this formula, the physical meaning of the structural confidence factor is: when the local gradient consistency at a certain pixel location is high (i.e., (relatively large) and the environmental background interference is relatively weak (i.e.) When it is smaller, A larger value indicates that the image information at that location is more likely to be valid structural information than scattering noise; conversely, a smaller value indicates that the image information at that location is more likely to be valid structural information than scattering noise. A smaller value indicates that the image information at that location is more likely to be dominated by scattering noise. It should be noted that in actual calculations, to avoid the denominator... In the case of zero, a very small positive constant can be added to the denominator (e.g., () is used as a numerically stable term.

[0051] Based on the ratio of the structural confidence factor to the pseudo polarization information, the pixel targets whose structural confidence factor is in the preset advantage range are determined, and the gain ratio of the phase compensation reference quantity corresponding to the pixel targets is reduced to construct the spatial weight distribution function.

[0052] In this step, the preset dominance interval refers to the range of values ​​where the structure confidence factor is relatively high. Pixels within this interval are considered to have relatively reliable structural information and do not require excessive phase compensation. The preset dominance interval is determined by statistically analyzing the structure confidence factors of all pixels in the image and taking the upper quartile (i.e., the 75th percentile) of its distribution as the lower bound threshold of the dominance interval. ,Will The interval is defined as the preset dominant interval. When a certain pixel... When the value is within this advantageous range, it indicates that the structural information of the pixel has a high confidence level. Applying excessive phase compensation to it may introduce unnecessary interference. Therefore, the gain ratio of the phase compensation reference value corresponding to this pixel should be reduced; conversely, Pixels whose values ​​are not in the dominant range have lower structural information confidence and more severe scattering degradation; their gain ratio should be maintained or increased. "Ratio relationship" refers to comparing the structural confidence factors at pixel (x,y). With pseudo-polarization information The relative magnitude determines the direction of spatial weight adjustment: when Compared to When in a dominant position, reduce the weight; when Compared to When in a dominant position, the weights are maintained or increased. It should be noted that the lower bound threshold of the preset dominant interval is not limited to the upper quartile value; other statistical quantile values ​​or fixed thresholds can also be used depending on the specific application scenario. This application does not impose any limitations on this. The spatial weight distribution function constructed in the above manner... The value ranges from 0 to 1, and the larger the value, the higher the compensation weight required at that position.

[0053] By adjusting the gain of the phase compensation reference value pixel by pixel using a spatial weight distribution function, differentiated phase correction is performed on different regions while preserving edge details, resulting in the locally corrected phase offset. The specific formula is as follows:

[0054] ;

[0055] in, This represents the phase offset (rad) at pixel (x,y) after local correction. This represents the spatial weight distribution function value at pixel (x, y) constructed from environmental characterization parameters, reflecting the weight required for phase compensation at that point. This represents the phase compensation reference quantity (rad) generated based on the pseudo polarization degree information, which varies in the positive direction.

[0056] The physical meaning of this formula is as follows: For each pixel location, the phase compensation reference value driven by pseudo-polarization degree is used as the initial estimate for compensation, and then the initial estimate is locally adjusted through the spatial weight distribution function—in regions with high structural confidence and weak environmental interference, the spatial weight distribution function value at pixel (x,y) is constructed from environmental characterization parameters. The smaller the value, the smaller the final phase shift, thus avoiding overcompensation for already clear areas; in areas with low structural confidence and strong environmental interference, the spatial weight distribution function value at pixel (x,y) is constructed from environmental characterization parameters. The phase shift is relatively large, and the final phase shift remains at a high level to adequately correct the wavefront distortion in this region.

[0057] A complex-domain rotation operator is constructed using the locally corrected phase offset to offset wavefront distortion in complex-domain analytic signals, serving as a virtual phase compensation operator. The specific formula is as follows:

[0058] ;

[0059] in, For virtual phase compensation operators, It represents the imaginary unit.

[0060] In this formula, The virtual phase compensation operator, which is a unit modulus rotation factor in the complex domain, has the physical meaning that when this operator is multiplied by a complex domain analytic signal, it only changes the phase (rotates) of the complex domain analytic signal. It adjusts the phase of the wavefront in radians without changing its amplitude, thus achieving directional offsetting of wavefront phase distortion. It is called a "virtual" phase compensation operator because it is not based on physical phase correction generated by optical hardware (such as deformable mirrors in adaptive optics systems), but on an equivalent correction operator constructed in the digital domain based on image features.

[0061] Following the aforementioned underwater inspection example, for pixel point A (coordinates (320, 240), pseudo-polarization degree P = 0.15, environmental characterization parameters... Assuming the phase compensation reference quantity The result obtained through linear mapping is: rad; its local gradient consistency characteristics:

[0062] Structural confidence factor Assuming the upper quartile threshold after statistical analysis of the entire graph. If the pixel is within the preset dominant range, then the corresponding spatial weight... Reduced, assuming a The phase offset after local correction rad, the corresponding virtual phase compensation operator:

[0063] For pixel B (coordinates (100, 400), the normalized pseudo-polarization degree value P = 0.72, and the environmental characterization parameters... =0.81), assuming rad; its , well below the threshold Not in the dominant region, spatial weight distribution function value Maintain a high value, assuming ,but rad, corresponding It can be seen that pixel B, which suffers from severe scattering degradation, receives a greater amount of phase compensation.

[0064] By using the above scheme, a virtual phase compensation operator that can be adaptively adjusted pixel by pixel is constructed by using pseudo polarization degree information to drive the compensation intensity and using environmental characterization parameters for local weight correction, providing a correction tool for subsequent phase rotation compensation in the complex domain.

[0065] After constructing the virtual phase compensation operator, the method proceeds to the signal compensation stage in the complex domain. The core of this stage lies in using the virtual phase compensation operator constructed in S2 to simultaneously perform phase rotation compensation and amplitude suppression on the analytic signal in the complex space. The reason for simultaneous phase rotation compensation and amplitude suppression is that the degradative effect of water scattering on the image manifests in two dimensions: firstly, scattering causes random phase distortion of the light wavefront, resulting in edge blurring and texture blurring on the imaging plane (corresponding to phase dimension degradation); secondly, backscattered light forms superimposed high-brightness background noise on the imaging plane, raising the signal's amplitude floor (corresponding to amplitude dimension degradation). Therefore, performing phase correction without amplitude suppression, while improving edge sharpness, still results in insufficient contrast due to background noise; performing amplitude suppression without phase correction, while reducing background noise, still leaves edge blurring and texture blurring. Performing both simultaneously helps to improve image clarity while suppressing background noise. Alternative solutions include performing denoising in the amplitude domain separately (such as wavelet thresholding, nonlocal mean filtering, etc.), but these methods only affect the signal amplitude and do not involve phase correction, thus having limited improvement on blur caused by wavefront distortion.

[0066] return Figure 1 In step S3: In the complex space, phase rotation compensation is performed on the complex domain analytic signal using a virtual phase compensation operator, including:

[0067] The virtual phase compensation operator is used to perform inverse phase mapping on the complex domain analytic signal, and the spatial phase distribution of the complex domain analytic signal is corrected to the preset coherent reference plane.

[0068] In this step, inverse phase mapping refers to applying the virtual phase compensation operator... When performing complex multiplication with analytic signals in the complex domain, since The phase shift direction is opposite to the phase distortion direction caused by scattering (i.e., "inverse"), so the effect of this multiplication operation is to correct the phase components in the analytical signal that have been disrupted by scattering back to the target reference state.

[0069] The pre-defined coherent reference plane refers to the reference state in which the spatial phase distribution of each pixel position on the imaging plane should be under ideal conditions (i.e., when the light wave is transmitted without scattering distortion). In physical optics, when a light wave propagates in a non-scattering medium, the wavefront reaching the imaging plane should remain coherent, that is, the phase distribution of each point satisfies certain spatial continuity and consistency conditions, and the phase distribution plane corresponding to this state is the coherent reference plane. The methods for determining the pre-defined coherent reference plane include, but are not limited to: (1) during the deployment phase, using the same optical imaging system to perform calibration acquisition on the same scene under clear water or low turbidity conditions, and using the phase distribution of the complex domain analytic signal of the calibration image as the coherent reference plane; (2) when the calibration image cannot be obtained, the phase distribution of the complex domain analytic signal corresponding to the frame with the smallest phase distribution variance in the original color image sequence can be selected as an approximate coherent reference plane; (3) the ideal phase distribution derived from the theoretical model can also be used as a reference. This application does not limit the specific method for determining the coherent reference plane.

[0070] In the coherent reference plane, extract the first and second orthogonal components of the complex domain analytic signal;

[0071] In this step, the first and second dimension components correspond to the real and imaginary components of the phase-corrected complex-domain analytic signal, respectively. After correcting the phase distribution of the analytic signal to the coherent reference plane, the real and imaginary components are in an orthogonal reference coordinate system, each carrying information about different dimensions of the signal.

[0072] The attenuation weight, determined based on pseudo-polarization information, is calculated using the following formula:

[0073] ;

[0074] in, For decay weights, The environmental extinction coefficient reflects the sensitivity of the current water body to polarization characteristics. This represents the pseudo-polarization degree information at pixel (x,y) (the pseudo-polarization degree value after normalization, with a value range of 0~1).

[0075] In this formula, Dimensionless coefficients, decay weights The physical meaning is as follows: The larger the value (i.e.) The smaller the value, the less severe the scattering degradation. This indicates that the direct light component at this pixel location accounts for a higher proportion of the total received light, and the first dimension component (real part) should be given a greater retention weight. The smaller the value (i.e.) The larger the value, the more severe the scattering degradation. This indicates that the scattering component at this pixel location accounts for a relatively high proportion, and more energy should be redistributed from the first-dimensional component to the second-dimensional component. The attenuation weight adopts an exponential decay form rather than a nonlinear form because the attenuation law of the polarization characteristics of light when it propagates in water approximately follows an exponential law (similar to the Beer-Lambert law). Using an exponential form can make the weight allocation more consistent with the actual physical attenuation characteristics.

[0076] Environmental extinction coefficient It is a dimensionless parameter related to the current characteristics of the water body, and its value reflects the sensitivity of water turbidity to the attenuation of polarization characteristics. The determination method is as follows: pre-calibration is performed based on the current turbidity level of the water body. Specifically, the turbidity of the target water body can be measured before deployment, and corresponding values ​​can be preset according to different turbidity levels. Value range: For low turbidity waters (turbidity value less than 10 NTU (scattering turbidity units)), Smaller values ​​can be used, for example, a range of 0.5 to 1.5; for moderately turbid waters (turbidity values ​​between 10 and 50 NTU). Medium values ​​can be taken, for example, a range of 1.5 to 3.0; for high turbidity waters (turbidity value greater than 50 NTU). Larger values ​​can be selected, for example, a range of 3.0 to 5.0. In practical applications, the values ​​can also be determined based on the statistical characteristics of the image sequence (e.g., the mean and variance of the pseudo-polarization degree of the entire image). This application performs adaptive estimation. The specific method of determination is not limited.

[0077] The energy is redistributed between the first and second dimension components to reconstruct the coherent wavefront of the target object. The reconstruction process follows a linear combination mapping relationship:

[0078] ;

[0079] in, This represents the reconstructed coherent wavefront analytic signal in the complex domain. For decay weights, This represents the first-dimensional component (real part numerical channel) of a complex-domain analytic signal in the coherent reference plane. This represents the second-dimensional component (imaginary part numerical channel) of the complex domain analytic signal in the coherent reference plane.

[0080] The physical meaning of this formula is that, based on the severity of scattering degradation at each pixel location, the real and imaginary parts of the corrected analytical signal are allocated differently, so that the reconstructed coherent wavefront complex domain analytical signal is closer to the coherent wavefront state of the target object under scattering-free conditions.

[0081] The amplitude of the complex domain analytic signal is suppressed synchronously based on pseudo-polarization information, including:

[0082] The instantaneous energy envelope of the complex domain analytic signal at each pixel location is extracted based on pseudo-polarization degree information;

[0083] In this step, the instantaneous energy envelope refers to the magnitude of the complex domain analytic signal at each pixel location. It reflects the instantaneous total energy of the signal at that location (including the superposition of the direct light component and the backscattered component). The instantaneous energy envelope is extracted by calculating the complex magnitude of the complex domain analytic signal at each pixel location after phase rotation compensation, i.e., taking the square root of the sum of the squares of the real and imaginary parts. The pseudo-polarization information serves as a basis for determining the proportion of backscattered components at each pixel location—pixel locations with higher pseudo-polarization values ​​have a larger proportion of backscattered components in their instantaneous energy envelope, requiring stronger amplitude suppression.

[0084] The dynamic gain dynamic range compression of the instantaneous energy envelope is performed using pseudo-polarization degree information to obtain the compressed energy distribution. The magnitude of the complex domain analytic signal is then remapped based on the compressed energy distribution.

[0085] In this step, dynamic range compression refers to compressing the numerical range of the instantaneous energy envelope based on pseudo-polarization information. Specifically, for pixel locations with high pseudo-polarization values ​​(large scattering component proportion), the energy envelope value is reduced by a larger compression ratio, thereby suppressing high-brightness background noise caused by backscattering; for pixel locations with low pseudo-polarization values ​​(large direct light component proportion), the energy envelope value is retained by a smaller compression ratio to maintain the integrity of the target object signal. The compressed energy distribution is used to replace the original instantaneous energy envelope, remapping the magnitude of the complex domain analytic signal so that the magnitude of the compressed signal reflects the effective signal energy after suppression of the scattering component. It should be noted that the specific implementation of dynamic range compression can employ logarithmic compression, power-law compression, piecewise linear compression, etc., and this application does not limit this approach.

[0086] The enhanced complex domain analytic signal is obtained.

[0087] In this step, the enhanced complex domain analytic signal refers to the complex signal obtained after two operations: phase rotation compensation and amplitude suppression. Its phase distribution has been corrected to the coherent reference plane, and the backscattering component in the amplitude has been suppressed.

[0088] Using the aforementioned underwater inspection example, assuming the current water turbidity is 35 NTU (moderate turbidity), the corresponding environmental extinction coefficient... The value is set to 2.0. For pixel A (P=0.15), the attenuation weight is... That is, the real component of the point retains approximately 74.1% of the weight, while the imaginary component receives approximately 25.9% of the weight; for pixel B (P=0.72). This means that the real component of the signal at this point retains only about 23.7% of the weight, while the imaginary component receives about 76.3% of the weight. This implies a significant adjustment in the energy allocation of pixel B, which suffers from severe scattering degradation. Simultaneously, in the amplitude suppression stage, pixel B, due to its higher pseudo-polarization degree, experiences a larger compression ratio on its instantaneous energy envelope, suppressing the high-brightness background caused by backscattering; while pixel A, due to its lower pseudo-polarization degree, has its energy envelope well preserved. After these processing steps, the enhanced complex domain analytic signal is obtained.

[0089] The above scheme simultaneously achieves phase rotation compensation and amplitude suppression in the complex domain, which helps to reduce backscattering noise while improving edge sharpness and texture clarity.

[0090] After completing phase rotation compensation and amplitude suppression in the complex domain, the method proceeds to signal conversion and color restoration. The core of this stage is to convert the enhanced complex domain analytic signal back to the real domain image signal and compensate for the energy attenuation of the red channel in underwater environments to restore color representations closer to real-world scenes. The reason for separate energy compensation for the red channel is that water selectively absorbs different wavelengths of light—the absorption coefficient of red light in water is much greater than that of blue and green light, resulting in underwater images generally exhibiting a bluish or greenish tint. Using the energy distribution of the blue and green channels as a reference to infer the degree of red light attenuation leverages the physical principle of differentiated attenuation of different wavelengths of light in water within the same scene. Compared to independently adjusting the color balance of the RGB channels, this method helps to make the color restoration results closer to real-world scenes; furthermore, the use of a logarithmic gain compensation coefficient helps to avoid amplifying noise due to overcompensation when red light attenuation is significant.

[0091] return Figure 1 In step S4: the modulus is extracted from the enhanced complex domain analytical signal. Based on the energy distribution characteristics of the blue-green channels in the original color image, and utilizing the physical property that red light in water exhibits exponential attenuation with increasing depth and turbidity, the energy attenuation ratio of the red light channel in the current water area is dynamically calculated. The specific formula is as follows:

[0092] ;

[0093] in, This indicates the proportion of energy attenuation in the red light channel at the current water level, used to quantify the degree of loss of the red light component. , and These represent the average pixel energy distribution characteristics of the blue, green, and red color channels in the original color image sequence, respectively. This represents the underwater spectral correlation correction factor, used to balance the differences in spectral absorption selectivity between different water qualities (such as seawater and freshwater);

[0094] In this formula, The physical meaning is: using the average pixel energy distribution characteristics of the blue and green channels as a reference, calculate the attenuation factor of the red channel relative to this reference. When red light attenuation is severe, The value is low. A larger value indicates that greater gain compensation is needed; when red light attenuation is relatively minor, The value is close to 1.

[0095] underwater spectral correlation correction factor The determination method is as follows: pre-calibration is performed based on the water quality type. Different water qualities have different selective characteristics in the absorption of different bands of the spectrum: dissolved salts in seawater absorb blue light relatively weakly but absorb red light strongly, while the presence of organic matter, silt and other suspended matter in freshwater makes the absorption characteristics of each band different. Its function is to correct for the effects of the aforementioned water quality differences; specifically, for marine environments, The typical value range is 0.85~1.01; for freshwater environments (such as rivers and reservoirs). The typical value range is 1.01~30; for mixed water quality (such as estuary areas). Transitional values ​​from the two ranges mentioned above can be taken. In practical applications, Alternatively, before deployment, the color can be determined by photographing and calibrating a calibration board with known colors in the target waters. This application addresses... The specific method of determination is not limited.

[0096] Based on the energy attenuation ratio of the red light channel in the current water area, a gain compensation coefficient is generated. The specific formula is as follows:

[0097] ;

[0098] The reason why the gain compensation coefficient is generated in logarithmic form in this formula, instead of directly using logarithmic form, is... As a compensation, it exists because the logarithmic function has the property of compressing large values ​​and retaining small values: when When the red light attenuation is significant (i.e., severe red light attenuation), The growth rate slows down, avoiding the problem of noise being amplified synchronously due to an excessively large gain compensation coefficient; when When smaller, The compensation amount is approximately linearly related to the attenuation ratio, ensuring compensation accuracy even with small attenuations. Gain compensation coefficient. It is then applied to the red light channel to compensate for the pixel values ​​of the red light channel, thereby improving the bluish or greenish hue of the underwater image.

[0099] Output enhanced image; in this step, output enhanced image refers to: extracting the modulus value from the enhanced complex domain analytic signal, mapping the modulus value back to the pixel intensity space of each color channel, and using gain compensation coefficients. After energy compensation of the red light channel, the final enhanced color image is obtained and output.

[0100] Continuing with the aforementioned underwater inspection example, let's assume the average pixel energy distribution characteristics of the blue channel in the original color image. Average pixel energy distribution characteristics of the green channel Average pixel energy distribution characteristics of the red channel The current water area is a freshwater environment (reservoir dam). Underwater spectral correlation correction factor. Take 1.15. Then the energy attenuation ratio of the red light channel is... Gain compensation coefficient The gain compensation coefficient is applied to the red light channel to compensate for the gain of the red light pixel values. After the above processing, the output enhanced image shows improvements in edge sharpness and color reproduction compared to the original image.

[0101] The above scheme uses the energy distribution of the blue-green channels as a physical benchmark to dynamically calculate the degree of red light attenuation, and uses a logarithmic gain compensation coefficient to restore red light energy, which helps to improve color performance while suppressing noise amplification introduced by overcompensation.

[0102] Based on the above S1-S4, the main process of the underwater robot turbid water vision enhancement method of this application has been completed. However, the above main process still has room for further optimization in the following aspects: On the one hand, when constructing the virtual phase compensation operator in S2, its gain value is determined only based on pseudo polarization information and environmental characterization parameters, without considering the influence of the spatial distribution difference of contrast features between different color channels on the compensation intensity. In the water scattering environment, the contrast distribution of the blue channel and the green channel often differs in space due to their different wavelengths. This difference contains channel selectivity information of scattering degradation. If it can be incorporated into the gain adjustment of the compensation operator, it will help to make the compensation intensity more consistent with the actual degradation status of each pixel position. On the other hand, in the complex domain analytical signal after phase rotation compensation, the signal components at different frequency scales may have scattering noise and effective texture mixing, and the contrast evolution trend of the image in the time domain sequence also reflects whether the enhancement effect has converged sufficiently. Therefore, it is necessary to introduce multi-scale frequency domain filtering and a detail enhancement mechanism based on contrast change gradient. Based on the above considerations, such as Figure 2 As shown, this application further proposes that after performing phase rotation compensation on the complex domain analytic signal, it also includes:

[0103] The signal contrast features of each color channel in the original color image are extracted. Based on the spatial distribution differences of the signal contrast features among different color channels, a compensation intensity adjustment factor is generated. The specific formula is as follows:

[0104] ;

[0105] in This represents the compensation intensity adjustment factor at pixel (x, y), reflecting the spatial distribution differences in contrast among different color channels. and These represent the local signal contrast characteristics of the blue and green channels at pixel (x,y) in the original color image, respectively. Indicates taking the absolute value;

[0106] In this formula, The physical meaning is that when the local contrast difference between the blue channel and the green channel is large (indicating that the color attenuation in that area has a strong selectivity). A value approaching 2 indicates that the gain value of the virtual phase compensation operator needs significant correction; when the contrast difference between the two is small, As the value approaches 1, the gain of the virtual phase compensation operator remains essentially unchanged. It should be noted that in actual calculations, to avoid the denominator... If the result is zero, a very small positive constant can be added to the denominator (e.g., ...). () is used as a numerically stable term.

[0107] The gain value of the virtual phase compensation operator is then corrected using a compensation intensity adjustment factor.

[0108] ;

[0109] in, This represents the final gain value of the corrected virtual phase compensation operator. This represents the operator gain value, driven by pseudo-polarization degree information and initially corrected by environmental characterization parameters.

[0110] Multi-scale frequency domain filtering is performed on the complex domain analytic signal to extract the high-frequency components in the preset frequency range, and the amplitude of the high-frequency components is reconstructed by weighting based on the pseudo-polarization information.

[0111] In this step, multi-scale frequency domain filtering refers to filtering and decomposing the complex domain analytic signal at multiple different frequency scales in the frequency domain space to separate signal components within different frequency ranges. The preset frequency range refers to the frequency range used to extract high-frequency components, and its upper and lower bounds are determined based on the image resolution and target texture scale. Specifically, for an image with a resolution of 640×480 pixels, the maximum spatial frequency (Nyquist frequency) is half the number of pixels in the image width direction, i.e., 320 cycles / image. The lower bound of the preset frequency range can be taken as a certain proportion of the Nyquist frequency (e.g., 30%~50%), and the upper bound is taken as the Nyquist frequency or a value close to the Nyquist frequency. That is, the preset frequency range can be set within the range of 30% to 100% of the Nyquist frequency. Signal components within this frequency range correspond to detailed textures and edge features in the image.

[0112] "Weighted reconstruction of the amplitude of high-frequency components based on pseudo-polarization information" means that for regions with high pseudo-polarization values ​​(severe scattering degradation), their high-frequency components contain more scattering noise components, and the amplitude weight of these high-frequency components needs to be reduced to suppress noise; for regions with low pseudo-polarization values ​​(clearer structural information), their high-frequency components are mainly effective texture and edge information, and their amplitude weight should be retained or enhanced.

[0113] The contrast change gradient of the enhanced complex domain analytic signal in the time domain sequence is obtained. The image detail enhancement operator is dynamically mapped based on the contrast change gradient, and the image detail enhancement operator is used to perform pixel-level mapping on the image signal after the modulus is extracted.

[0114] In this step, the contrast change gradient refers to the rate of change of local contrast of the enhanced complex domain analytic signal in the time domain sequence (i.e., between adjacent frames), reflecting the contrast evolution trend of the image sequence in the time dimension. The image detail enhancement operator is a pixel-level mapping function dynamically generated based on the contrast change gradient, used to further enhance the details of the image signal after the modulus is extracted: regions with large contrast change gradients (rapid contrast changes) indicate that there may be details in the region that have not been fully recovered, and the image detail enhancement operator applies a stronger enhancement mapping to the region; regions with small contrast change gradients (stable contrast) indicate that the enhancement in the region has converged, and a weaker enhancement mapping is applied to avoid overprocessing.

[0115] In actual underwater imaging scenarios, the signal-to-noise ratio carried by different frequency components varies, and the attenuation of underwater ambient light with depth also affects the compensation intensity required by the virtual phase compensation operator. In addition, the uniformity of contrast distribution in local spatial regions of the image also reflects the richness of texture information in that region, and regions with poor texture may require additional texture compensation. Therefore, this application further proposes the following preferred schemes to supplement and optimize in three dimensions: frequency domain gain adjustment, depth adaptive calibration, and spatial texture compensation.

[0116] This application further proposes that, after performing multi-scale frequency domain filtering, it also includes:

[0117] The energy distribution characteristics of each frequency component in the complex domain analytic signal are extracted, and a gain correlation is established between the energy distribution characteristics and the enhanced complex domain analytic signal. Based on the gain correlation, the gain coefficient of each frequency component is dynamically corrected. The specific formula is as follows:

[0118] ;

[0119] in, This represents the dynamic gain coefficient of the corrected frequency component f. Indicates the reference frequency gain value. This represents the characteristic value of the energy distribution at frequency component f. This represents the average energy eigenvalue of all frequency components. This represents the modulus energy of the enhanced complex-domain analytic signal;

[0120] In this formula, "gain correlation" refers to the dynamic adjustment of the gain coefficient of a frequency component based on its energy distribution characteristics and the relationship between the overall signal energy: when the energy of a certain frequency component... Above average energy When the exponent term is greater than 1, the gain coefficient is amplified to enhance the signal component corresponding to that frequency; when Below When the exponent term is less than 1, the gain coefficient is reduced to suppress any noise that may exist in that frequency component. The reference frequency gain value is determined by: presetting it according to the energy distribution ratio that each frequency component should have in the undegraded image, or taking it as the initial uniform value of the gain of each frequency component (e.g., setting it to 1.0). This application does not limit this.

[0121] The attenuation gradient information of underwater ambient light in the depth dimension is obtained, and the compensation intensity of the virtual phase compensation operator is calibrated twice based on the attenuation characteristic curve determined by the attenuation gradient information.

[0122] In this step, the underwater ambient light attenuation gradient information in the depth dimension refers to the rate of change of underwater light intensity along the depth direction. Because the absorption and scattering of light by water accumulates with increasing depth, the underwater ambient light intensity exhibits an attenuation trend along the depth direction. The attenuation gradient information can be obtained by combining the current depth value measured by the depth sensor onboard the underwater robot with a preset light attenuation model, or it can be indirectly estimated by the overall brightness change trend in an image sequence. The attenuation characteristic curve determined based on the attenuation gradient information refers to fitting the light attenuation gradient along the depth direction into a continuous attenuation curve (e.g., an exponential attenuation curve or a piecewise linear attenuation curve), used to describe the degree of light attenuation at different depths. The meaning of using this attenuation characteristic curve to perform secondary gain calibration on the compensation intensity of the virtual phase compensation operator is: when the underwater robot is at a greater depth and the light attenuation is more severe, the compensation intensity of the virtual phase compensation operator is increased accordingly; when the depth is shallower and the light attenuation is less severe, the compensation intensity is appropriately reduced to match the actual degradation degree at the current depth.

[0123] The entropy value of the contrast distribution in a local region of the image signal is extracted using the following formula:

[0124] ;

[0125] in, This represents the contrast distribution entropy value within a local region. This represents the total number of contrast quantization levels. This represents the probability of the i-th level of contrast occurring within the local neighborhood.

[0126] In this formula, The total number of contrast quantization levels is determined by dividing the contrast values ​​in a local neighborhood of the image into equal intervals based on their magnitude. The system quantifies the number of levels and then calculates the frequency of each level as a probability estimate. The value can be set according to the dynamic range of the image and the required quantization precision, for example, it can be set to... or wait. The physical meaning is that if the contrast distribution in a local area is relatively uniform (the probability of each level appearing is similar), then the entropy value is high, indicating that the area contains rich texture information; if the contrast distribution is concentrated in a few levels, then the entropy value is low, indicating that the area has less texture information.

[0127] An adaptive filtering operator is generated based on the contrast distribution entropy value, and the specific formula is as follows:

[0128] ;

[0129] in, This represents the adaptive filtering operator generated at pixel (x, y), used to adjust the local texture compensation intensity. The contrast distribution entropy value represents the contrast distribution within a local region, reflecting the richness of feature information in that region. This represents the total number of contrast quantization levels. This represents the maximum theoretical entropy value, used as a normalization benchmark.

[0130] In this formula, The physical meaning is: the higher the contrast distribution entropy value (the richer the texture information) of a region, The smaller the value (approaching 0), the weaker the texture compensation should be to avoid over-enhancing existing rich textures; the lower the contrast distribution entropy value (the poorer the texture information) of a region, the better. The larger the value (closer to 1), the stronger the texture compensation is applied to make up for the missing details in that area.

[0131] An adaptive filtering operator is used to compensate for the spatial texture of the image signal.

[0132] In this step, an adaptive filtering operator is used. Compensating for the spatial texture of an image signal refers to... As a pixel-by-pixel gain adjustment weight, the texture component in the image signal is weighted and enhanced. Specifically, the image signal can be decomposed into a basis component (low-frequency overall brightness distribution) and a texture component (high-frequency detail information), and then... The texture components are weighted, and finally the weighted texture components are superimposed with the base components to obtain the image signal after spatial texture compensation.

[0133] Following the aforementioned underwater inspection example, in the further preferred scheme described above, it is assumed that the local signal contrast characteristics of the blue channel of pixel A (at the edge of the crack) are... Local signal contrast characteristics of the green channel ,but Pixel B (the muddy background area) , ,but It can be seen that at pixel B, where the color attenuation selectivity differs significantly, the compensation intensity adjustment factor... A larger value results in a greater correction to the gain value of the virtual phase compensation operator. For the calculation of the contrast distribution entropy value, it is assumed that… With a contrast level of 16, the contrast distribution in the local area where pixel A is located is relatively uniform. bit, then The region receives relatively weak texture compensation; the contrast distribution in the region where pixel B is located is concentrated at only a few levels. bit, then Stronger texture compensation is applied to this area.

[0134] In summary, this application constructs an analytical signal by mapping the original image to complex space, constructs a virtual phase compensation operator using pseudo-polarization information and environmental characterization parameters, synchronously performs phase rotation compensation and amplitude suppression in the complex domain, and dynamically compensates for red light energy attenuation using the blue-green channel as the physical reference. This forms a complete processing link from complex domain signal construction, phase-amplitude joint compensation to cross-channel color restoration, providing a feasible enhancement scheme for the visual perception of underwater robots in turbid waters.

[0135] Example 2:

[0136] This is one embodiment of the present invention, which differs from the previous embodiment in that:

[0137] The underwater robot's visual enhancement system for turbid water includes:

[0138] The complex construction module is used to acquire the original color image sequence collected in real time by the underwater robot, map the pixel intensity of the original color image sequence to the complex space, construct the complex domain analytic signal, extract the multi-directional spatial gradient tensor of the original color image, and calculate the pseudo polarization information and environmental characterization parameters accordingly.

[0139] The operator generation module is used to construct a virtual phase compensation operator using pseudo-polarization degree information and environmental characterization parameters; wherein, the pseudo-polarization degree information is used to drive the compensation intensity of the virtual phase compensation operator, and the environmental characterization parameters are used to perform local weight correction on the compensation intensity;

[0140] The complex domain compensation module is used to perform phase rotation compensation on the complex domain analytic signal in the complex space using a virtual phase compensation operator, and simultaneously suppress the amplitude of the complex domain analytic signal according to the pseudo polarization degree information to obtain an enhanced complex domain analytic signal.

[0141] The energy gain output module is used to extract the modulus of the enhanced complex domain analytical signal. Based on the energy distribution characteristics of the blue and green channels in the original color image, it dynamically calculates the energy attenuation ratio of the red channel in the current water area, and generates a gain compensation coefficient accordingly to output the enhanced image.

[0142] Example 3:

[0143] In one embodiment of the present invention, which differs from the previous embodiment, the electronic device includes one or more processors and a memory.

[0144] A processor can be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and can control other components in an electronic device to perform desired functions.

[0145] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0146] In one example, the electronic device may also include input devices and output devices, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown). In addition, depending on the specific application, the electronic device may include any other suitable components.

[0147] Example 4:

[0148] Embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps described in the "Exemplary Methods" section above according to the various embodiments of this application.

[0149] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0150] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not restrict the application from being implemented using the specific details described above.

[0151] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0152] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0153] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0154] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for enhancing the vision of underwater robots in turbid waters, characterized in that, include: The original color image sequence collected in real time by the underwater robot is obtained, and the pixel intensity of the original color image sequence is mapped to the complex space to construct the complex domain analytic signal. Extract the multi-directional spatial gradient tensor of the original color image, and calculate the pseudo-polarization information and environmental characterization parameters accordingly. A virtual phase compensation operator is constructed using the pseudo polarization degree information and the environmental characterization parameters; wherein, the compensation intensity of the virtual phase compensation operator is driven by the pseudo polarization degree information, and the compensation intensity is locally weighted and corrected using the environmental characterization parameters; In the complex space, the virtual phase compensation operator is used to perform phase rotation compensation on the complex domain analytic signal, and the amplitude of the complex domain analytic signal is suppressed simultaneously according to the pseudo polarization degree information to obtain an enhanced complex domain analytic signal. The modulus is extracted from the enhanced complex domain analytical signal. Based on the energy distribution characteristics of the blue and green channels in the original color image, the energy attenuation ratio of the red channel in the current water area is dynamically calculated. Based on this, a gain compensation coefficient is generated, and the enhanced image is output.

2. The underwater robot visual enhancement method for turbid waters according to claim 1, characterized in that, After performing phase rotation compensation on the complex domain analytic signal, the method further includes: The signal contrast features of each color channel in the original color image are extracted. Based on the spatial distribution differences of the signal contrast features among different color channels, a compensation intensity adjustment factor is generated, and the gain value of the virtual phase compensation operator is corrected using the compensation intensity adjustment factor. The complex domain analytic signal is subjected to multi-scale frequency domain filtering to extract the high-frequency components in the preset frequency range, and the amplitude of the high-frequency components is reconstructed by weighting according to the pseudo-polarization information. The contrast change gradient of the enhanced complex domain analytic signal in the time domain sequence is obtained, and an image detail enhancement operator is dynamically mapped based on the contrast change gradient. The image detail enhancement operator is then used to perform pixel-level mapping on the image signal after the modulus value is extracted.

3. The underwater robot visual enhancement method for turbid waters according to claim 2, characterized in that, After multi-scale frequency domain filtering, the process also includes: Extract the energy distribution characteristics of each frequency component in the complex domain analytic signal, establish the gain correlation between the energy distribution characteristics and the enhanced complex domain analytic signal, and dynamically correct the gain coefficient of each frequency component based on the gain correlation; The attenuation gradient information of underwater ambient light in the depth dimension is obtained, and the compensation intensity of the virtual phase compensation operator is calibrated twice based on the attenuation characteristic curve determined by the attenuation gradient information. Extract the contrast distribution entropy value of the image signal in a local region, generate an adaptive filtering operator based on the contrast distribution entropy value, and use the adaptive filtering operator to compensate for the spatial texture of the image signal.

4. The underwater robot visual enhancement method for turbid waters according to claim 1, characterized in that, The construction of the virtual phase compensation operator includes: Based on the pseudo-polarization degree information, a phase compensation reference quantity that varies with the positive direction of the pseudo-polarization degree is generated; A spatial weight distribution function is constructed using the environmental characterization parameters. The gain of the phase compensation reference quantity is adjusted pixel by pixel using the spatial weight distribution function to obtain the locally corrected phase offset. A complex domain rotation operator is constructed using the locally corrected phase offset, which serves as a virtual phase compensation operator.

5. The underwater robot visual enhancement method for turbid waters according to claim 4, characterized in that, The construction of the spatial weight distribution function includes: Local gradient consistency features are extracted from the original color image sequence, and the local gradient consistency features are spatially mapped with the environmental characterization parameters to obtain the structural confidence factor of each pixel. Based on the ratio of the structural confidence factor to the pseudo polarization information, pixel targets whose structural confidence factor is in a preset advantage range are determined, and the gain ratio of the phase compensation reference quantity corresponding to the pixel target is reduced to construct a spatial weight distribution function.

6. The underwater robot visual enhancement method for turbid waters according to claim 1, characterized in that, The phase rotation compensation includes: The virtual phase compensation operator is used to perform inverse phase mapping on the complex domain analytic signal to correct the spatial phase distribution of the complex domain analytic signal to a preset coherent reference plane; Within the coherent reference plane, the first and second orthogonal components of the complex domain analytic signal are extracted. Based on the attenuation weight determined by the pseudo-polarization information, the energy of the first and second components is redistributed to reconstruct the coherent wavefront of the target object.

7. The underwater robot visual enhancement method for turbid waters according to claim 1, characterized in that, The suppression of the amplitude of the complex domain analytic signal includes: Based on the pseudo-polarization degree information, the instantaneous energy envelope of the complex domain analytic signal at each pixel position is extracted; The instantaneous energy envelope is compressed using the pseudo-polarization information to obtain a compressed energy distribution. The magnitude of the complex domain analytic signal is then remapped based on the compressed energy distribution.

8. A visual enhancement system for underwater robots in turbid waters, characterized in that, include: The complex construction module is used to acquire the original color image sequence collected in real time by the underwater robot, map the pixel intensity of the original color image sequence to the complex space, construct the complex domain analytic signal, extract the multi-directional spatial gradient tensor of the original color image, and calculate the pseudo polarization information and environmental characterization parameters accordingly. An operator generation module is used to construct a virtual phase compensation operator using the pseudo polarization degree information and the environmental characterization parameters; wherein, the pseudo polarization degree information is used to drive the compensation intensity of the virtual phase compensation operator, and the environmental characterization parameters are used to perform local weight correction on the compensation intensity; The complex domain compensation module is used to perform phase rotation compensation on the complex domain analytic signal in the complex space using the virtual phase compensation operator, and simultaneously suppress the amplitude of the complex domain analytic signal according to the pseudo polarization information to obtain an enhanced complex domain analytic signal. The energy gain output module is used to extract the modulus of the enhanced complex domain analytical signal. Based on the energy distribution characteristics of the blue and green channels in the original color image, it dynamically calculates the energy attenuation ratio of the red channel in the current water area, and generates a gain compensation coefficient accordingly to output the enhanced image.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 7.

10. A computer storage medium storing computer-executable instructions thereon, characterized in that: When the computer-executable instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 7.

Images