Underwater dredging robot terrain recognition method and system

Abstract

The invention belongs to the technical field of image processing, and discloses an underwater dredging robot terrain recognition method and system, and the method comprises the steps: obtaining an illumination attenuation coefficient through the shooting depth and plankton information, accurately correcting the influence of illumination on an image, and carrying out the brightness adjustment of the image based on the illumination attenuation coefficient, so as to enable the terrain features to be more prominent. Utilizing Laplace decomposition to obtain a multi-layer topographic image, and realizing multi-scale analysis; the accuracy of terrain analysis is improved by screening an edge texture region; the defogging weight is determined in combination with the illumination attenuation coefficient, and the dark channel prior is adopted for defogging recovery, so that the defogging effect is effectively removed, and the image visibility is improved. And reconstructing and synthesizing a complete image through an inverse Laplacian pyramid to obtain an underwater topographic image of which the quality is comprehensively improved, and further obtaining underwater topographic information. The method not only improves the image quality, but also achieves the precise analysis and recognition of the underwater topography, and provides powerful technical support for the research of the underwater topography.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and relates to a method and system for underwater dredging robot terrain recognition. Background Art

[0002] Underwater dredging robots usually carry tools such as sensors (such as sonar, laser, camera, etc.) and robotic arms to conduct terrain detection and cleaning work underwater. Their main task is to remove silt and sediment at the bottom of the water, especially for dredging operations in waters such as reservoirs, ports, and rivers. When conducting operations such as dredging, exploration, and engineering construction in the underwater environment, terrain recognition technology can help the robot accurately understand the underwater terrain and obstacle conditions, so as to achieve more precise and efficient operations.

[0003] Generally, underwater dredging robots are usually in environments such as rivers, lakes, ports, and the seabed. Currently, when recognizing the terrain during the dredging process based on underwater terrain images, due to the poor underwater light conditions, the image clarity is poor. The current dehazing algorithm based on dark channel prior usually restores the image clarity through global optimization, but the global optimization method may cause local distortion of the image, affecting the accurate recognition of the terrain. Summary of the Invention

[0004] The purpose of the present invention is to solve the problem that the dehazing algorithm based on dark channel prior in the prior art may cause local distortion of the image and affect terrain recognition, and to provide a method and system for underwater dredging robot terrain recognition.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: A method for underwater dredging robot terrain recognition, including: Based on the underwater dredging robot, obtain the underwater real-time terrain image and preprocess the obtained terrain image; Based on the shooting depth of the underwater dredging robot and the influence of plankton in the water, obtain the light attenuation coefficient; Based on the light attenuation coefficient, brighten the gray levels of different pixels in the preprocessed terrain image; Perform Laplace decomposition on the adjusted image to obtain multi-layer terrain images; Based on any layer of terrain image, obtain the connected domain corresponding to the terrain image and screen out the existing edge texture regions; Based on the proportion of the edge texture regions in each layer of image, obtain the weight of the importance of high-frequency information in each layer of scale image, and determine the final dehazing weight in combination with the light attenuation coefficient; Based on the final defogging weight, the image is defogged and restored by the dark channel prior, and the image information at multiple scales is synthesized into a complete image through inverse Laplacian pyramid reconstruction, thereby obtaining underwater terrain information.

[0006] A further improvement of the present invention lies in: Further, preprocess the obtained terrain image, specifically: perform grayscale processing on the obtained terrain image to obtain a grayscale image.

[0007] Further, based on the shooting depth of the underwater dredging robot and the influence of plankton in the water, obtain the light attenuation coefficient, specifically: The light attenuation situation in general underwater areas is affected by the shooting depth and plankton in the water. If the shooting depth is deeper and the plankton density is greater, the corresponding light attenuation situation is greater; at the same time, the color wavelength of the water body also affects the light absorption rate of the water. The shorter the wavelength, the weaker the absorption ability and the slower the light attenuation. Then there is:

[0008] Where, is the light attenuation coefficient corresponding to a certain terrain image ; is the shooting depth of the terrain image ; is the plankton density at the corresponding depth of this water body, is the color wavelength value of this water body, is the longest wavelength value of this water body; since the longer the wavelength, the stronger the corresponding absorption ability, that is, is smaller, then the greater the light attenuation degree of this water body. Combining the shooting depth and the plankton density , then the greater its value, the greater the corresponding light attenuation coefficient, and it is normalized to the range of [0,1].

[0009] Further, based on the light attenuation coefficient, brighten the gray levels of different pixels in the terrain image, specifically: For each pixel of this terrain image, the weaker the light, the darker the pixel gray level, that is, the gray value tends to 0. Therefore, based on the light attenuation coefficient, brighten the gray levels of different pixels. The greater the light attenuation degree, the greater the adjustment degree, and vice versa. Then there is:

[0010] Where, is the adjusted pixel value of the pixel point in the terrain image ; is the original pixel value of the pixel point , is a topographic image The corresponding light attenuation coefficient.

[0011] Furthermore, based on the topographic image of any layer, obtain the connected components corresponding to the topographic image, and screen out the existing edge texture regions, specifically: In general images, the gray level of the edge texture region is significantly different from that of other regions, often showing high contrast. At the same time, the shape of the edge texture region usually appears in the form of linear or curved extension; for the image at a certain scale, obtain the corresponding connected components, and then screen out the possible edge texture regions, which is expressed as:

[0012] Among them, is a certain connected component in the image at this scale is the confidence of the edge texture region; is the connected component The average gray value of, is the average gray value of this image; is the modulus length of the maximum principal component vector after PCA analysis of this connected component, is the average modulus length of all principal component vectors of this connected component; Therefore represents the gray level difference of the connected component in the image at this scale. Since the gray level of the edge texture region has strong contrast, therefore The larger the value, the more likely it is to be an edge texture region; represents the morphological extension trend of this connected component. Since the largest principal component vector in PCA represents its most obvious extension degree, that is, if the extension trend of this connected domain is more obvious, corresponding relative to The larger the value of, the more likely it is to be an edge texture region. Finally, normalize the result to the range of [0, 1], and select The corresponding region is the high-frequency edge texture region in the image.

[0013] Furthermore, the proportion of the edge texture region in each layer of the image, specifically:

[0014] Among them, is the area of the edge texture region of the th layer of the image, is the th layer of the overall area of the image; The weight for obtaining the importance of high-frequency information in each layer of the scale image is specifically: Put the proportion of the edge texture area in each layer of the image into a sequence For the images at any scale, there is:

[0015] Among them, is the importance weight of the high-frequency information in the -layer scale image; is the proportion of the area of the high-frequency information region in the -layer scale image, is the proportion of the area of the high-frequency information region in the -layer scale image, is the number of decomposition layers; is the relative value of the area of the high-frequency information in the -layer scale image; the larger its value, the more high-frequency information in the corresponding layer of the image, and then the greater the importance weight; represents the relationship between its layer number and high-frequency information. Since the lower the layer number, the more important the corresponding high-frequency information, and the greater the corresponding importance weight.

[0016] Furthermore, combine the light attenuation coefficient to determine the final defogging weight, specifically:

[0017] Among them, is the defogging weight of the -layer scale image; is the importance weight of the high-frequency information in the -layer scale image, is the light attenuation coefficient of this original image; that is, if the light attenuation coefficient of this image is larger, the overexposure situation in the adjusted high-frequency information region is more serious, then a larger defogging weight is required to reduce this situation, and finally the result is normalized to the range of [0, 1].

[0018] Furthermore, based on the final defogging weight, perform defogging restoration on the image through the dark channel prior, specifically: Take the image at each scale as the input image, and calculate the dark channel of each pixel point to generate a dark channel image ; By calculating the maximum value position in the dark channel image , estimate the atmospheric light value and the transmittance ; Through the estimated transmittance and the atmospheric light and the defogging weight, restore a clear fog-free image:

[0019] Among them, is the -th layer scale image after haze removal, is the original -th layer scale hazy image, is the haze removal weight of the -th layer scale image; that is, for images of different layer scales, the dark channel prior method is optimized according to their haze removal weights to improve the haze removal effect.

[0020] Furthermore, the image at each scale is used as the input image, and the dark channel of each pixel is calculated to generate a dark channel image , specifically: Assume the input image is , where is each pixel in the image; since some color channels in natural scene images are very dark, the minimum value of the RGB channels of each pixel is used to represent the "dark channel"; For an RGB image, the dark channel is defined as:

[0021] Among them, , , are the pixel values of the image on the red, green, and blue channels respectively; By calculating the maximum value position in the dark channel image , the atmospheric light value and the transmittance are estimated, specifically: In the dark channel image , find the pixel position with the maximum value; the pixel position with the maximum value corresponds to the darkest part of the image; by obtaining the pixel value in the original image corresponding to the maximum value position in the dark channel image, the atmospheric light value is estimated. Usually, the RGB pixel value corresponding to the maximum dark channel value is selected as the atmospheric light value ; Assume the maximum value position in the of the image corresponds to the pixel value in the original image, then the atmospheric light value is:

[0022] Among them, is the pixel value corresponding to the maximum dark channel position in the image ; The transmittance reflects the haze degree of each pixel in the image. The transmittance It is usually by estimating the dark primary color of each pixel of the image; the formula for calculating the transmittance is as follows:

[0023] wherein, is a constant, and its value range is between [0, 1].

[0024] An underwater dredging robot terrain recognition system, comprising: A first acquisition module, which acquires a real-time underwater terrain image based on an underwater dredging robot and preprocesses the acquired terrain image; A second acquisition module, which acquires a light attenuation coefficient based on the shooting depth of the underwater dredging robot and the influence of plankton in the water; An adjustment module, which brightens and adjusts the gray levels of different pixels in the preprocessed terrain image based on the light attenuation coefficient; A decomposition module, which performs Laplacian decomposition on the adjusted image to obtain multi-layer terrain images; A screening module, which acquires the connected domain corresponding to the terrain image based on any layer of the terrain image and screens out the existing edge texture regions; A determination module, which acquires the weight of the importance of high-frequency information in each layer of scale image based on the proportion of the edge texture region in each layer of image, and determines the final defogging weight in combination with the light attenuation coefficient; A reconstruction module, which performs defogging restoration on the image through the dark channel prior based on the final defogging weight, and synthesizes the image information of multiple scales into a complete image through inverse Laplacian pyramid reconstruction, thereby obtaining underwater terrain information.

[0025] Compared with the prior art, the present invention has the following beneficial effects: The present invention utilizes the shooting depth and plankton information to acquire the light attenuation coefficient, accurately corrects the influence of light on the image, and brightens and adjusts the image based on the light attenuation coefficient, making the terrain features more prominent. Multi-layer terrain images are obtained by Laplacian decomposition to achieve multi-scale analysis and comprehensively understand the underwater terrain structure. By screening the edge texture regions, the accuracy of terrain analysis is improved. The defogging weight is determined in combination with the light attenuation coefficient, and defogging restoration is performed using the dark channel prior to effectively remove the fog effect and improve the image visibility. Finally, a complete image is synthesized through inverse Laplacian pyramid reconstruction to achieve efficient information fusion, obtain an underwater terrain image with comprehensively improved quality, and thereby obtain underwater terrain information. The present invention not only improves the image quality, but also realizes the accurate analysis and recognition of the underwater terrain, providing strong technical support for underwater terrain research. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0027] Figure 1 It is a schematic flowchart of a method for improving the clarity of underwater terrain images of the present invention; Figure 2 It is a schematic structural diagram of a system for improving the clarity of underwater terrain images of the present invention. Detailed implementation manners

[0028] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the drawings here can be arranged and designed in various different configurations.

[0029] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0030] It should be noted that similar reference numerals and letters indicate similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0031] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper", "lower", "horizontal", "inner", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the invention product is usually placed during use, it is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, terms such as "first", "second", etc. are only used for differential description and cannot be understood as indicating or implying relative importance.

[0032] In addition, when the term "horizontal" appears, it does not mean that the component is required to be absolutely horizontal, but it can be slightly inclined. For example, "horizontal" only means that its direction is more horizontal relative to "vertical", and it does not mean that the structure must be completely horizontal, but it can be slightly inclined.

[0033] In the description of the embodiments of the present invention, it should also be noted that unless otherwise clearly specified and limited, when the terms "arranged", "installed", "connected", "coupled" appear, they should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0034] The present invention will be further described in detail below with reference to the accompanying drawings: See Figure 1 , the present invention discloses a method for underwater terrain recognition of a dredging robot, including: S101, acquiring an underwater real-time terrain image based on the underwater dredging robot, and preprocessing the acquired terrain image; An underwater dredging robot is an automated robot system specifically used for tasks such as dredging, salvage, surveying, and repair in underwater environments. It is widely used in the silt cleaning work of waters such as reservoirs, rivers, ports, and waterways. During the working process, the robot generally collects underwater images through a high-definition camera to help the operator observe the terrain conditions and cleaning effects in real time.

[0035] Since the water quality in the underwater environment where the underwater dredging robot works is usually turbid and the light is poor, the image will be unclear, with severe fogging and difficult to capture details, which affects the accurate recognition of the terrain. And the current defogging algorithms, such as dark channel prior, generally restore the clarity of the image through global optimization. In underwater images, local features such as the texture of the terrain, the edges of objects, and water flow are crucial for terrain recognition. The global optimization method may ignore the local texture information in the image, resulting in an unsatisfactory defogging effect.

[0036] After the underwater dredging robot captures an underwater real-time terrain image through a high-definition camera, it preprocesses the captured terrain image, performs grayscale processing on the acquired terrain image, and obtains a grayscale image.

[0037] S102, obtaining a light attenuation coefficient based on the shooting depth of the underwater dredging robot and the influence of plankton in the water; The dark channel prior is a prior knowledge based on the statistical laws of image colors and was first used in the research of image dehazing. The dark channel prior can be used to estimate the atmospheric light and transmittance in the image, thereby restoring the clarity of the image. However, the method with global effect has a poor degree of detail restoration for underwater terrain. To improve the image quality, this step is based on the idea of multi-scale fusion. The image is decomposed into representations of different scales through the Laplacian pyramid, and then the image information of each scale is processed and fused.

[0038] Since the underwater terrain in the dredging area generally has a low light intensity, the image often appears blurred, dull, and details are lost. These characteristics will cause important high-frequency information to be lost in the image after Laplacian pyramid decomposition. Therefore, it is necessary to perform light compensation on the image first to obtain clearer high-frequency detail processing.

[0039] Generally, the light attenuation in the underwater area is affected by the shooting depth and plankton in the water. If the shooting depth is deeper and the plankton density is greater, the corresponding light attenuation is greater. At the same time, the color wavelength of the water body also affects the light absorption rate of the water. The shorter the wavelength, the weaker the absorption ability, and the slower the light attenuation. Then there is:

[0040] Among them, is the light attenuation coefficient corresponding to a certain terrain image ; is the shooting depth of the terrain image ; is the plankton density at the corresponding depth of this water body, is the color wavelength value of this water body, is the longest wavelength value of this water body; since the longer the wavelength, the stronger the corresponding absorption ability, that is, is smaller, then the light attenuation degree of this water body is greater. Combining the shooting depth and the plankton density , then the greater its value, the greater the corresponding light attenuation coefficient, and it is normalized to the range of [0, 1].

[0041] S103, brighten the gray levels of different pixels in the preprocessed terrain image based on the light attenuation coefficient; For each pixel of this terrain image, the weaker the light, the darker the pixel gray level, that is, the gray value tends to 0. Therefore, based on the light attenuation coefficient, the gray levels of different pixels are brightened. The greater the light attenuation degree, the greater the corresponding adjustment degree, and vice versa. Then there is:

[0042] Among them, is the terrain image Adjusted pixel value of the middle pixel point ; is the original pixel value of the pixel point , is the light attenuation coefficient corresponding to the terrain image .

[0043] S104. Perform Laplacian decomposition on the adjusted image to obtain multi-layer terrain images.

[0044] The image resolutions of different layers are different and the information they contain is different. Generally, the top layer contains more smooth low-frequency information, and the lower each layer is, the more high-frequency detail information it gradually contains, and finally the original image is restored at the bottom layer.

[0045] In order to accurately distinguish the underwater terrain, it is necessary to use the dark channel prior method to perform defogging restoration on the underwater terrain image, and the high-frequency information corresponding to the texture edges in the image can better reflect the terrain height trend. Therefore, it is necessary to adaptively defog the high-frequency information of the image at different scales and then fuse it. Thus, the edge texture of the image is retained during the defogging process, and the accuracy of terrain recognition is improved.

[0046] Since the high-frequency information in the multi-layer terrain images after decomposition gradually decreases from the bottom layer to the top layer, it means that the edge texture of the graph gradually decreases. At the same time, the high-frequency information in the lower-layer images represents more important terrain edge textures, which is more important for improving the recognition accuracy of underwater terrain. Therefore, it is necessary to determine the high-frequency information regions of the images at different scales.

[0047] S105. Based on the terrain image of any layer, obtain the connected domain corresponding to the terrain image and screen out the existing edge texture regions; Generally, the gray level of the edge texture region in the image is significantly different from other regions, often showing the characteristics of high contrast. At the same time, the shape of the edge texture region usually appears in the form of linear or curved extension; obtaining the corresponding connected domain for the image at a certain scale and then screening out the possible edge texture regions from it is expressed as:

[0048] wherein, is the confidence level of a certain connected domain in this scale image ; is the average gray level value of the connected domain , is the average gray level value of this image; is the modulus length of the maximum principal component vector after PCA analysis of this connected domain is the average modulus length of all principal component vectors of this connected domain; Therefore represents the connected regions in this scale image of the gray-scale difference. Since there is a strong contrast in the gray-scale of the edge texture region, therefore the larger the value, the more likely it is to be an edge texture region; represents the morphological extension trend of this connected region. Since the largest principal component vector in PCA represents its most obvious extension degree, that is, if the extension trend of this connected region is more obvious, corresponding to the larger the value, the more likely it is to be an edge texture region. Finally, the result is normalized to the range of [0, 1], and the region corresponding to

[0049] is selected as the high-frequency edge texture region in the image. After obtaining the high-frequency information corresponding to the multi-scale images of different layers, when performing defogging, these high-frequency regions are used as the target regions for optimization through the dark channel prior. Since the dark channel prior restores the clarity and contrast of the image by estimating the transmission rate and atmospheric light and applying the defogging algorithm, and optimizes the defogging effect by fusing multiple terrain images of different scales, and the importance of the high-frequency information in different scale images is different, then it is necessary to determine the corresponding defogging weights for fusion according to different scale images.

[0050] S106, based on the proportion of the edge texture region in each layer of the image, obtain the weight of the importance of the high-frequency information in each layer of the scale image, and combine the light attenuation coefficient to determine the final defogging weight; For the image at any scale, generally, the high-frequency information retained in the lower-layer image can better represent the main information of this image. Therefore, a larger defogging weight needs to be set for the lower-layer image. At the same time, the larger the proportion of the high-frequency information (edge texture region) in the image, the more high-frequency information this image has, and the more obvious defogging optimization is required, and the corresponding weight should also be larger. The proportion of the edge texture region in each layer of the image is specifically:

[0051] where, is the area of the edge texture region of the -th layer image, is the overall area of the -th layer image; The method for obtaining the weight of the importance of the high-frequency information in each layer of the scale image is specifically: Put the proportion of the edge texture region in each layer of the image into the sequence , and for the image at any scale, there is:

[0052] where, is the The importance weight of high-frequency information in the layer-scale image; is the proportion of the area of the high-frequency information region in the layer-scale image, is the proportion of the area of the high-frequency information region in the layer-scale image, is the number of decomposition layers; is the relative value of the high-frequency information area of the layer-scale image; the larger its value, the more high-frequency information the corresponding layer image has, and thus the greater the importance weight; represents the relationship between its layer number and high-frequency information. Since the lower the layer number, the more important the corresponding high-frequency information and the greater the corresponding importance weight.

[0053] Combine the light attenuation coefficient to determine the final defogging weight, specifically:

[0054] where is the defogging weight of the layer-scale image; is the importance weight of high-frequency information in the layer-scale image, is the light attenuation coefficient of this original image; that is, if the light attenuation coefficient of this image is larger, the overexposure situation in the adjusted high-frequency information region is more serious, then a larger defogging weight is required to reduce this situation, and finally the result is normalized to the range of [0, 1].

[0055] S107. Based on the final defogging weight, perform defogging restoration on the image through the dark channel prior, and synthesize the image information of multiple scales into a complete image through inverse Laplacian pyramid reconstruction, thereby obtaining underwater terrain information.

[0056] Take the image at each scale as the input image, and calculate the dark channel of each pixel point to generate a dark channel image ; Assume the input image is where is each pixel in the image; since some color channels of the image of the natural scene are very dark, the minimum value of the RGB channels of each pixel is used to represent the "dark channel"; For an RGB image, the dark channel is defined as:

[0057] where , , are the pixel values of the image on the red, green, and blue channels respectively; By calculating the dark channel image to estimate the atmospheric light value and the transmission rate ; In the dark channel image find the pixel position with the maximum value; the pixel with the maximum value corresponds to the darkest part of the image; by obtaining the pixel value in the original image corresponding to the position of the maximum value in the dark channel image, the atmospheric light value is estimated. Usually, the RGB pixel value corresponding to the maximum dark channel value is selected as the atmospheric light value ; Assume that the position of the maximum value in the image corresponds to the pixel value in the original image , then the atmospheric light value is:

[0058] where is the pixel value corresponding to the position of the maximum dark channel in the image ; The transmission rate reflects the haze degree of each pixel in the image. The transmission rate is usually estimated by the dark channel of each pixel of the image; the formula for calculating the transmission rate is as follows:

[0059] where is a constant, and the value range is between [0,1].

[0060] Through the estimated transmission rate and the atmospheric light and the defogging weight, a clear haze-free image is restored:

[0061] where is the defogged image of the th layer scale, is the original hazy image of the th layer scale, is the defogging weight of the th layer scale image; that is, for images of different layer scales, the dark channel prior method is optimized according to their defogging weights to improve the defogging effect.

[0062] After the dehazing optimization for images of different layer scales is completed, the images of different layers are fused and reconstructed. Here, the image information of multiple scales is synthesized back into a complete image through inverse Laplacian pyramid reconstruction, so as to ensure that the global brightness, contrast, and details of the image are well preserved. Through the obtained complete underwater image, underwater terrain information can be obtained.

[0063] By analyzing the underwater terrain image after real-time dehazing, the underwater dredging robot can generate the shortest path or the optimal cleaning strategy. For example, for areas with a large amount of silt accumulation, the robot can be set to concentrate on cleaning, while for areas with many obstacles, it can choose to avoid and mark them to avoid wasting cleaning time. At the same time, a three-dimensional model of the underwater environment can also be established by combining sensors such as sonar sensors and lidar to help the robot understand complex terrains and further improve the dredging efficiency.

[0064] See Figure 2 , the present invention discloses an underwater dredging robot terrain recognition system, including: A first acquisition module, which acquires an underwater real-time terrain image based on the underwater dredging robot and preprocesses the acquired terrain image; A second acquisition module, which acquires a light attenuation coefficient based on the shooting depth of the underwater dredging robot and the influence of plankton in the water; An adjustment module, which brightens the gray levels of different pixels in the preprocessed terrain image based on the light attenuation coefficient; A decomposition module, which performs Laplacian decomposition on the adjusted image to obtain multi-layer terrain images; A screening module, which based on an arbitrary layer of terrain image, acquires the connected regions corresponding to the terrain image and screens out the existing edge texture regions; A determination module, which based on the proportion of the edge texture regions in each layer of image, acquires the weights of the importance of high-frequency information in each layer scale image, and combines the light attenuation coefficient to determine the final dehazing weight; A reconstruction module, which based on the final dehazing weight, performs dehazing restoration on the image through the dark channel prior, and synthesizes the image information of multiple scales into a complete image through inverse Laplacian pyramid reconstruction, thereby obtaining underwater terrain information.

[0065] The terminal device provided by the embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned various method embodiments are implemented. Or, when the processor executes the computer program, the functions of each module / unit in the above-mentioned various device embodiments are implemented.

[0066] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention.

[0067] The terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory.

[0068] The processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0069] The memory can be used to store the computer program and / or module. By running or executing the computer program and / or module stored in the memory, and by invoking the data stored in the memory, the various functions of the terminal device can be realized.

[0070] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0071] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A terrain recognition method for an underwater dredging robot, characterized in that: include: Acquire underwater real-time terrain images based on the underwater dredging robot, and pre-process the acquired terrain images; Based on the shooting depth of the underwater dredging robot and the influence of plankton in the water, the light attenuation coefficient is obtained; Brighten the grayscale of different pixels in the preprocessed terrain image based on the light attenuation coefficient; Perform Laplace decomposition on the adjusted image to obtain a multi-layer terrain image; Based on the terrain image of any layer, the connected domain corresponding to the terrain image is obtained, and the existing edge texture area is screened out; Based on the proportion of edge texture areas in each layer of image, the importance weight of high-frequency information in each layer of scale image is obtained, and the final dehazing weight is determined in combination with the light attenuation coefficient; Based on the final dehazing weight, the image is dehazed and restored through the dark color prior, and the image information of multiple scales is synthesized into a complete image through inverse Laplacian pyramid reconstruction to obtain the underwater terrain information.

2. The underwater dredging robot terrain recognition method according to claim 1, characterized in that: The preprocessing of the acquired terrain image specifically includes: performing grayscale processing on the acquired terrain image to obtain a grayscale image.

3. The underwater dredging robot terrain recognition method according to claim 2, characterized in that: The light attenuation coefficient is obtained based on the shooting depth of the underwater dredging robot and the influence of plankton in the water, specifically: The light attenuation in the underwater area is affected by the shooting depth and the plankton in the water. The deeper the shooting depth and the greater the density of plankton, the greater the corresponding light attenuation. At the same time, the color wavelength of the water body also affects the water's absorption rate of light. The shorter the wavelength, the weaker the absorption capacity and the slower the light attenuation. In this case, there are: in, For a terrain image The corresponding light attenuation coefficient; For terrain images The depth of shooting, The plankton density at the corresponding depth of this water column is, For this water body’s color wavelength value, This is the longest wavelength value of the water body; since the longer the wavelength, the stronger the corresponding absorption capacity, that is, The smaller it is, the greater the light attenuation of the water body. Combined with the shooting depth and plankton density , then the larger the value, the larger the corresponding light attenuation coefficient, and it is normalized to the range of [0,1].

4. The underwater dredging robot terrain recognition method according to claim 3 is characterized in that: The brightening adjustment of different pixel grayscales in the terrain image based on the light attenuation coefficient is specifically as follows: For each pixel of this terrain image, the weaker the light, the darker the corresponding pixel grayscale, that is, the grayscale value is closer to 0. Therefore, based on the light attenuation coefficient, the different pixel grayscales are brightened and adjusted. The greater the light attenuation, the greater the corresponding adjustment, and vice versa. Then there exists: in, For terrain images Medium pixel The adjusted pixel value of Pixel The original pixel value, For terrain images The corresponding light attenuation coefficient.

5. The underwater dredging robot terrain recognition method according to claim 4, characterized in that: The method of obtaining a connected domain corresponding to the terrain image based on any layer and filtering out the existing edge texture areas is as follows: The grayscale of the edge texture area in the image is significantly different from that of other areas, showing the characteristics of high contrast. At the same time, the shape of the edge texture area appears in the form of linear or curved extension. For an image at a certain scale, the corresponding connected domain is obtained, and then the possible edge texture area is screened out, which is expressed as: in, A connected region in this scale image is the confidence of the edge texture area; Connected domain The average gray value of The average gray value of this image; For this connected domain, the modulus of the largest principal component vector after PCA analysis is: The average modulus of all principal component vectors in this connected domain; therefore Represents the connected domain in this scale image The grayscale difference of the edge texture area has a strong contrast. The larger the value, the more likely it is to be an edge texture area; Represents the morphological extension trend of this connected domain. Since the largest principal component vector in PCA Represents the most obvious extension degree, that is, if the extension trend of this Unicom domain is more obvious, the corresponding Relative to The larger the value, the more likely it is an edge texture area; finally, the result is normalized to the range of [0, 1] and selected The corresponding area is the high-frequency edge texture area in the image.

6. The underwater dredging robot terrain recognition method according to claim 5, characterized in that: The proportion of edge texture area in each layer of image is specifically: in, For the The edge texture area of ​​the layer image, For the The overall area of ​​the layer image; The weight of the importance of high-frequency information in each layer of scale image is obtained as follows: Put the proportion of edge texture area in each layer image into the sequence For images of any scale, there exists: in, For the Importance weight of high-frequency information in layer-scale images; For the The area ratio of high-frequency information in layer-scale images, For the The area ratio of high-frequency information in layer-scale images, is the number of decomposition layers; For the The relative value of the high-frequency information area of ​​the layer-scale image; the larger the value, the more high-frequency information the corresponding layer image has, and the greater the importance weight; It represents the relationship between the number of layers and high-frequency information. Since the lower the number of layers, the more important the corresponding high-frequency information is, the greater the corresponding importance weight is.

7. The method for terrain recognition of an underwater dredging robot according to claim 6, characterized in that: The final defogging weight is determined by combining the light attenuation coefficient, specifically: in, For the Dehazing weights for layer-scale images; For the The importance weight of high-frequency information in layer-scale images, is the illumination attenuation coefficient of the original image; that is, if the illumination attenuation coefficient of this image is larger, the overexposure in the adjusted high-frequency information area will be more serious, and a larger dehazing weight will be needed to reduce this situation, and finally the result will be normalized to the range of [0,1].

8. The method for terrain recognition of an underwater dredging robot according to claim 7, characterized in that: The defogging and restoration of the image is performed based on the final defogging weight through the dark primary color prior, specifically: Take the image at each scale as the input image, calculate the dark primary color of each pixel, and generate the dark primary color image ; By calculating the dark primary image The maximum value position in the atmospheric illumination value is estimated And the transmittance ; By estimating the transmittance and atmospheric lighting And the defogging weights to restore a clear fog-free image: in, After defogging Layer scale images, For the original Layer-scale foggy images, For the The defogging weight of the layer-scale image; that is, the dark color prior method is optimized according to the defogging weight of the different layer-scale images to improve the defogging effect.

9. The underwater dredging robot terrain recognition method according to claim 8, characterized in that: The image at each scale is used as the input image, and the dark primary color of each pixel is calculated to generate a dark primary color image. , specifically: Assume the input image is ,in, It is each pixel in the image; since some color channels of natural scene images are very dark, the minimum value of the RGB channel of each pixel is used to represent the "dark primary color"; For an RGB image, the dark primary is defined as: in, , , are the pixel values ​​of the image in the red, green, and blue channels respectively; By calculating the dark primary color image The maximum value position in the atmospheric illumination value is estimated And the transmittance , specifically: In dark color image Find the pixel position with the maximum value in the image; the pixel position with the maximum value corresponds to the darkest part of the image; infer the atmospheric illumination value by obtaining the pixel value in the original image corresponding to the maximum value position in the dark primary color image, and select the RGB pixel value corresponding to the maximum dark primary color value as the atmospheric illumination value ; Assuming the image The maximum position in Corresponding pixel value in the original image , then the atmospheric illumination value for: in, is an image The pixel value corresponding to the maximum dark primary color position in ; Transmittance Reflects the haze level of each pixel in the image, transmittance It is estimated by estimating the dark primary color of each pixel of the image; the formula for calculating the transmittance is as follows: in, is a constant whose value range is [0,1].

10. A terrain recognition system for an underwater dredging robot, characterized in that: include: A first acquisition module, wherein the first acquisition module acquires a real-time underwater terrain image based on the underwater dredging robot and pre-processes the acquired terrain image; A second acquisition module, wherein the second acquisition module acquires a light attenuation coefficient based on the shooting depth of the underwater dredging robot and the influence of plankton in the water; An adjustment module, wherein the adjustment module brightens and adjusts the grayscales of different pixels in the preprocessed terrain image based on the light attenuation coefficient; A decomposition module, wherein the decomposition module performs Laplace decomposition on the adjusted image to obtain a multi-layer terrain image; A screening module, wherein the screening module obtains a connected domain corresponding to a terrain image of any layer and screens out existing edge texture areas; A determination module, wherein the determination module obtains the weight of the importance of high-frequency information in each layer of scale image based on the proportion of edge texture areas in each layer of image, and determines the final defogging weight in combination with the illumination attenuation coefficient; A reconstruction module is provided, wherein the reconstruction module performs defogging and restoration on the image through a dark color prior based on the final defogging weight, and synthesizes image information of multiple scales into a complete image through inverse Laplace pyramid reconstruction, thereby obtaining underwater terrain information.

Images