Underwater production system-oriented attenuation scattering dual-path three-dimensional reconstruction method and system

Abstract

The invention belongs to the field of underwater three-dimensional reconstruction, and provides an attenuation scattering dual-path three-dimensional reconstruction method and system for an underwater production system, and the method comprises the steps: obtaining a multi-view underwater image for preprocessing, and obtaining an initial point cloud; performing three-dimensional Gaussian splashing processing on the initial point cloud to obtain a three-dimensional Gaussian distribution depth map; performing multi-scale depth perception scattering processing on the three-dimensional Gaussian distribution depth map by using an attenuation scattering dual-path three-dimensional reconstruction model to obtain a scattering map, and performing attenuation processing on the multi-view underwater image and the three-dimensional Gaussian distribution depth map in parallel to obtain an attenuation map; and performing physical imaging on the attenuation image and the scattering image to obtain a three-dimensional reconstruction result. According to the method, the geometric accuracy, the texture fidelity and the physical reality of a reconstruction result are remarkably improved by introducing a feature-guided double-path physical modeling framework and a scene self-adaptive mechanism.

Description

Technical Field

[0001] This invention belongs to the field of underwater three-dimensional reconstruction technology, specifically relating to a method and system for attenuated scattering dual-path three-dimensional reconstruction of underwater production systems. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In the wave of intelligent manufacturing, high-precision 3D reconstruction of industrial objects and scenarios is a crucial technological foundation for building digital twins and realizing automated operation and maintenance as well as intelligent quality inspection. High-precision 3D models provide accurate geometric and appearance data for product design, production simulation, defect detection, and autonomous robot operation, serving as a bridge between the physical and digital worlds. Remote inspection and intelligent maintenance based on high-precision 3D models are particularly important in extreme or inaccessible working environments.

[0004] High-precision 3D reconstruction plays a crucial role, especially in intelligent underwater production systems. For example, in marine ranching, high-precision 3D reconstruction of aquaculture cages, underwater sensors, and the ecological environment is fundamental to achieving automated feeding, growth status monitoring, and disease early warning. In the offshore energy sector, high-fidelity 3D reconstruction and defect detection of submarine cables and oil pipelines are core components for ensuring production safety and predictive maintenance. The core objective of these tasks is to accurately reconstruct the 3D geometric structure and texture details of underwater production systems using 2D image datasets taken from multiple perspectives, providing critical data support for intelligent production, operation, and management.

[0005] In recent years, traditional methods have primarily relied on Multi-View Stereo (MVS) theory to reconstruct 3D point clouds through steps such as feature matching, camera pose estimation, and dense matching. Deep learning methods, represented by Neural Radiance Fields (NeRF), have brought new ideas to underwater reconstruction. NeRF, through implicit neural representations and volumetric rendering techniques, can simulate the propagation of light in a medium, mitigating the negative effects of underwater optical effects to some extent. However, NeRF-like methods inherently suffer from extremely slow training and rendering speeds and enormous computational overhead, making them unsuitable for real-time applications. To pursue real-time rendering performance, the latest 3D Gaussian Splatting (3DGS) technology has been introduced into underwater reconstruction. 3DGS represents the scene using explicit 3D Gaussian primitives, achieving unprecedented rendering speed and high-quality detail.

[0006] However, these methods are all designed based on atmospheric optical models, and their core assumption—that "light travels in straight lines and its color is constant in a uniform transparent medium"—is completely broken in the underwater environment. First, because water selectively absorbs light wavelengths (i.e., longer wavelengths of red light attenuate first), underwater images generally suffer from severe color distortion and a bluish-green tint, greatly affecting the stable matching of feature points and the fidelity of the final reconstructed texture. Furthermore, suspended particles in the water (such as plankton and silt) produce complex scattering effects on light. Forward scattering blurs image details, while backscattering creates a "fog-like" background light, significantly reducing image contrast and signal-to-noise ratio, leading to geometric reconstruction errors or even collapse in areas at slightly greater distances.

[0007] In summary, existing methods struggle to effectively decouple the physically tightly coupled effects of attenuation and scattering. This leads to a systematic overestimation of attenuation and underestimation of scattering in turbid waters, and the opposite in clear waters, resulting in cumulative errors and ultimately causing distortions in both color and geometry. The models lack adaptability to diverse water bodies; from clear coral reefs to turbid nearshore waters, the optical parameters vary significantly, and existing methods mostly employ fixed parameter models, lacking the ability to automatically adjust optimization strategies based on scene characteristics, resulting in weak generalization capabilities. Multi-scale information representation has limitations; single-scale models struggle to simultaneously capture the high-frequency texture details of near-field objects and the macroscopic structural contours of the far-field environment, often leading to overly smoothed near-field scenes or loss of far-field structure. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes a dual-path 3D reconstruction method and system for attenuation and scattering in underwater production systems. This invention accurately decouples and models the complex physical processes of attenuation and scattering underwater, and can adaptively respond to diverse aquatic environments, thereby achieving truly high-fidelity and robust 3D reconstruction of underwater scenes and providing technical support for intelligent underwater operations.

[0009] According to some embodiments, the first aspect of the present invention provides a method for attenuated scattering dual-path three-dimensional reconstruction of underwater production systems, employing the following technical solution: A dual-path 3D reconstruction method for attenuated scattering in underwater production systems includes: A water body optical type classifier is trained by acquiring multi-view underwater images in different water areas. A scene adaptive optimization controller is used to generate a control signal based on the water body optical type classification result, and an adaptive loss for water body optical type is generated based on the control signal. Preprocessing is performed on multi-view underwater images that determine the optical type of the water body to obtain an initial point cloud; The initial point cloud was subjected to three-dimensional Gaussian splashing processing to obtain a three-dimensional Gaussian distribution depth map that determines the optical type of the water body. A multi-scale depth-sensing scattering processing method is used to obtain a scattering map by performing multi-scale scattering processing on a 3D Gaussian distribution depth map using a dual-path attenuation scattering model. In parallel, attenuation processing is performed on multi-view underwater images and 3D Gaussian distribution depth maps to obtain an attenuation map. The attenuated scattering dual-path 3D reconstruction model was trained using three-dimensional Gaussian splash loss, attenuated scattering consistency loss, depth-aware edge loss, multi-scale feature loss and water body optical type adaptive loss as the total loss, and the trained attenuated scattering dual-path 3D reconstruction model was obtained. The attenuation map and scattering map are obtained by using a trained attenuation and scattering dual-path 3D reconstruction model. The attenuation map and scattering map are then physically imaged to obtain the 3D reconstruction result.

[0010] Furthermore, underwater images from different water bodies are acquired from multiple perspectives to train a water body optical type classifier, specifically as follows: Acquire multi-view underwater images in different water areas; Multi-view underwater images are processed by a global average pooling layer to compress their global color information into feature vectors. A two-layer fully connected network is used to perform a non-linear mapping on the feature vector; By processing the nonlinear mapping results using activation functions, multi-view underwater images that determine the optical type of the water body are obtained.

[0011] Furthermore, attenuation processing is performed on the multi-view underwater images and the 3D Gaussian depth map to obtain an attenuation map, specifically:

[0012] in, This indicates that the attenuation model is based on RGB guidance, and the output is an attenuation map. It is a wavelength-dependent weight vector. It is the basic attenuation coefficient. It is an edge-sensing factor. It is the edge modulation intensity constant. It is a 3D Gaussian depth map. These are multi-view underwater images.

[0013] Furthermore, a multi-scale depth-sensing scattering processing method is used to process the 3D Gaussian distribution depth map using a decaying scattering dual-path 3D reconstruction model to obtain a scattering map, specifically:

[0014] in, It is a multi-scale depth-sensing scattering model that outputs a scattering map. It is the fundamental scattering coefficient. It is the depth confidence factor. It is the depth edge factor. Multi-scale feature weights, and It is the modulation factor. It is a 3D Gaussian depth map.

[0015] Furthermore, the calculation process of the multi-scale weights is as follows:

[0016] in, Represents the depth map of a three-dimensional Gaussian distribution. Perform downsampling times, , Indicates upsampling times, , Represents feature extraction functions at different scales. , This indicates the attention enhancement module.

[0017] Furthermore, the attenuation map and scattering map are physically imaged to obtain the three-dimensional reconstruction result, specifically:

[0018] in, It is the result of three-dimensional reconstruction. This is the actual radiation level. It is the attenuation map output by the attenuation path. It is the scattering map output by the scattering path. These are multi-view underwater images.

[0019] According to some embodiments, a second aspect of the present invention provides an attenuated scattering dual-path three-dimensional reconstruction system for underwater production systems, employing the following technical solution: A dual-path 3D reconstruction system for attenuated scattering in underwater production systems includes: The water classification module is configured to train a water optical type classifier using multi-view underwater images from different water bodies. The scene adaptive optimization controller generates a control signal based on the water optical type classification result and generates an adaptive loss for water optical type based on the control signal. The image preprocessing module is configured to preprocess multi-view underwater images that determine the optical type of the water body to obtain an initial point cloud. The 3D Gaussian splashing module is configured to perform 3D Gaussian splashing processing on the initial point cloud to obtain a 3D Gaussian distribution depth map that determines the optical type of the water body; The model training module is configured to use a decaying scattering dual-path 3D reconstruction model to perform multi-scale depth-sensing scattering processing on a 3D Gaussian distribution depth map to obtain a scattering map, and to perform decaying processing on multi-view underwater images and a 3D Gaussian distribution depth map in parallel to obtain a decaying map. The loss function construction module is configured to use three-dimensional Gaussian splash loss, attenuated scattering consistency loss, depth-aware edge loss, multi-scale feature loss and water body optical type adaptive loss as the total loss to train the attenuated scattering dual-path three-dimensional reconstruction model, and obtain the trained attenuated scattering dual-path three-dimensional reconstruction model. The 3D reconstruction module is configured to use a trained attenuation and scattering dual-path 3D reconstruction model to obtain attenuation and scattering maps, and then perform physical imaging on the attenuation and scattering maps to obtain the 3D reconstruction results.

[0020] According to some embodiments, a third aspect of the present invention provides a computer-readable storage medium.

[0021] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in the first embodiment above.

[0022] According to some embodiments, a fourth aspect of the present invention provides a computer device.

[0023] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in the first embodiment above.

[0024] According to some embodiments, a fifth aspect of the present invention provides a computer program product or computer program.

[0025] A computer program product or computer program includes computer instructions stored in a computer-readable storage medium, wherein a processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in the first embodiment above.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention significantly improves the geometric accuracy, texture fidelity, and physical realism of the reconstruction results by introducing a feature-guided dual-path physical modeling framework and a scene adaptation mechanism. This method not only accurately decouples and simulates attenuation and scattering optical effects in complex underwater environments but also maintains excellent generalization ability and stable performance in diverse water bodies ranging from clear to turbid. It not only promotes the development of 3D reconstruction technology in non-ideal homogeneous media but also provides technical support for intelligent underwater production systems such as marine ranching and underwater pipeline inspection, and offers high-fidelity, robust 3D reconstruction solutions for cutting-edge applications such as underwater autonomous navigation and environmental monitoring. Attached Figure Description

[0027] 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 improper limitation of the invention.

[0028] Figure 1 This describes the training process of the attenuated scattering dual-path three-dimensional reconstruction model in this embodiment of the invention. Figure 2 This is an overall architecture diagram of the attenuated scattering dual-path three-dimensional reconstruction model in an embodiment of the present invention. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0032] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0033] Example 1 This embodiment provides a method for attenuated scattering dual-path three-dimensional reconstruction of underwater production systems. This embodiment uses the application of this method to a server as an example for illustration. It is understood that this method can also be applied to terminals, and can also be applied to systems including terminals, servers, and other components, and can be implemented through interaction between the terminal and the server. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communication, middleware services, domain name services, CDN security services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. In this embodiment, the method includes the following steps: A water body optical type classifier is trained by acquiring multi-view underwater images in different water areas. A scene adaptive optimization controller is used to generate a control signal based on the water body optical type classification result, and an adaptive loss for water body optical type is generated based on the control signal. Preprocessing is performed on multi-view underwater images that determine the optical type of the water body to obtain an initial point cloud; The initial point cloud was subjected to three-dimensional Gaussian splashing processing to obtain a three-dimensional Gaussian distribution depth map that determines the optical type of the water body. A multi-scale depth-sensing scattering processing method is used to obtain a scattering map by performing multi-scale scattering processing on a 3D Gaussian distribution depth map using a dual-path attenuation scattering model. In parallel, attenuation processing is performed on multi-view underwater images and 3D Gaussian distribution depth maps to obtain an attenuation map. The attenuated scattering dual-path 3D reconstruction model was trained using three-dimensional Gaussian splash loss, attenuated scattering consistency loss, depth-aware edge loss, multi-scale feature loss and water body optical type adaptive loss as the total loss, and the trained attenuated scattering dual-path 3D reconstruction model was obtained. The attenuation map and scattering map are obtained by using a trained attenuation and scattering dual-path 3D reconstruction model. The attenuation map and scattering map are then physically imaged to obtain the 3D reconstruction result.

[0034] like Figure 1 As shown, the training process of the attenuated scattering dual-path 3D reconstruction model in this embodiment is as follows: Step S1: Obtain a multi-view underwater image dataset and divide it proportionally. Train a water body optical type classifier as follows: Step S1.1: Collect multi-view underwater image datasets. To fully evaluate the model's performance in various underwater scenarios, multi-view underwater image datasets were collected from different water bodies. These datasets comprehensively cover a wide range of water optical types and depth variations, from clear to moderately turbid, and from near-view coral reefs to distant seabeds, ensuring the diversity and representativeness of the datasets and enabling effective evaluation of the model's performance and robustness in complex underwater environments.

[0035] The multi-view underwater image dataset is divided into training, testing, and validation sets according to a set ratio.

[0036] Step S1.2: Construct a lightweight water body optical type classifier. Different water bodies have significantly different optical properties, and using a fixed single parameter model is difficult to adapt to diverse underwater environments. To address this issue, this embodiment proposes and constructs a lightweight water body optical type classifier, as follows:

[0037] in, For water body optical type classifier, The input consists of multi-view underwater images in RGB image format. and These are the weight parameters of the water body optical type classifier.

[0038] Specifically, the processing procedure of the water body optical type classifier is as follows: RGB images (multi-view underwater images) First, the global color information is compressed into a compact feature vector by a global average pooling layer (AvgPool). Then, this feature vector is fed into a two-layer fully connected network (MLP) with a ReLU activation function for non-linear mapping. Finally, a Softmax activation function is used to output the probability distribution of the current scene belonging to various predefined water body optical types (such as "clear", "turbid", etc.). Then, by taking the category corresponding to the maximum value in the probability distribution (i.e., Argmax operation), the unique water body optical type of the scene is determined, thus obtaining the final classification result.

[0039] Step S1.3: As Figure 2 As shown, this embodiment also proposes a scene-adaptive optimization controller for training and optimizing the water body optical type classifier. During training, the training set is input into the water body optical type classifier, and the output water body optical type prior will be utilized by the controller.

[0040] Specifically, the controller automatically identifies different water body optical types based on prior knowledge of water body optical types, and dynamically adjusts two key aspects during subsequent training: first, it fine-tunes the model's learning rate, for example, appropriately reducing the learning rate in clear water to promote model convergence to more accurate results; second, it dynamically allocates the weights of the water body optical type adaptive loss function proposed in this embodiment, for example, increasing the constraint on attenuation loss in clear water and strengthening the constraint on scattering loss in turbid water.

[0041] The scene adaptive optimization controller generates control signals based on the water body optical type classification results, and generates water body optical type adaptive loss based on the control signals. By constructing a water body optical type classifier, it aims to automatically identify the water body optical type of the current scene, providing dynamic and adaptive prior guidance for subsequent physical modeling and loss function optimization.

[0042] Step S2: Preprocess the multi-view underwater images that determine the optical type of the water body to obtain the initial point cloud.

[0043] Specifically, this embodiment preprocesses multi-view underwater images to determine the optical type of the water body. This process automatically completes feature point extraction and matching, camera pose estimation, and 3D point cloud generation of the scene to obtain initial point clouds for different optical types of water bodies.

[0044] During training, this step obtains the geometric prior information necessary for subsequent 3D reconstruction. The final output of accurate camera intrinsic and extrinsic parameters and initial point cloud data will serve as input for the subsequent physical modeling and optimization stages of this method. This step utilizes, but is not limited to, motion reconstruction techniques to preprocess multi-view underwater images to obtain scene point cloud data and accurate camera intrinsic and extrinsic parameters.

[0045] Step S3: Based on the initial point cloud that determines the optical type of the water body, train the attenuated scattering dual-path 3D reconstruction model, as follows: It should be noted that the attenuated scattering dual-path 3D reconstruction model includes a 3D Gaussian splash model and a parallel dual-path physical modeling framework.

[0046] Step S3.1: Build a parallel dual-path physics modeling framework. For example... Figure 2 As shown, traditional underwater physical imaging models are relatively simple and cannot effectively cope with the complexity of underwater scenes, especially when dealing with edges, depth-varying regions, and different water conditions. This embodiment proposes a parallel dual-path physical modeling framework, which includes an RGB-guided attenuation model and a multi-scale scattering model, accurately decomposing the underwater optical propagation process into two key physical processes: attenuation and scattering.

[0047] Step 3.2: Construct an RGB-guided attenuation model.

[0048] First, to address the color distortion problem caused by wavelength-selective attenuation, this embodiment defines an attenuation model based on RGB guidance:

[0049] in, This indicates that it is based on the RGB guided decay model. It is a wavelength-dependent weight vector. It is the basic attenuation coefficient. It is an edge-sensing factor. It is the edge modulation intensity constant. It is a 3D Gaussian depth map. These are multi-view underwater images.

[0050] After the initial point cloud is processed by the 3D Gaussian splash model, a 3D Gaussian distribution depth map with a defined water optical type is obtained. Then, through the designed RGB feature extraction network, color distribution and texture information are extracted from the multi-view underwater images and the 3D Gaussian distribution depth map to ensure that the attenuation model can identify color distortion areas in the scene caused by wavelength selective attenuation. Combining this RGB information and depth features, the model can more accurately estimate the true color of each area.

[0051] Step 3.3: Construct a multi-scale depth-sensing scattering model.

[0052] Considering the scattering effect caused by underwater suspended particles, which leads to a decrease in reconstruction quality, this embodiment defines a multi-scale depth-sensing scattering model, as follows:

[0053] in, It is a multi-scale depth-sensing scattering model that outputs a scattering map. It is the fundamental scattering coefficient. It is the depth confidence factor. It is the depth edge factor. Multi-scale feature weights, and It is the modulation factor.

[0054] Among them, for Multi-scale feature weighting: This embodiment addresses the difficulty of single-scale models in accurately handling both near-field details and far-field overall characteristics simultaneously. It introduces local structural features and captures depth information at different scales through a feature pyramid network. This includes three feature extraction branches processing original, 1 / 2, and 1 / 4 resolutions, and defines a multi-scale feature calculation formula:

[0055] in, Represents the depth map of a three-dimensional Gaussian distribution. Perform downsampling times, , Indicates upsampling times, , Represents feature extraction functions at different scales. , This indicates the attention enhancement module.

[0056] This step aims to forcibly decouple the complex underwater optical effects into two independent physical processes: attenuation and scattering, to achieve refined modeling. After processing in step S3, attenuation and scattering maps are obtained.

[0057] Step S4: Based on the attenuation map output by the attenuation path and the scattering map output by the scattering path, the two maps are synthesized using an underwater physical imaging model to obtain a three-dimensional reconstruction result.

[0058] To achieve high-fidelity rendering of underwater scenes, after modeling the RGB-based attenuation model and the multi-scale scattering model using feature enhancement methods, this embodiment, based on a traditional underwater physical imaging model, compares the modeling results of the two physical processes with the actual radiometric properties of the scene. To combine them organically:

[0059] in, It is the result of three-dimensional reconstruction, that is, the underwater image after observation. It is the true radiometric value, which is the clear image of a scene under conditions where there is no water. It is the attenuation map output by the attenuation path. It is the scattering map output by the scattering path. It provides multi-view underwater images. This dual-path physical modeling mechanism avoids mutual interference of parameters by decoupling the optimization processes of attenuation and scattering. At the same time, it introduces a variety of feature-guided strategies to enhance the model's expressive power: RGB image-guided attenuation optimization and depth information-guided scattering optimization. This enables accurate simulation of optical phenomena in the underwater environment, significantly improving reconstruction quality and physical accuracy.

[0060] Based on steps S1-S4 above, this embodiment combines the physical model with the 3D Gaussian Splash (3DGS) rendering mechanism to construct a complete, end-to-end differentiable rendering pipeline from the Gaussian primitive parameters of the 3D scene to the final 2D rendered image, achieving high-fidelity scene reconstruction. This allows the gradient signal generated by the loss between the rendered image and the real image in subsequent optimization steps to propagate unimpeded back to every parameter (position, color, etc.) of each initial Gaussian primitive, thereby achieving precise and automatic optimization of the entire 3D scene.

[0061] Step S5: Construct various loss functions for the training process, and optimize the training process of the attenuated scattering dual-path 3D reconstruction model based on the constructed loss functions, specifically as follows: A dual-path 3D reconstruction model for attenuated scattering of underwater production systems has been built. The model uses attenuated scattering consistency loss to ensure the correctness of physical laws; water body adaptive loss to dynamically adjust weights; edge perception loss to maintain the clarity of object structure; and multi-scale feature loss to take into account both global structure and local details.

[0062] As depth increases, the scattering effect of water intensifies, while the attenuation of directly transmitted light also increases. Due to the lack of physical constraints, near-field objects exhibit high scattering and low attenuation, while distant objects exhibit low scattering and high attenuation—a phenomenon not inherent to physical systems. To address this challenge, this embodiment defines an attenuation-scattering uniformity loss:

[0063] in, Indicates the expected value. Represents the scattering component. Represents a 3D Gaussian depth map. Indicates the attenuation component. This represents the transmittance map, which shows the proportion of the original light rays that successfully pass through the medium and reach the camera in an underwater scene. It is the mean squared error loss function. It's a hyperparameter.

[0064] To achieve adaptive optimization of the model for different water bodies, this embodiment proposes an adaptive loss function based on water body optical type:

[0065] Among them, the weighting coefficient and Water optical type Dynamically determined, It is the loss weighting coefficient. It measures the scattering diagram The reconstruction error caused by inaccuracies is not an independently defined loss function, but is determined by weighting coefficients. The model can dynamically adjust the emphasis on optimizing the scattering process based on the optical type of the water body.

[0066] This embodiment uses a defined water body optical type label. This guides how to better train the unique 3D Gaussian distribution model in this embodiment. During training, the 3D Gaussian distribution depth map rendered from this model will be used repeatedly. This is used to calculate various physical effects and losses, representing their information that collectively affects the entire subsequent optimization process.

[0067] The purpose of depth-sensing edge loss is to address the problems of blurred object edges and structural distortion caused by underwater scattering effects.

[0068] in, The depth edge weights are dynamically allocated based on the magnitude of the local depth gradient. To smooth out constraint weights, used to maintain the sharpness of structural edges, Represents the scattering component. It is the absolute value difference loss function.

[0069] To ensure that the reconstruction results simultaneously capture both the macroscopic global structure and the microscopic local details, this embodiment employs a multi-scale feature loss mechanism:

[0070] in, It's rendering an image. It is a real image. This is a set of scale factors for downsampling, including original resolution, 1 / 2 resolution, and 1 / 4 resolution. These are multi-scale weighting coefficients. Indicated in scale The input image and the rendered image are downsampled. This indicates calculation at the original resolution. The loss term, followed by the term, indicates the loss calculated at each downsampling scale.

[0071] The total loss is calculated using three-dimensional Gaussian splash loss, attenuated scattering uniformity loss, depth-aware edge loss, multi-scale feature loss, and water body optical type adaptive loss, as detailed below:

[0072] in, It is a 3D Gaussian splash loss; Weighting coefficients for five different losses.

[0073] The attenuated scattering dual-path 3D reconstruction model is trained by iteratively repeating steps S2-S4 using the training set. The loss function in step S5 is used to adjust the parameters of the attenuated scattering dual-path 3D reconstruction model in each iteration until the attenuated scattering dual-path 3D reconstruction model converges, thus obtaining the trained attenuated scattering dual-path 3D reconstruction model.

[0074] The trained attenuated scattering dual-path 3D reconstruction model was tested and validated using the test set and validation set.

[0075] Using the trained attenuated scattering dual-path 3D reconstruction model, repeat steps S2-S4 to perform 3D reconstruction on multi-view underwater images and obtain the 3D reconstruction results.

[0076] Example 2 This embodiment provides an attenuated scattering dual-path three-dimensional reconstruction system for underwater production systems, including: The water classification module is configured to train a water optical type classifier using multi-view underwater images from different water bodies. The scene adaptive optimization controller generates a control signal based on the water optical type classification result and generates an adaptive loss for water optical type based on the control signal. The image preprocessing module is configured to preprocess multi-view underwater images that determine the optical type of the water body to obtain an initial point cloud. The 3D Gaussian splashing module is configured to perform 3D Gaussian splashing processing on the initial point cloud to obtain a 3D Gaussian distribution depth map that determines the optical type of the water body; The model training module is configured to use a decaying scattering dual-path 3D reconstruction model to perform multi-scale depth-sensing scattering processing on a 3D Gaussian distribution depth map to obtain a scattering map, and to perform decaying processing on multi-view underwater images and a 3D Gaussian distribution depth map in parallel to obtain a decaying map. The loss function construction module is configured to use three-dimensional Gaussian splash loss, attenuated scattering consistency loss, depth-aware edge loss, multi-scale feature loss and water body optical type adaptive loss as the total loss to train the attenuated scattering dual-path three-dimensional reconstruction model, and obtain the trained attenuated scattering dual-path three-dimensional reconstruction model. The 3D reconstruction module is configured to use a trained attenuation and scattering dual-path 3D reconstruction model to obtain attenuation and scattering maps, and then perform physical imaging on the attenuation and scattering maps to obtain the 3D reconstruction results.

[0077] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1 above. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0078] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0079] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0080] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in Embodiment 1 above.

[0081] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in Embodiment 1 above.

[0082] Example 5 This embodiment provides a computer program product or computer program, including computer instructions stored in a computer-readable storage medium. The processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems described in Embodiment 1 above.

[0083] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0084] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0085] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0086] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0087] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0088] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for attenuated scattering dual-path three-dimensional reconstruction of underwater production systems, characterized in that, include: A water body optical type classifier is trained by acquiring multi-view underwater images in different water areas. A scene adaptive optimization controller is used to generate a control signal based on the water body optical type classification result, and an adaptive loss for water body optical type is generated based on the control signal. Preprocessing is performed on multi-view underwater images that determine the optical type of the water body to obtain an initial point cloud; The initial point cloud was subjected to three-dimensional Gaussian splashing processing to obtain a three-dimensional Gaussian distribution depth map that determines the optical type of the water body. A multi-scale depth-sensing scattering processing method is used to obtain a scattering map by performing multi-scale scattering processing on a 3D Gaussian distribution depth map using a dual-path attenuation scattering model. In parallel, attenuation processing is performed on multi-view underwater images and 3D Gaussian distribution depth maps to obtain an attenuation map. The attenuated scattering dual-path 3D reconstruction model was trained using three-dimensional Gaussian splash loss, attenuated scattering consistency loss, depth-aware edge loss, multi-scale feature loss and water body optical type adaptive loss as the total loss, and the trained attenuated scattering dual-path 3D reconstruction model was obtained. The attenuation map and scattering map are obtained by using a trained attenuation and scattering dual-path 3D reconstruction model. The attenuation map and scattering map are then physically imaged to obtain the 3D reconstruction result.

2. The attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in claim 1, characterized in that, The water body optical type classifier is trained by acquiring multi-view underwater images from different water bodies. Specifically: Acquire multi-view underwater images in different water areas; Multi-view underwater images are processed by a global average pooling layer to compress their global color information into feature vectors. A two-layer fully connected network is used to perform a non-linear mapping on the feature vector; By processing the nonlinear mapping results using activation functions, multi-view underwater images that determine the optical type of the water body are obtained.

3. The attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in claim 1, characterized in that, Attenuation processing is performed on multi-view underwater images and 3D Gaussian depth maps to obtain attenuation maps, specifically: in, This indicates that the attenuation model is based on RGB guidance, and the output is an attenuation map. It is a wavelength-dependent weight vector. It is the basic attenuation coefficient. It is an edge-sensing factor. It is the edge modulation intensity constant. It is a 3D Gaussian depth map. These are multi-view underwater images.

4. The attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in claim 1, characterized in that, A multi-scale depth-sensing scattering map is obtained by performing multi-scale depth-sensing scattering processing on a 3D Gaussian distribution depth map using a dual-path attenuated scattering model. Specifically: in, It is a multi-scale depth-sensing scattering model that outputs a scattering map. It is the fundamental scattering coefficient. It is the depth confidence factor. It is the depth edge factor. Multi-scale feature weights, and It is the modulation factor. It is a 3D Gaussian depth map.

5. The attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in claim 4, characterized in that, The calculation process of the multi-scale weights is as follows: in, Represents the depth map of a three-dimensional Gaussian distribution. Perform downsampling times, , Indicates upsampling times, , Represents feature extraction functions at different scales. , This indicates the attention enhancement module.

6. The attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in claim 1, characterized in that, Physical imaging is performed on the attenuation map and scattering map to obtain the three-dimensional reconstruction result, specifically: in, It is the result of three-dimensional reconstruction. This is the actual radiation level. It is the attenuation map output by the attenuation path. It is the scattering map output by the scattering path. These are multi-view underwater images.

7. A dual-path attenuated scattering three-dimensional reconstruction system for underwater production systems, characterized in that, include: The water classification module is configured to train a water optical type classifier using multi-view underwater images from different water bodies. The scene adaptive optimization controller generates a control signal based on the water optical type classification result and generates an adaptive loss for water optical type based on the control signal. The image preprocessing module is configured to preprocess multi-view underwater images that determine the optical type of the water body to obtain an initial point cloud. The 3D Gaussian splashing module is configured to perform 3D Gaussian splashing processing on the initial point cloud to obtain a 3D Gaussian distribution depth map that determines the optical type of the water body; The model training module is configured to use a decaying scattering dual-path 3D reconstruction model to perform multi-scale depth-sensing scattering processing on a 3D Gaussian distribution depth map to obtain a scattering map, and to perform decaying processing on multi-view underwater images and a 3D Gaussian distribution depth map in parallel to obtain a decaying map. The loss function construction module is configured to use three-dimensional Gaussian splash loss, attenuated scattering consistency loss, depth-aware edge loss, multi-scale feature loss and water body optical type adaptive loss as the total loss to train the attenuated scattering dual-path three-dimensional reconstruction model, and obtain the trained attenuated scattering dual-path three-dimensional reconstruction model. The 3D reconstruction module is configured to use a trained attenuation and scattering dual-path 3D reconstruction model to obtain attenuation and scattering maps, and then perform physical imaging on the attenuation and scattering maps to obtain the 3D reconstruction results.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in any one of claims 1-6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps in the attenuated scattering dual-path three-dimensional reconstruction method for underwater production systems as described in any one of claims 1-6.

Images