Muddy water image processing method based on image analysis
By calculating saturation, saliency, and Laplacian weights, and combining them with a CNN-GAN filtering model, the problems of low contrast and complex backgrounds in underwater image processing are solved, improving the clarity and positioning accuracy of underwater robots in observing the diameter of cable pipes.
Patent Information
- Application Number
- CN202511448351.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-11
- Publication Date
- 2026-01-13
AI Technical Summary
Traditional underwater high-definition cameras struggle to accurately locate cable and duct openings due to underwater debris and low contrast, and the complex backgrounds present numerous interference factors that traditional image processing algorithms cannot handle.
An image analysis-based approach is used to calculate saturation, saliency, and Laplacian weights. A CNN-GAN filtering model is then used for denoising to enhance image contrast and target region details. Multi-scale fusion technology is combined to improve image clarity.
It significantly improves the clarity and positioning accuracy of underwater robots in observing underwater targets, reduces the impact of complex background interference, and improves measurement reliability.
Smart Images
Figure CN121330481A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing, and in particular to a method for processing muddy water images based on image analysis. Background Technology
[0002] In recent years, underwater robots (AUVs / ROVs) have gradually become core tools for underwater inspection due to their autonomy, flexibility, and environmental adaptability. However, when using traditional high-definition cameras for underwater observation, it is difficult to achieve clear observation of objects in the water due to debris and contrast. The main reasons for this are: (1) Insufficient underwater lighting results in low contrast and indistinct color difference between black and white parts. This makes it difficult to accurately locate the edge of the cable pipe diameter when using edge detection algorithm for direct measurement, which affects the subsequent measurement of pipe diameter.
[0003] (2) In underwater cable wells, there are many factors that cause interference, such as robot connection lines, steel legs in the cable well, and human-caused interference such as leaves, plastic and other garbage and dust suspended in the water. Traditional image processing algorithms (such as edge detection and template matching) are difficult to handle feature extraction in complex backgrounds. Summary of the Invention
[0004] In view of this, the present invention aims to propose a muddy water image processing method based on image analysis to solve the problems in the prior art.
[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows: This invention proposes a method for processing murky water images based on image analysis, the method comprising the following steps: S1, acquire raw, turbid underwater images taken by the underwater robot; S2, calculate the weights of saturation and saliency features in the original image; S3, normalize the saturation weight, significance weight and Laplacian weight to generate an enhanced fusion weight map; S4 inputs the enhanced fused image into the CNN-GAN filtering model for denoising.
[0006] Furthermore, the saturation weight in S2 is calculated using the following formula: In the formula: The red channel value at coordinates (x, y) For brightness, The green channel value at coordinates (x, y) The blue channel value at coordinates (x, y).
[0007] Furthermore, the significance weights in S2 are calculated using the following formula: In the formula: The brightness value of the current pixel. To reference the average brightness of the background or surrounding area, Reflects brightness and contrast. This represents the red-green hue value of the current pixel. The average red-green hue of the reference background is used. The yellow-blue tint value of the current pixel. The average yellow-blue tint of the reference background.
[0008] Furthermore, in S3, the normalization process is calculated using the following formula: To represent the k-th weight at coordinates (x, y), Summing all weights at the same position (x, y); in: In the formula: For saturation weight, For significance weight, For Laplace weights.
[0009] Furthermore, in S3, a multi-scale fusion method is used for image fusion, decomposing the original image into a Laplacian pyramid and the weight map into a Gaussian pyramid, and then reconstructing and outputting the enhanced image after fusion according to scale. Image fusion is performed using the following formula: In the formula, F1(x,y) is the final output image pyramid. The coefficient of the first level of the Laplace pyramid at (x,y) Let represent the coefficient of the I-th level Gaussian pyramid at (x, y). For the weights of the Laplace pyramid, The weights are those of the Gaussian pyramid.
[0010] Furthermore, after obtaining the image pyramid, it can be upsampled to obtain the output image; This is done using the following formula: In the formula, This represents the final output image, where U indicates the downsampling operation. This method can better restore the true colors of the image and improve its contrast.
[0011] Furthermore, in S4, the training of the CNN-GAN model includes: During the training phase, 4000 labeled images were input into the GAN network, and its optimal filtering parameters and corresponding PSNR / SSIM values were recorded. During the test set phase, after obtaining all the training set parameters, 1000 test images are input into the CNN network. The CNN network is trained to obtain the GAN input parameters for these 1000 images and then passed to the GAN. GANs filter images using a filtering model that includes the input, obtain the filtered image, and record the corresponding image parameters.
[0012] Furthermore, PSNR, or Peak Signal-to-Noise Ratio, is derived using the following formula: Where max2 is the maximum possible pixel value of the original image I, usually 255; MSE represents the mean square error between the original image I of size M*N and the processed image K; where: In the formula: M represents the height of the image, and N represents the width of the image. Let be the pixel value of the original image at coordinates (i,j). The pixel value of the processed image at coordinates (i,j).
[0013] Furthermore, SSIM, or structural similarity, is derived using the following formula: In the formula: These represent the average pixel values within local windows of images x and y, respectively. , These represent the standard deviations of pixel values within local windows of the image (x and y, respectively). This represents the covariance of the image x and y within a local window; , It is the stability constant.
[0014] Compared with the prior art, the present invention has the following advantages: In this invention, the method effectively overcomes the problems of low contrast and complex background interference faced by traditional underwater visual inspection, and significantly improves the observation clarity, positioning accuracy and measurement reliability of underwater robots for underwater targets.
[0015] By fusing saturation weights, saliency weights, and Laplacian weights, this method enhances the details and boundary information of key target regions (such as pipe edges) in an image. The normalized fusion weight map guides the generation of enhanced images with higher local contrast, making the originally blurry target edges clear and sharp, laying the foundation for subsequent accurate edge detection and localization. Attached Figure Description
[0016] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is the image before enhancement and fusion in this invention; Figure 3 This is the enhanced fused image of the present invention. Detailed Implementation
[0017] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0018] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "back," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0019] Furthermore, in the description of this invention, unless otherwise explicitly defined, the terms "installation," "connection," "linking," and "connector" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention in light of the specific circumstances.
[0020] The following will refer to the appendix. Figures 1 to 3 The present invention will be described in detail with reference to the embodiments.
[0021] It should be understood that the transmission characteristics of underwater optical images captured by underwater robots differ from those in the atmosphere. In water, images are subject to absorption and scattering. In addition to the influence of the water itself, suspended particles such as sand, dust, and plankton in the underwater environment, along with dissolved organic matter, combine to cause double optical interference to the raw images acquired underwater.
[0022] Among them, Rayleigh scattering effect produced by light on the surface of suspended particles will cause image edge blurring and reduction of texture details, while the selective absorption characteristics of water to different wavelengths of light will cause chromatographic shift, specifically manifested as significant attenuation of the red spectrum, accompanied by compression of the overall image brightness and darkness and a decrease in saturation.
[0023] Overall, this invention proposes a method for processing murky water images based on image analysis, which includes the following steps: S1, acquire raw, turbid underwater images taken by the underwater robot; S2, calculate the weights of saturation and saliency features in the original image; S3, normalize the saturation weight, significance weight and Laplacian weight to generate an enhanced fusion weight map; S4 inputs the enhanced fused image into the CNN-GAN filtering model for denoising.
[0024] In this embodiment, this method effectively overcomes the problems of low contrast and complex background interference faced by traditional underwater visual inspection, and significantly improves the observation clarity, positioning accuracy and measurement reliability of underwater robots (AUV / ROV) for underwater targets (such as the diameter of cable pipes).
[0025] In detail, by fusing saturation weights, saliency weights, and Laplacian weights, this method can enhance the details and boundary information of key target regions (such as pipe edges) in an image. The normalized fusion weight map guides the generation of enhanced images with higher local contrast, making the originally blurry target edges clear and sharp, laying the foundation for subsequent accurate edge detection and localization.
[0026] Combined with the subsequent CNN-GAN filtering model, it can intelligently learn and remove complex noises unique to turbid water bodies, such as suspended particle noise and light scattering noise, reducing the risk of false detection and false negatives caused by interference.
[0027] In this invention, saturation weights help distinguish color differences between the target and a muddy background. Saliency weights automatically focus on visually prominent areas in the image that may represent the target. Laplacian weights enhance high-frequency information in the image (i.e., edges and textures).
[0028] It should be noted that the saturation weight in S2 is calculated using the following formula: In the formula: The red channel value at coordinates (x, y) For brightness, The green channel value at coordinates (x, y) The blue channel value at coordinates (x, y).
[0029] Furthermore, the significance weights in S2 are calculated using the following formula: In the formula: The brightness value of the current pixel. To reference the average brightness of the background or surrounding area, Reflects brightness and contrast. This represents the red-green hue value of the current pixel. The average red-green hue of the reference background is used. The yellow-blue tint value of the current pixel. The average yellow-blue tint of the reference background.
[0030] The introduction of saliency weights enables the algorithm to prioritize potential target areas in the image that are significantly different from the background, rather than interference objects such as robot cables, metal structures, and floating debris.
[0031] In S3, the normalization process is calculated using the following formula: To represent the k-th weight at coordinates (x, y), Summing all weights at the same position (x, y); in: In the formula: For saturation weight, For significance weight, For Laplace weights.
[0032] In S3, a multi-scale fusion method is used for image fusion. The original image is decomposed into a Laplacian pyramid and the weight map is decomposed into a Gaussian pyramid. After fusion by scale, the enhanced image is reconstructed and output. Image fusion is performed using the following formula: In the formula, F1(x,y) is the final output image pyramid. The coefficient of the first level of the Laplace pyramid at (x,y) Let represent the coefficient of the I-th level Gaussian pyramid at (x, y). For the weights of the Laplace pyramid, The weights are those of the Gaussian pyramid.
[0033] It should be further explained that the fusion process is not a global average enhancement, but rather a selective local enhancement based on the weight map, which highlights the target while relatively weakening the interference of insignificant background areas.
[0034] This calculation can improve the accuracy of underwater robot vision systems in identifying targets such as cables and pipes in turbid environments, as well as their size measurement accuracy, and also provides strong anti-interference capabilities.
[0035] Based on the above, the training of the CNN-GAN model in S4 includes: During the training phase, 4000 labeled images were input into the GAN network, and its optimal filtering parameters and corresponding PSNR / SSIM values were recorded. During the test set phase, after obtaining all the training set parameters, 1000 test images are input into the CNN network. The CNN network is trained to obtain the GAN input parameters for these 1000 images and then passed to the GAN. GANs filter images using a filtering model that includes the input, obtain the filtered image, and record the corresponding image parameters.
[0036] During the training phase, 4000 labeled images were input into the GAN network to provide sample features for the model. By recording the optimal filtering parameters, the model can accurately capture the filtering settings that perform best under different image features, laying a solid foundation for subsequent image processing tasks. For example, when processing images with complex textures, the optimal filtering parameters can ensure that noise is removed while preserving the image's details to the greatest extent possible, avoiding image blurring due to over-filtering.
[0037] During the test set phase, a CNN network was used to train on 1000 test images to obtain the input parameters of the GAN for these 1000 images.
[0038] This approach allows the model to be personalized based on the unique features of the test images, rather than simply applying the generic parameters obtained during the training phase.
[0039] Furthermore, recording the corresponding PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index) values provides objective and quantitative indicators for evaluating the filtering effect of the model.
[0040] PSNR is the peak signal-to-noise ratio, which is calculated using the following formula: Where max2 is the maximum possible pixel value of the original image I, usually 255; MSE represents the mean square error between the original image I of size M*N and the processed image K; where: In the formula: M represents the height of the image, and N represents the width of the image. Let be the pixel value of the original image at coordinates (i,j). The pixel value of the processed image at coordinates (i,j).
[0041] SSIM is the structural similarity score, calculated using the following formula: In the formula: These represent the average pixel values within local windows of images x and y, respectively. , These represent the standard deviations of pixel values within local windows of the image (x and y, respectively). This represents the covariance of the image x and y within a local window; , It is the stability constant.
[0042] The image processing method described above can be applied to the processing systems of different underwater robots and can be programmed into a computer-readable storage medium. In practical applications, the computer-readable storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
[0043] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
[0044] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.
Claims
1. A method for processing murky water images based on image analysis, characterized in that: The method includes the following steps: S1, acquire raw, turbid underwater images taken by the underwater robot; S2, calculate the weights of saturation and saliency features in the original image; S3, normalize the saturation weight, significance weight and Laplacian weight to generate an enhanced fusion weight map; S4 inputs the enhanced fused image into the CNN-GAN filtering model for denoising.
2. A method for processing murky water images based on image analysis according to claim 1, characterized in that: The saturation weight in S2 is calculated using the following formula: In the formula: The red channel value at coordinates (x, y) For brightness, The green channel value at coordinates (x, y) The blue channel value at coordinates (x, y).
3. The image analysis-based muddy water image processing method according to claim 1, characterized in that: The significance weight in S2 is calculated using the following formula: In the formula: The brightness value of the current pixel. To reference the average brightness of the background or surrounding area, Reflects brightness and contrast. This represents the red-green hue value of the current pixel. The average red-green hue of the reference background is used. The yellow-blue tint value of the current pixel. The average yellow-blue tint of the reference background.
4. The image analysis-based muddy water image processing method according to claim 1, characterized in that: In S3, the normalization process is calculated using the following formula: To represent the k-th weight at coordinates (x, y), Summing all weights at the same position (x, y); in: In the formula: For saturation weight, For significance weight, For Laplace weights.
5. The image analysis-based muddy water image processing method according to claim 4, characterized in that: In S3, a multi-scale fusion method is used for image fusion. The original image is decomposed into a Laplacian pyramid and the weight map is decomposed into a Gaussian pyramid. After fusion by scale, the enhanced image is reconstructed and output. Image fusion is performed using the following formula: In the formula, F1(x,y) is the final output image pyramid. The coefficient of the first level of the Laplace pyramid at (x,y) Let represent the coefficient of the I-th level Gaussian pyramid at (x, y). For the weights of the Laplace pyramid, The weights are those of the Gaussian pyramid.
6. The image analysis-based muddy water image processing method according to claim 7, characterized in that: After obtaining the image pyramid, it can be upsampled to obtain the output image; This is done using the following formula: In the formula, This represents the final output image, where U indicates the downsampling operation. This method can better restore the true colors of the image and improve its contrast.
7. The image analysis-based muddy water image processing method according to claim 1, characterized in that: In S4, the training of the CNN-GAN model includes: During the training phase, 4000 labeled images were input into the GAN network, and its optimal filtering parameters and corresponding PSNR / SSIM values were recorded. During the test set phase, after obtaining all the training set parameters, 1000 test images are input into the CNN network. The CNN network is trained to obtain the GAN input parameters for these 1000 images and then passed to the GAN.
8. The image analysis-based muddy water image processing method according to claim 7, characterized in that: GANs filter images using a filtering model that includes the input, obtain the filtered image, and record the corresponding image parameters.
9. The image analysis-based muddy water image processing method according to claim 8, characterized in that: PSNR is the peak signal-to-noise ratio, which is calculated using the following formula: Where max2 is the maximum possible pixel value of the original image I, usually 255; MSE represents the mean square error between the original image I of size M*N and the processed image K; where: In the formula: M represents the height of the image, and N represents the width of the image. Let be the pixel value of the original image at coordinates (i,j). The pixel value of the processed image at coordinates (i,j).
10. A method for processing murky water images based on image analysis according to claim 8, characterized in that: SSIM is the structural similarity score, calculated using the following formula: In the formula: These represent the average pixel values within local windows of images x and y, respectively. , These represent the standard deviations of pixel values within local windows of the image (x and y, respectively). This represents the covariance of the image x and y within a local window; , It is the stability constant.