High-precision underwater structured light three-dimensional point cloud imaging method and system

Abstract

The invention relates to the technical field of three-dimensional imaging, and discloses a high-precision underwater structured light three-dimensional point cloud imaging method and system, and the method comprises the following steps: deploying a plurality of sensor nodes, monitoring the temperature, salinity and turbidity key parameters of a water body in real time, selecting proper light source wavelength and power according to the collected data, and carrying out the three-dimensional point cloud imaging of the underwater structured light. It is ensured that the optimal signal-to-noise ratio is obtained under the current condition, and meanwhile the exposure time and gain setting of the camera are initialized. According to the method, the optimal signal-to-noise ratio is obtained by monitoring water body conditions in real time and dynamically adjusting light source characteristics, the optimal structured light pattern is predicted by using a machine learning model, scattering noise is removed, advanced registration and fusion algorithms and an accurate phase unwrapping method are applied, and a high-quality comprehensive structured light pattern is generated. Point cloud gaps are filled through a geometric correction algorithm and an interpolation / filtering algorithm, so that a reconstruction model is closer to the surface of a real object.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional imaging, and more specifically, to a high-precision underwater structured light three-dimensional point cloud imaging method and system. Background Art

[0002] With the development of environmental protection, resource development and other fields, the demand for technologies that can accurately obtain three-dimensional information of underwater environments is growing. Although traditional sonar technology and LiDAR can meet these needs to a certain extent, they have limitations in resolution, speed and detail capture. In contrast, structured light 3D imaging technology has shown great potential in underwater applications due to its high precision, real-time and flexibility.

[0003] Due to the complexity of the water medium itself, including but not limited to water type (such as fresh water, salt water), changes in optical properties (such as turbidity), and scattering effects caused by suspended particles, structured light 3D imaging faces many challenges. Summary of the invention

[0004] The present invention provides a high-precision underwater structured light three-dimensional point cloud imaging method and system to solve the technical problems in related technologies.

[0005] The present invention provides a high-precision underwater structured light three-dimensional point cloud imaging method, comprising the following steps:

[0006] S100, environmental monitoring and parameter initialization: deploy multiple sensor nodes to monitor the key parameters of water temperature, salinity and turbidity in real time, select the appropriate light source wavelength and power based on the collected data to ensure the best signal-to-noise ratio under the current conditions, and initialize the camera exposure time and gain settings;

[0007] S200, dynamic light source adjustment and multi-band emission: based on real-time feedback of water conditions, dynamically adjusts light source characteristics to optimize imaging effects, and uses light of different wavelengths to alternately emit structured light patterns;

[0008] S300, structured light pattern generation and optimization: predicts the optimal structured light pattern through a machine learning model that takes into account the scattering effect and absorption characteristics of the current water conditions;

[0009] S400, image acquisition and preprocessing: uses a camera array equipped with a fast shutter mechanism to synchronously capture structured light patterns emitted and reflected by multi-band light sources, and applies deep learning algorithms to remove scattered noise;

[0010] S500, data fusion and phase unwrapping: align and fuse image data of different frequency bands to generate a comprehensive structured light pattern, extract complete phase information from the fused image, and then calculate the three-dimensional coordinates of the object surface;

[0011] S600, anti-interference 3D reconstruction: Taking into account possible non-ideal factors, a geometric correction algorithm is applied to further improve the reconstruction accuracy, and gaps in the point cloud are filled through interpolation or filtering algorithms;

[0012] S700, performance evaluation and feedback adjustment: Comprehensively evaluate the reconstructed 3D point cloud, including point cloud density, geometric error, and surface detail retention, and automatically adjust system parameters based on the evaluation results to form a closed-loop control mechanism.

[0013] Furthermore, S100 also includes the following contents:

[0014] S110, deploying the sensor network: select multiple representative monitoring points based on the geographical features of the target underwater area, cover the entire imaging range, and distribute them evenly. Each node is equipped with a thermometer, conductivity meter, and turbidity meter;

[0015] S120, real-time environmental monitoring: start the sensor and start collecting temperature at set time intervals ,salinity and turbidity Set thresholds for the data to trigger alarms when the monitored data exceeds the normal range, alerting you to potential problems.

[0016] S130, data processing and analysis: Apply filtering methods to the original collected data to remove high-frequency noise, obtain a more stable trend line, and calculate the average value and standard deviation statistics over a period of time.

[0017] Furthermore, S200 also includes the following contents:

[0018] S210, real-time water condition feedback: Get the latest temperature from the sensor network deployed by S100 ,salinity and turbidity Data, based on real-time data, evaluates the impact of the current water environment on light propagation;

[0019] S220, adaptive light source adjustment: select the most suitable light source wavelength according to water conditions , in high turbidity environments, longer wavelengths are chosen to reduce the effects of scattering, while in low turbidity environments, shorter wavelengths are chosen to achieve higher resolution;

[0020] Adjust the output power of the light source according to the calculated optimal signal-to-noise ratio SNR , providing sufficient lighting intensity without causing excessive scattering or reflection;

[0021] S230, multi-band emission design: determine multiple working frequency bands, each with different penetration capabilities and scattering characteristics, and design the emission sequence of the structured light pattern so that light of different frequency bands can be emitted sequentially in a short period of time.

[0022] Furthermore, S300 also includes the following contents:

[0023] S310, Data Preparation and Feature Extraction: Get the latest water condition data from S100 and S200, including temperature ,salinity , turbidity Key parameters, based on the above data, calculate indicators describing the current optical properties of the water body;

[0024] At the same time, other factors that may affect the imaging effect are extracted;

[0025] S320, Intelligent Pattern Generation: uses a pre-trained model that has learned the best design principles for structured light patterns under different water conditions;

[0026] S330, optimization algorithm application: determine the design space of structured light patterns, i.e., all possible encoding vectors The set is realized by random sampling or grid division, and the algorithm is used to find the optimal solution that maximizes the objective function in the search space. In each iteration, the wavelength and power of the light source are adjusted according to the current water conditions.

[0027] S340, Pattern Verification and Fine-tuning: Use simulation tools to virtually test the generated structured light pattern to evaluate its performance under specific water conditions. If conditions permit, deploy the initially selected structured light pattern in a real environment and collect image data through experiments for further verification. Based on the results of simulation and field tests, make necessary fine-tuning to the structured light pattern until it meets the expected performance requirements.

[0028] Furthermore, the architecture of the pre-trained model is as follows:

[0029] Input layer: receives the environment feature vector , which contains the temperature ,salinity , turbidity Water condition parameters;

[0030] Convolutional layer: Use multiple convolution kernels to locally perceive the input features and extract spatial feature representations;

[0031] Pooling layer: reduces the spatial dimension of the feature map through downsampling operations while retaining key information;

[0032] Fully connected layer: maps the features extracted by the convolutional layer and the pooling layer to the final output, the structured light pattern encoding vector ;

[0033] Output layer: Generate structured light pattern encoding vector , used to control the emission characteristics of the light source;

[0034] The calculation formula of the pre-trained model is as follows:

[0035] Forward propagation: Assume that the weight of the lth layer of the model is , the bias is , the activation function is , then the output of the l+1th layer It is expressed as:

[0036] ;

[0037] in is the input feature vector;

[0038] in represents the weight matrix, represents the bias vector; is the activation function ReLU function;

[0039] Loss function: Use mean square error as the loss function to measure the predicted value and the true value The difference between:

[0040] ;

[0041] in is the sample size;

[0042] Backpropagation and optimization: The gradient is calculated through the backpropagation algorithm and the optimizer is used to update the model parameters to minimize the loss function.

[0043] Furthermore, S400 also includes the following contents:

[0044] S410, high-speed image capture: ensuring that all cameras are properly initialized and that exposure time and gain are set according to the parameters determined in S100, and checking the synchronization mechanism between the cameras to be able to capture the structured light pattern simultaneously;

[0045] Coordinate the working sequence of the light source and the camera to ensure that the camera can immediately capture the reflected light signal after each light source emission. For multi-band emission, use a reasonable alternating sequence;

[0046] S420, image data acquisition: achieving precise synchronization of light source emission and camera exposure through an external trigger or an internal timer;

[0047] Continuously collect multiple frames of images within a certain period of time for subsequent statistical analysis and noise suppression;

[0048] S430, Denoising and Enhancement: Apply pre-trained models to remove scattered noise;

[0049] The specific formula of the model is:

[0050] ;

[0051] in is the original acquired image, and Represents a high-quality image after processing. represents a deep neural network;

[0052] Super-resolution reconstruction: Use super-resolution reconstruction technology to improve image resolution and restore more details;

[0053] S440, image quality assessment: evaluate the denoising effect, calculate the signal-to-noise ratio of the image before and after processing, and ensure that the noise is effectively suppressed;

[0054] Contrast measurement: Quantify image contrast to ensure that the imaging system can maintain good visual effects in different water conditions.

[0055] Furthermore, S500 also includes the following contents:

[0056] S510, multi-band data registration: pre-process the original image of each frequency band, including correcting geometric distortion and color balance, using corner detection or edge detection algorithms to find stable feature points in each image, using feature point matching methods to find the correspondence between images of different frequency bands, and aligning all images to the same coordinate system through affine transformation or perspective transformation;

[0057] S520, data fusion: assign a weight to each frequency band based on the quality of the image in each frequency band , the multi-band images are fused into a comprehensive structured light pattern by weighted averaging;

[0058] The specific formula is as follows:

[0059] ;

[0060] in is the fused image, The image of the i-th frequency band is a sequence of images captured after being emitted and reflected by a multi-band light source. Represents the weight factor, which is a non-negative real number. It indicates the relative importance of the i-th frequency band image in the fusion process and needs to be dynamically adjusted according to the quality of each frequency band image. is the total number of frequency bands;

[0061] S530, Refractive Index Correction: Temperature-dependent and salinity The measured value of the water body is calculated using the empirical formula The formula is:

[0062] ;

[0063] here , , is a predetermined empirical constant;

[0064] Considering the optical path deviation caused by the refraction effect, adjust the camera intrinsic parameter matrix K;

[0065] S540, Phase Unwrapping: Extract the complete phase information from the fused image, assuming the camera intrinsic matrix is ​​known , then the three-dimensional coordinates [X, Y, Z] are calculated by the following formula:

[0066]

[0067] in is the pixel position, It is the distance value of the corresponding point, which refers to the actual distance from the camera to a certain point on the surface of the object.

[0068] Furthermore, S600 also includes the following contents:

[0069] S610, geometric error correction: using a pre-calibrated camera intrinsic parameter matrix and distortion coefficient, geometric distortion correction is performed on the captured image, and the posture of the camera relative to the object surface is estimated by a multi-view geometry method;

[0070] S620, point cloud density optimization: use interpolation algorithms to fill sparse areas in the point cloud, apply filtering algorithms to remove isolated noise points in the point cloud, smooth surface details, and maintain edge features. For a target point, find its nearest known point and assign the attribute value of the point to the target point. For each point, take a weighted average based on the spatial distance and brightness difference between it and its neighboring points, and calculate the 3D coordinates based on the distance value obtained by the camera intrinsic parameter matrix and phase unwrapping.

[0071] S630, surface detail preservation: uses local feature enhancement technology to highlight the key structure and texture information of the object surface, combines filters of different scales or transform domain methods to process point cloud data at different resolution levels.

[0072] The present invention provides a high-precision underwater structured light three-dimensional point cloud imaging system, comprising:

[0073] Environmental monitoring and parameter initialization module: deploy multiple sensor nodes to monitor key parameters such as water temperature, salinity, and turbidity in real time, select the appropriate light source wavelength and power based on the collected data to ensure the best signal-to-noise ratio under current conditions, and initialize the camera exposure time and gain settings;

[0074] Dynamic light source adjustment and multi-band emission module: Based on real-time feedback of water conditions, the light source characteristics are dynamically adjusted to optimize the imaging effect, and different wavelengths of light are used to alternately emit structured light patterns;

[0075] Structured light pattern generation and optimization module: Predicts the optimal structured light pattern through a pre-trained machine learning model, taking into account scattering effects and absorption characteristics to maximize signal-to-noise ratio and contrast;

[0076] Image acquisition and preprocessing module: uses a camera array equipped with a fast shutter mechanism to synchronously capture structured light patterns emitted and reflected by multi-band light sources, applies deep learning algorithms to remove scattered noise, and improves image quality through super-resolution reconstruction technology;

[0077] Data fusion and phase unwrapping module: aligns and fuses image data of different frequency bands to generate a comprehensive structured light pattern, extracts complete phase information from the fused image, and then calculates the three-dimensional coordinates of the object surface;

[0078] Anti-interference 3D reconstruction module: Taking into account possible non-ideal factors, the geometric correction algorithm is applied to further improve the reconstruction accuracy, and the gaps in the point cloud are filled through interpolation or filtering algorithms;

[0079] Performance evaluation and feedback adjustment module: Comprehensively evaluate the reconstructed 3D point cloud, including point cloud density, geometric error, and surface detail retention, and automatically adjust system parameters based on the evaluation results to form a closed-loop control mechanism.

[0080] The present invention also provides a storage medium storing non-temporary computer-readable instructions for executing one or more steps in the aforementioned high-precision underwater structured light three-dimensional point cloud imaging method.

[0081] The beneficial effects of the present invention are:

[0082] The present invention ensures the best signal-to-noise ratio under various water conditions by monitoring water conditions in real time and dynamically adjusting light source characteristics. It uses a machine learning model to predict the optimal structured light pattern, combines a deep learning algorithm to remove scattered noise, applies advanced registration and fusion algorithms, and precise phase unwrapping methods to generate high-quality comprehensive structured light patterns. It fills point cloud gaps through geometric correction algorithms and interpolation / filtering algorithms, greatly improving the accuracy and smoothness of 3D reconstruction and making the reconstructed model closer to the real object surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 It is a flow chart of a high-precision underwater structured light three-dimensional point cloud imaging method proposed by the present invention;

[0084] Figure 2 It is a structural block diagram of a high-precision underwater structured light three-dimensional point cloud imaging system proposed by the present invention;

[0085] Figure 3 It is a process guide diagram of a high-precision underwater structured light three-dimensional point cloud imaging method proposed by the present invention.

[0086] In the figure: 101, environmental monitoring and parameter initialization module; 102, dynamic light source adjustment and multi-band emission module; 103, structured light pattern generation and optimization module; 104, image acquisition and preprocessing module; 105, data fusion and phase unwrapping module; 106, anti-interference three-dimensional reconstruction module; 107, performance evaluation and feedback adjustment module. DETAILED DESCRIPTION

[0087] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed can be changed without departing from the scope of protection of the present specification. Each example can omit, replace or add various processes or components as needed. In addition, the features described relative to some examples can also be combined in other examples.

[0088] like Figure 1 and Figure 3 As shown, a high-precision underwater structured light three-dimensional point cloud imaging method includes the following steps:

[0089] S100, Environmental Monitoring and Parameter Initialization: Deploy multiple sensor nodes to monitor key parameters such as water temperature, salinity, and turbidity in real time. Select the appropriate light source wavelength λ and power P based on the collected data to ensure the best signal-to-noise ratio (SNR) under the current conditions. At the same time, initialize the camera exposure time and gain settings.

[0090] In one embodiment of the present invention, the following specific steps are also included:

[0091] S110, deploying sensor networks:

[0092] Select suitable locations: Select multiple representative monitoring points based on the geographical features of the target underwater area, ensuring that these points can cover the entire imaging range and are evenly distributed;

[0093] Install sensor nodes: Each node is equipped with a thermometer, conductivity meter (for measuring salinity), turbidity meter and other equipment;

[0094] It should be noted that the sensor should be waterproof, pressure-resistant, and able to work stably for a long time;

[0095] Data transmission and synchronization: Establish a reliable communication link so that the data of each sensor node can be transmitted to the central control system in real time, ensuring consistent timestamps for subsequent data analysis.

[0096] S120, real-time environmental monitoring:

[0097] Continuous data collection: Start the sensor and start collecting temperature at set time intervals ,salinity and turbidity data;

[0098] Anomaly detection and alarm: Set thresholds to trigger alarms when monitored data exceeds normal ranges, alerting operators to potential problems;

[0099] S130, Data processing and analysis

[0100] Smoothing filter: Apply low-pass filter or other appropriate filtering methods to the original collected data to remove high-frequency noise and obtain a smoother trend line;

[0101] Statistical analysis: Calculate the average value, standard deviation and other statistics over a period of time to describe the overall state of water conditions;

[0102] In order to choose the most suitable light source wavelength and power , the following factors need to be considered:

[0103] Light attenuation coefficient : It reflects the degree of energy lost due to absorption and scattering when light propagates in water and can be expressed as:

[0104] ;

[0105] in is the total attenuation coefficient, is the temperature ,salinity and turbidity Function is the absorption coefficient, which depends on temperature and salinity is the scattering coefficient, which is mainly affected by turbidity Influence;

[0106] Signal-to-noise ratio (SNR): defined as the ratio of signal intensity to background noise, for structured light imaging systems, it can be estimated by the following formula:

[0107] ;

[0108] in is the signal-to-noise ratio, is the wavelength and power The function of Indicates a given wavelength and power The signal strength under is the standard deviation of the background noise; the wavelength of the light source : Choose one or more wavelengths that are within the visible spectrum (e.g. 450nm to 650nm) or extend into the near infrared region (e.g. 850nm). The specific choice depends on the water conditions and the intended application requirements.

[0109] Light source power It determines the total energy output of the emitted light. The power level needs to be balanced between sufficient illumination brightness and avoiding overexposure.

[0110] Camera exposure time : Controls the length of time the camera's sensor receives light. Longer exposure times can brighten images in low-light environments, but may also introduce motion blur.

[0111] Camera Gain : Adjusts the gain level of the camera's internal amplifier circuit. Higher gain helps capture weak signals, but also amplifies noise.

[0112] In one embodiment of the present invention, the parameter initialization process is as follows:

[0113] Select initial parameters based on environmental data: based on the currently monitored temperature ,salinity and turbidity , combined with the above formula to calculate the corresponding light attenuation coefficient , using a pre-trained machine learning model to predict the optimal light source wavelength and power , in order to maximize the signal-to-noise ratio SNR.

[0114] Camera setting adjustment: Preliminarily set the camera exposure time based on the light source characteristics and expected reflected light intensity and gain This step may require several trials to find the optimal combination. In actual operation, these parameters can be further optimized through the automatic exposure and automatic gain control (AE / AGC) functions.

[0115] Verification and fine-tuning: Perform a test run to check the quality of the generated image and the 3D reconstruction effect. If the image is found to be too dark or too bright, adjust the light source power appropriately. and the camera exposure time , Gain , until a satisfactory result is obtained.

[0116] S200, dynamic light source adjustment and multi-band emission: Based on real-time feedback of water conditions, the light source characteristics are dynamically adjusted to optimize the imaging effect, and different wavelengths of light (such as visible light and near-infrared) are used to alternately emit structured light patterns for subsequent fusion processing;

[0117] For example, in a high turbidity environment, the frequency of the light source can be lowered to reduce the effect of scattering; while in a low turbidity environment, the frequency can be appropriately increased to increase the resolution;

[0118] Light in each frequency band has different penetrating power and scattering characteristics, which helps to obtain richer information.

[0119] In one embodiment of the present invention, the following steps are also included:

[0120] S210, real-time water condition feedback:

[0121] Data collection: Get the latest temperature from the sensor network deployed by S100 ,salinity and turbidity data.

[0122] State assessment: Based on these real-time data, the impact of the current water environment on light propagation, especially the scattering and absorption characteristics, is evaluated.

[0123] S220, adaptive light source adjustment:

[0124] Dynamic wavelength selection: select the most suitable light source wavelength according to water conditions In high turbidity environments, longer wavelengths (such as near infrared) are selected to reduce the effects of scattering, while in low turbidity environments, shorter wavelengths (such as visible light) can be selected to obtain higher resolution.

[0125] Power adjustment: adjust the output power of the light source according to the calculated optimal signal-to-noise ratio SNR , ensuring sufficient lighting intensity without causing excessive scattering or reflections.

[0126] The light attenuation coefficient : Considering the absorption and scattering effects, it is defined as follows:

[0127] ;

[0128] in is the total attenuation coefficient, is the temperature ,salinity and turbidity Function is the absorption coefficient, which is mainly determined by the temperature and salinity Decide, is the scattering coefficient, affected by turbidity The impact is greater;

[0129] The signal-to-noise ratio (SNR) is used to evaluate the imaging quality and is calculated as follows:

[0130] ;

[0131] in is the signal-to-noise ratio, is the wavelength and power The function of Indicates a given wavelength and power The signal strength under is the standard deviation of the background noise;

[0132] The best wavelength selection: find the optimal wavelength by optimizing the objective function , the function can be expressed as:

[0133] ;

[0134] in is the optimal wavelength, is the signal-to-noise ratio, is the wavelength and power Function of

[0135] Power adjustment: adjust the power according to the selected wavelength In order to maintain the best signal-to-noise ratio, it can be estimated by the following empirical formula:

[0136] ;

[0137] in is the light source power, is a scaling factor, is the minimum signal strength required to ensure adequate image brightness;

[0138] S230, multi-band emission design: determine multiple working frequency bands (such as visible light, near infrared, etc.), each with different penetration capabilities and scattering characteristics, so as to cover a wider range of information, and design the emission sequence of the structured light pattern so that light of different frequency bands can be emitted in sequence in a short time to avoid mutual interference and provide sufficient information for subsequent data fusion.

[0139] S300, Structured Light Pattern Generation and Optimization: predicts the optimal structured light pattern through a machine learning model that takes into account factors such as scattering effects and absorption characteristics under current water conditions;

[0140] In one embodiment of the present invention, the following contents are also included:

[0141] S310, Data Preparation and Feature Extraction

[0142] Environmental data input: Get the latest water condition data, including temperature, from S100 and S200 ,salinity , turbidity And other key parameters.

[0143] Feature engineering: Based on these data, calculate indicators that describe the current optical properties of the water, such as the absorption coefficient and scattering coefficient .

[0144] At the same time, other factors that may affect the imaging effect, such as the concentration distribution of suspended particles, are extracted.

[0145] S320, intelligent pattern generation

[0146] Model selection: Use a pre-trained machine learning or deep learning model (such as convolutional neural network (CNN) or reinforcement learning (RL)) that has learned the best design principles for structured light patterns under different water conditions.

[0147] The model architecture is as follows:

[0148] Input layer: receives the environment feature vector , which contains the temperature ,salinity , turbidity Water condition parameters.

[0149] Convolutional layer: Use multiple layers of convolution kernels to perform local perception on the input features and extract spatial feature representations.

[0150] Pooling layer: Reduces the spatial dimension of the feature map through downsampling operations while retaining key information.

[0151] Fully connected layer: maps the features extracted by the convolutional layer and the pooling layer to the final output - the structured light pattern encoding vector .

[0152] Output layer: Generate structured light pattern encoding vector , used to control the emission characteristics of the light source.

[0153] The calculation formula of this model is as follows:

[0154] Forward propagation: Assume that the weight of the lth layer of the model is , the bias is , the activation function is , then the output of the l+1th layer Can be expressed as:

[0155] ;

[0156] in is the output of the l+1th layer, is the weight matrix of the lth layer, is the bias vector of the lth layer, is the activation function, is the output of layer l, is the input feature vector;

[0157] in represents the weight matrix, Represents the bias vector, activation function Including ReLU and Sigmoid functions.

[0158] Loss function: Use mean square error as the loss function to measure the predicted value and the true value The difference between:

[0159] ;

[0160] in is the loss function value, is the sample size, and are the true value and predicted value of the i-th sample respectively;

[0161] Backpropagation and optimization: The gradient is calculated through the backpropagation algorithm, and the model parameters are updated using an optimizer (such as Adam, SGD) to minimize the loss function.

[0162] S330, optimization algorithm application

[0163] Search space initialization: determine the design space of structured light patterns, i.e., all possible encoding vectors can be achieved by random sampling or grid partitioning.

[0164] Iterative optimization: Use evolutionary algorithms such as genetic algorithm (GA) and particle swarm optimization (PSO), or traditional optimization methods such as gradient descent to find the optimal solution for the objective function in the search space. The optimal solution to maximize In each iteration, the wavelength of the light source is adjusted according to the current water conditions. and power , to ensure that the optimization process is always targeted at actual application scenarios.

[0165] Objective function definition: In order to find the optimal structured light pattern, define an optimization objective function , which comprehensively considers the signal-to-noise ratio (SNR) and contrast (Contrast).

[0166] The specific formula is:

[0167] ;

[0168] in is the optimization objective function, The encoding vector representing the structured light pattern, and is the weight factor of the two indicators, reflecting the degree of attention paid to signal-to-noise ratio and contrast, respectively. is the signal-to-noise ratio, is the contrast;

[0169] The structured light pattern encoding vector It is composed of a series of binary bits, each of which represents an element in the structured light pattern (such as stripe width, spacing, etc.). The specific encoding method depends on the optimization algorithm used and the hardware limitations.

[0170] Signal-to-noise ratio (SNR): used to evaluate imaging quality, the calculation method is:

[0171] ;

[0172] in is the signal-to-noise ratio, represents the average value of the signal strength under a given structured light pattern, is the standard deviation of the background noise;

[0173] Contrast: A measure of the degree of brightness difference between adjacent areas in an image, which can be expressed as:

[0174] ;

[0175] in is the contrast, and are the grayscale values ​​of the brightest and darkest pixels in the image, respectively;

[0176] S340, pattern verification and fine-tuning:

[0177] Simulation testing: Use simulation tools to virtually test the generated structured light patterns to evaluate their performance in specific water conditions.

[0178] Field verification: If conditions permit, the initially selected structured light pattern can be deployed in a real environment, and image data can be collected through experiments to further verify its effectiveness and stability.

[0179] Feedback Adjustment: Based on the results of simulation and field testing, make necessary fine-tuning of the structured light pattern until the expected performance requirements are met.

[0180] S400, image acquisition and preprocessing: uses a camera array equipped with a fast shutter mechanism to synchronously capture structured light patterns emitted and reflected by multi-band light sources, applies deep learning algorithms to remove scattered noise, and improves image quality through super-resolution reconstruction technology;

[0181] In one embodiment of the present invention, the following specific steps are also included:

[0182] S410, high-speed image capture:

[0183] Camera array preparation: Ensure that all cameras are properly initialized and exposure times are set according to the parameters determined in S100 and gain , check the synchronization mechanism between cameras to ensure that they can capture the structured light pattern at the same time.

[0184] Light source emission control: coordinate the working timing of the light source and the camera to ensure that the camera can immediately capture the reflected light signal after each light source emission. For multi-band emission, a reasonable alternating sequence needs to be designed to avoid interference between frequency bands.

[0185] S420, image data acquisition:

[0186] Synchronous triggering: Through external trigger or internal timer, precise synchronization of light source emission and camera exposure is achieved to obtain clear images without motion blur.

[0187] Continuous acquisition: Multiple frames of images are acquired continuously within a certain period of time for subsequent statistical analysis and noise suppression. This step is particularly important because in underwater environments, a single shot may introduce random errors due to factors such as water flow.

[0188] S430, denoising and enhancement:

[0189] Deep learning denoising: Apply a pre-trained convolutional neural network (CNN) or other deep learning models to remove scattered noise.

[0190] The specific formula is:

[0191] ;

[0192] in It is a high-quality image after processing. is the original collected image, represents a deep neural network;

[0193] Super-resolution reconstruction: Use super-resolution reconstruction technology to improve image resolution and restore more details.

[0194] Common methods include those based on interpolation, sparse coding, or generative adversarial networks (GANs).

[0195] S440, image quality assessment:

[0196] Signal-to-noise ratio calculation: Evaluate the denoising effect and calculate the signal-to-noise ratio (SNR) of the image before and after processing to ensure that the noise is effectively suppressed. The formula is as follows:

[0197] ;

[0198] in is the signal-to-noise ratio, represents the mean square of the signal strength, is the square of the standard deviation of the noise;

[0199] Contrast measurement: quantify image contrast to ensure that the imaging system can maintain good visual effects in different water conditions;

[0200] Contrast ratio can be calculated using the following formula:

[0201] ;

[0202] in is the contrast, and are the grayscale values ​​of the brightest and darkest pixels in the image, respectively;

[0203] In one embodiment of the present invention, the image acquisition and preprocessing process is as follows:

[0204] Preparation phase: Ensure that the camera array and light source equipment have been initialized according to the settings in the previous steps, verify the synchronization mechanism between the camera and the light source, and ensure that the two can work together;

[0205] Image capture: trigger the light source to emit structured light patterns, start the camera array for synchronous exposure, capture the reflected light signals, and continuously collect multiple frames of images within a certain period of time for subsequent statistical analysis and noise suppression;

[0206] Denoising: The original image Input into the pre-trained deep learning model to get the denoised image ,Using super-resolution reconstruction technology to further improve image resolution and restore more details;

[0207] Quality assessment: Calculate the signal-to-noise ratio (SNR) of the image before and after denoising to ensure that the noise is effectively suppressed, measure the contrast of the image, and evaluate whether its visual effect meets the expected standards;

[0208] Feedback adjustment: If the image quality does not meet the requirements, the camera exposure time can be adjusted appropriately , Gain , and the wavelength of the light source and power ;

[0209] Through continuous iterative optimization, we ensure that the final image has a high signal-to-noise ratio and good contrast, providing a solid foundation for subsequent data fusion and three-dimensional reconstruction.

[0210] S500, data fusion and phase unwrapping: align and fuse image data of different frequency bands to generate a comprehensive structured light pattern, extract complete phase information from the fused image, and then calculate the three-dimensional coordinates of the object surface;

[0211] In one embodiment of the present invention, the following steps are also included:

[0212] S510, multi-band data registration:

[0213] Image preprocessing: Preprocess the original image of each frequency band, including correcting geometric distortion, color balance, etc., to ensure good consistency between images in different frequency bands.

[0214] Feature point extraction: Use corner detection (such as Harris corner detection) or edge detection algorithms to find stable feature points in each image. These feature points will be used in the subsequent image registration process.

[0215] Matching and registration: Use feature point matching methods (such as SIFT, SURF, or ORB) to find the correspondence between images in different frequency bands, and align all images to the same coordinate system through affine transformation or perspective transformation.

[0216] S520, Data Fusion:

[0217] Weight allocation: Assign a weight to each frequency band based on the quality of the image in each frequency band (such as signal-to-noise ratio, contrast, etc.) This helps to fully utilize high-quality data during the fusion process while reducing the impact of low-quality data.

[0218] Weighted average fusion: Use weighted average to fuse multi-band images into a comprehensive structured light pattern. The specific formula is as follows:

[0219] ;

[0220] in is the fused image, The image of the i-th frequency band is a sequence of images captured after being emitted and reflected by a multi-band light source. Represents the weight factor, which is a non-negative real number. It indicates the relative importance of the i-th frequency band image in the fusion process and needs to be dynamically adjusted according to the quality of each frequency band image. is the total number of frequency bands;

[0221] S530, refractive index correction:

[0222] Refractive index calculation: based on temperature and salinity The measured value of the water body is calculated using the empirical formula The formula is:

[0223] ;

[0224] in is the refractive index, is the temperature and salinity The function of is the temperature, is the salinity, , , They are compensation empirical constant, temperature empirical constant and salinity empirical constant;

[0225] Optical path deviation correction: Consider the optical path deviation caused by the refraction effect and adjust the camera intrinsic parameter matrix , to ensure that the reconstructed three-dimensional coordinates are more accurate.

[0226] S540, Phase Unwrapping:

[0227] Phase information extraction: Extract complete phase information from the fused image using Fourier transform or fringe analysis.

[0228] 3D coordinate calculation: Assuming the camera intrinsic matrix is ​​known , then the three-dimensional coordinates [X, Y, Z] can be calculated by the following formula:

[0229] ;

[0230] in is the pixel position, It is the distance value of the corresponding point, which refers to the actual distance from the camera to a certain point on the surface of the object.

[0231] In one embodiment of the present invention, the data fusion and phase unwrapping process is as follows:

[0232] Image registration: Preprocess the original images of each frequency band to ensure the consistency between images, extract feature points and match them, and align all images to the same coordinate system through affine or perspective transformation.

[0233] Data fusion: Assign appropriate weights to each frequency band based on the quality of the images in each frequency band , using the weighted average method to fuse multi-band images into a comprehensive structured light pattern

[0234] Refractive index correction: Based on real-time monitored temperature and salinity Calculate the refractive index of water , adjust the camera intrinsic parameter matrix , in order to correct the optical path deviation caused by the refraction effect.

[0235] Phase unwrapping: Extract the complete phase information from the fused image using the camera intrinsic parameter matrix and phase information to calculate the three-dimensional coordinates [X, Y, Z] of the object surface.

[0236] Continuous optimization: In actual operation, the imaging effect is continuously monitored, and the fusion and phase unwrapping processes are continuously optimized based on the latest data to form a closed-loop control system to ensure the adaptability and robustness of the system.

[0237] S600, anti-interference 3D reconstruction: Taking into account possible non-ideal factors (such as camera lens distortion), the geometric correction algorithm is applied to further improve the reconstruction accuracy, and the gaps in the point cloud are filled by interpolation or filtering algorithms to ensure the integrity and smoothness of the model;

[0238] In one embodiment of the present invention, the following steps are also included:

[0239] S610, Geometric Error Correction:

[0240] Lens distortion correction: using a pre-calibrated camera intrinsic matrix and distortion coefficient , perform geometric distortion correction on the captured image. This step can effectively eliminate the barrel or pincushion distortion caused by the camera lens.

[0241] Relative pose estimation: Estimate the pose of the camera relative to the object surface through multi-view geometry methods (such as the PnP algorithm) to ensure the accuracy of subsequent 3D coordinate calculations.

[0242] In order to achieve anti-interference 3D reconstruction, we need to rely on the following key formulas:

[0243] Geometric distortion correction: Assume that the original pixel position is , the corrected pixel position is , it can be corrected by the following formula:

[0244] ;

[0245] in is the distortion model function, is the camera intrinsic parameter matrix.

[0246] Relative pose estimation (PnP problem): Given a set of 3D points and the corresponding 2D image points , solve the camera rotation matrix by minimizing the reprojection error and translation vector :

[0247] ;

[0248] in Represents the function of projecting a 3D point onto the image plane, the rotation matrix and translation vector Transformation parameters representing the pose of the camera relative to the surface of the object.

[0249] S620, point cloud density optimization:

[0250] Gap filling: Use interpolation algorithms (such as nearest neighbor interpolation, bilinear interpolation, or spline interpolation) to fill sparse areas in the point cloud to improve the integrity and continuity of the point cloud.

[0251] Noise filtering: Apply filtering algorithms (such as bilateral filtering, mean filtering, or Gaussian filtering) to remove isolated noise points in the point cloud and smooth surface details while maintaining edge features.

[0252] Interpolation formula (taking nearest neighbor interpolation as an example): For the target point , find its nearest known point , and assign the attribute value of the point to the target point:

[0253] ;

[0254] ;

[0255] Filtering formula (taking bilateral filtering as an example): For each point , according to its relationship with the neighboring points The weighted average of spatial distance and brightness difference:

[0256] ;

[0257] in and are spatial weight and range weight, Denotes the difference-weighted mean.

[0258] Calculation of three-dimensional coordinates: Based on the camera intrinsic parameter matrix The distance value obtained by phase unwrapping , calculate the three-dimensional coordinates [X, Y, Z]:

[0259] ;

[0260] Distortion coefficient : A vector describing the lens distortion characteristics, including radial distortion coefficients and tangential distortion coefficients, and the camera intrinsic parameter matrix : A 3x3 matrix containing parameters such as focal length and principal point offset, used to convert two-dimensional image coordinates into three-dimensional space coordinates.

[0261] S630, surface detail preservation:

[0262] Feature enhancement: Local feature enhancement technology (such as Harris corner detection, FAST feature detection, etc.) is used to highlight the key structure and texture information of the object surface to ensure that the reconstructed model can faithfully reflect the shape of the original object.

[0263] Multi-scale processing: Combine filters of different scales or transform domain methods (such as wavelet transform) to process point cloud data at different resolution levels to take into account both global shape and local details.

[0264] In one embodiment of the present invention, the anti-interference 3D reconstruction process is as follows:

[0265] Geometric Error Correction: Using a Pre-calibrated Camera Intrinsic Matrix and distortion coefficient , perform geometric distortion correction on the captured images, estimate the posture of the camera relative to the object surface through multi-view geometry methods, and ensure the accuracy of subsequent three-dimensional coordinate calculations;

[0266] Point cloud density optimization: use interpolation algorithms to fill in sparse areas in the point cloud to improve the integrity and continuity of the point cloud, and apply filtering algorithms to remove isolated noise points in the point cloud, smooth surface details, and maintain edge features;

[0267] Surface detail preservation: Use local feature enhancement technology to highlight the key structure and texture information of the object surface, ensure that the reconstructed model can faithfully reflect the shape of the original object, combine filters or transform domain methods of different scales, process point cloud data at different resolution levels, and take into account both global shape and local details;

[0268] Continuous optimization: In actual operation, the imaging effect is continuously monitored, and the reconstruction process is continuously optimized based on the latest data to form a closed-loop control system to ensure the adaptability and robustness of the system.

[0269] S700, performance evaluation and feedback adjustment: Comprehensively evaluate the reconstructed 3D point cloud, including point cloud density, geometric error, surface detail retention, etc., automatically adjust system parameters according to the evaluation results, form a closed-loop control mechanism, and continuously optimize the imaging effect.

[0270] In one embodiment of the present invention, the following steps are also included:

[0271] S710, Quality Assessment:

[0272] Point cloud density analysis: Count the number of points in each area of ​​the generated 3D point cloud to ensure that the model has enough details to measure the density by counting the number of points per unit area or volume.

[0273] Geometric error measurement: Compare the difference between the reconstructed model and the real object and calculate the geometric error. Commonly used methods include root mean square error (RMSE), Hausdorff distance, etc.

[0274] Surface detail preservation: Evaluate whether the reconstructed model retains the key structure and texture information of the original object. The edge clarity of the model can be checked using edge detection algorithms, or visually judged through visual comparison.

[0275] In one embodiment of the present invention, the point cloud density :

[0276] ;

[0277] in is the number of points in a region, is the area or volume of the region, Represents the point cloud density.

[0278] Root Mean Square Error (RMSE):

[0279] ;

[0280] here Represents the coordinates of a point on a real object, represents the coordinates of the corresponding points on the reconstructed model, is the number of points, represents the root mean square error.

[0281] Hausdorff Distance:

[0282] ;

[0283] in and Represent the point sets of the real object and the reconstructed model respectively, represents the Hausdorff distance. represents the supremum (smallest upper bound), that is, the maximum value in the set, represents the infimum (maximum lower bound), that is, the minimum value in the set, Indicate point Belongs to point set , Indicate point Belongs to point set , Indicate point and Point The Euclidean distance between Indicates taking the larger of the two values;

[0284] Edge clarity evaluation: Extract the model edge through edge detection algorithms (such as Canny operator) and calculate the standard deviation of edge intensity distribution , a larger standard deviation indicates a clearer edge.

[0285] Edge strength standard deviation : A statistic that reflects the edge clarity of the model, in grayscale values.

[0286] S720, system feedback:

[0287] Parameter adjustment: Based on the results of the quality assessment, the system parameters are automatically adjusted to optimize the imaging effect. This may involve the wavelength of the light source ,power , camera exposure time and gain And other key parameters.

[0288] Light source wavelength : The unit is nanometer (nm), located in the visible spectrum (such as 450nm to 650nm) or extending to the near-infrared region (such as 850nm).

[0289] Light source power : The unit is watt (W), which represents the total energy output of the emitted light.

[0290] Camera exposure time : The unit is seconds (s), which controls the length of time the camera's photosensitive element receives light.

[0291] Camera Gain : The unit is decibel (dB) or multiple, which adjusts the gain level of the camera's internal amplifier circuit.

[0292] Closed-loop control: Form a continuously improving closed-loop control system that enables the system to dynamically adjust its behavior based on the latest environmental data and imaging results, gradually improving performance.

[0293] S730, continuous optimization: Using machine learning or deep learning algorithms, the system accumulates experience from each imaging process, continuously optimizes its own prediction and processing capabilities, collects users' subjective evaluations and suggestions, and further improves the system's functions and user experience.

[0294] In one embodiment of the present invention, the performance evaluation and feedback adjustment process:

[0295] Quality Assessment:

[0296] Point cloud density analysis: Calculate the density of the generated 3D point cloud in different areas to ensure that the model has sufficient details.

[0297] Geometric error measurement: The geometric error between the reconstructed model and the real object is evaluated using methods such as RMSE and Hausdorff distance.

[0298] Surface detail preservation: The edge detection algorithm is used to evaluate the clarity of the model edges to ensure that surface details are well preserved.

[0299] System feedback:

[0300] Parameter adjustment: Automatically adjust the wavelength of the light source according to the quality assessment results ,power , camera exposure time and gain etc. to optimize the imaging effect.

[0301] Closed-loop control: Establish a closed-loop control system that enables the system to dynamically adjust its behavior based on the latest environmental data and imaging results, gradually improving performance.

[0302] Continuous Optimization:

[0303] Adaptive learning: Using machine learning or deep learning algorithms, the system accumulates experience from each imaging process and continuously optimizes its prediction and processing capabilities.

[0304] User feedback: Collect users' subjective evaluations and suggestions to further improve the system's functions and user experience.

[0305] like Figure 2 As shown, the present invention also proposes a high-precision underwater structured light three-dimensional point cloud imaging system, which performs one or more steps of the above-mentioned high-precision underwater structured light three-dimensional point cloud imaging method, including:

[0306] Environmental monitoring and parameter initialization module 101: deploy multiple sensor nodes to monitor key parameters such as water temperature, salinity, and turbidity in real time, select the appropriate light source wavelength and power based on the collected data to ensure the best signal-to-noise ratio (SNR) under current conditions, and initialize the camera exposure time and gain settings;

[0307] Dynamic light source adjustment and multi-band emission module 102: based on real-time feedback of water conditions, dynamically adjust light source characteristics to optimize imaging effects, and use lights of different wavelengths to alternately emit structured light patterns for subsequent fusion processing;

[0308] Structured light pattern generation and optimization module 103: predicts the optimal structured light pattern through a pre-trained machine learning model (such as CNN or RL), taking into account factors such as scattering effects and absorption characteristics, and maximizing the signal-to-noise ratio and contrast;

[0309] Image acquisition and preprocessing module 104: using a camera array equipped with a fast shutter mechanism to synchronously capture structured light patterns emitted and reflected by multi-band light sources, applying a deep learning algorithm to remove scattered noise, and improving image quality through super-resolution reconstruction technology;

[0310] Data fusion and phase unwrapping module 105: aligns and fuses image data of different frequency bands to generate a comprehensive structured light pattern, extracts complete phase information from the fused image, and then calculates the three-dimensional coordinates of the object surface;

[0311] Anti-interference 3D reconstruction module 106: taking into account possible non-ideal factors (such as camera lens distortion), applying a geometric correction algorithm to further improve reconstruction accuracy, and filling gaps in the point cloud through an interpolation or filtering algorithm to ensure the integrity and smoothness of the model;

[0312] Performance evaluation and feedback adjustment module 107: comprehensively evaluates the reconstructed three-dimensional point cloud, including point cloud density, geometric error, surface detail retention, etc., automatically adjusts system parameters according to the evaluation results, forms a closed-loop control mechanism, and continuously optimizes the imaging effect.

[0313] At least one embodiment of the present disclosure provides a storage medium storing non-transitory computer-readable instructions for executing one or more steps of the aforementioned method for sustainable communication optimization processing in a restricted connection environment.

[0314] The computer program can be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems. Any reference signs in the claims should not be construed as limiting the scope.

[0315] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A high-precision underwater structured light three-dimensional point cloud imaging method, characterized in that: The following steps are involved: S100: Deploy multiple sensor nodes to monitor key parameters of water temperature, salinity and turbidity in real time, select the appropriate light source wavelength and power based on the collected data to ensure the best signal-to-noise ratio under current conditions, and initialize the camera exposure time and gain settings; S200: Based on real-time feedback of water conditions, it dynamically adjusts the light source characteristics to optimize the imaging effect, and uses light of different wavelengths to alternately emit structured light patterns; S300: Predict the optimal structured light pattern through a machine learning model that takes into account the scattering effect and absorption characteristics of the current water conditions; S400: uses a camera array equipped with a fast shutter mechanism to synchronously capture structured light patterns emitted and reflected by multi-band light sources, and applies deep learning algorithms to remove scattered noise; S500: aligns and fuses image data of different frequency bands to generate a comprehensive structured light pattern, extracts complete phase information from the fused image, and then calculates the three-dimensional coordinates of the object surface; S600: Taking into account possible non-ideal factors, a geometric correction algorithm is applied to further improve the reconstruction accuracy, and gaps in the point cloud are filled through interpolation or filtering algorithms; S700: Comprehensively evaluate the reconstructed 3D point cloud, including point cloud density, geometric error, and surface detail retention, and automatically adjust system parameters based on the evaluation results to form a closed-loop control mechanism.

2. A high-precision underwater structured light three-dimensional point cloud imaging method according to claim 1, characterized in that: The S100 also includes the following: S110, deploying the sensor network: select multiple representative monitoring points based on the geographical features of the target underwater area, cover the entire imaging range, and distribute them evenly. Each node is equipped with a thermometer, conductivity meter, and turbidity meter; S120, real-time environmental monitoring: start the sensor and start collecting temperature at set time intervals ,salinity and turbidity Set thresholds for the data to trigger alarms when the monitored data exceeds the normal range, alerting you to potential problems. S130, data processing and analysis: Apply filtering methods to the original collected data to remove high-frequency noise, obtain a more stable trend line, and calculate the average value and standard deviation statistics over a period of time.

3. The high-precision underwater structured light three-dimensional point cloud imaging method according to claim 2, characterized in that: The S200 also includes the following: S210, real-time water condition feedback: Get the latest temperature from the sensor network deployed by S100 ,salinity and turbidity Data, based on real-time data, evaluates the impact of the current water environment on light propagation; S220, adaptive light source adjustment: select the most suitable light source wavelength according to water conditions , in high turbidity environments, longer wavelengths are chosen to reduce the effects of scattering, while in low turbidity environments, shorter wavelengths are chosen to achieve higher resolution; Adjust the output power of the light source according to the calculated optimal signal-to-noise ratio SNR , providing sufficient lighting intensity without causing excessive scattering or reflection; S230, multi-band emission design: determine multiple working frequency bands, each with different penetration capabilities and scattering characteristics, and design the emission sequence of the structured light pattern so that light of different frequency bands can be emitted sequentially in a short period of time.

4. The high-precision underwater structured light three-dimensional point cloud imaging method according to claim 3, characterized in that: The S300 also includes the following: S310, Data Preparation and Feature Extraction: Get the latest water condition data from S100 and S200, including temperature ,salinity , turbidity Key parameters, based on the above data, calculate indicators describing the current optical properties of the water body; At the same time, other factors that may affect the imaging effect are extracted; S320, Intelligent Pattern Generation: uses a pre-trained model that has learned the best design principles for structured light patterns under different water conditions; S330, Optimization algorithm application: Determine the design space of structured light patterns, all possible encoding vectors The algorithm is used to find the optimal solution that maximizes the objective function in the search space through random sampling or grid division. During each iteration, the wavelength and power of the light source are adjusted according to the current water conditions. S340, Pattern Verification and Fine-tuning: Use simulation tools to virtually test the generated structured light pattern to evaluate its performance under specific water conditions. If conditions permit, deploy the initially selected structured light pattern in a real environment and collect image data through experiments for further verification. Based on the results of simulation and field tests, make necessary fine-tuning to the structured light pattern until it meets the expected performance requirements.

5. The high-precision underwater structured light three-dimensional point cloud imaging method according to claim 4, characterized in that: The architecture of the pre-trained model in S320 is as follows: Input layer: receives the environment feature vector , which contains the temperature ,salinity , turbidity Water condition parameters; Convolutional layer: Use multiple convolution kernels to locally perceive the input features and extract spatial feature representations; Pooling layer: reduces the spatial dimension of the feature map through downsampling operations while retaining key information; Fully connected layer: maps the features extracted by the convolutional layer and the pooling layer to the final output, the structured light pattern encoding vector ; Output layer: Generate structured light pattern encoding vector , used to control the emission characteristics of the light source; The calculation formula of the pre-trained model is as follows: Forward propagation: Assume that the weight of the lth layer of the model is , the bias is , the activation function is , then the output of the l+1th layer It is expressed as: ； in is the output of the l+1th layer, is the weight matrix of the lth layer, is the bias vector of the lth layer, is the activation function, is the output of layer l, is the input feature vector; in represents the weight matrix, represents the bias vector, is the Sigmoid function; Loss function: Use mean square error as the loss function to measure the predicted value and the true value The difference between: ； in is the loss function value, is the sample size, and are the true value and predicted value of the i-th sample respectively; Backpropagation and optimization: The gradient is calculated through the backpropagation algorithm and the optimizer is used to update the model parameters to minimize the loss function.

6. The high-precision underwater structured light three-dimensional point cloud imaging method according to claim 5, characterized in that: The S400 also includes the following: S410, high-speed image capture: ensuring that all cameras are properly initialized and that exposure time and gain are set according to the parameters determined in S100, and checking the synchronization mechanism between the cameras to be able to capture the structured light pattern simultaneously; Coordinate the working sequence of the light source and the camera to ensure that the camera can immediately capture the reflected light signal after each light source emission. For multi-band emission, use a reasonable alternating sequence; S420, image data acquisition: achieving precise synchronization of light source emission and camera exposure through an external trigger or an internal timer; Continuously collect multiple frames of images within a certain period of time for subsequent statistical analysis and noise suppression; S430, Denoising and Enhancement: Apply pre-trained models to remove scattered noise; The specific formula of the model is: ； in It is a high-quality image after processing. is the original collected image, represents a deep neural network; Super-resolution reconstruction: Use super-resolution reconstruction technology to improve image resolution and restore more details; S440, image quality assessment: evaluate the denoising effect, calculate the signal-to-noise ratio of the image before and after processing, and ensure that the noise is effectively suppressed; Contrast measurement: Quantify image contrast to ensure that the imaging system can maintain good visual effects in different water conditions.

7. The high-precision underwater structured light three-dimensional point cloud imaging method according to claim 6, characterized in that: The S500 also includes the following: S510, multi-band data registration: pre-process the original image of each frequency band, including correcting geometric distortion and color balance, using corner detection or edge detection algorithms to find stable feature points in each image, using feature point matching methods to find the correspondence between images of different frequency bands, and aligning all images to the same coordinate system through affine transformation or perspective transformation; S520, data fusion: assign a weight to each frequency band based on the quality of the image in each frequency band , the multi-band images are fused into a comprehensive structured light pattern by weighted averaging; The specific formula is as follows: ； in is the fused image, The image of the i-th frequency band is a sequence of images captured after being emitted and reflected by a multi-band light source. Represents the weight factor, which is a non-negative real number. It indicates the relative importance of the i-th frequency band image in the fusion process and needs to be dynamically adjusted according to the quality of each frequency band image. is the total number of frequency bands; S530, Refractive Index Correction: Temperature-dependent and salinity The measured value of the water body is calculated using the empirical formula The formula is: ； in is the refractive index, is the temperature, is the salinity, , , They are compensation empirical constant, temperature empirical constant and salinity empirical constant; Considering the optical path deviation caused by the refraction effect, adjust the camera intrinsic parameter matrix K; S540, Phase Unwrapping: Extract the complete phase information from the fused image, assuming the camera intrinsic matrix is ​​known , then the three-dimensional coordinates [X, Y, Z] are calculated by the following formula: ； in is the pixel position, It is the distance value of the corresponding point, which refers to the actual distance from the camera to a certain point on the surface of the object.

8. The high-precision underwater structured light three-dimensional point cloud imaging method according to claim 7, characterized in that: The S600 also includes the following: S610, geometric error correction: using a pre-calibrated camera intrinsic parameter matrix and distortion coefficient, geometric distortion correction is performed on the captured image, and the posture of the camera relative to the object surface is estimated by a multi-view geometry method; S620, point cloud density optimization: use interpolation algorithms to fill sparse areas in the point cloud, apply filtering algorithms to remove isolated noise points in the point cloud, smooth surface details, and maintain edge features. For a target point, find its nearest known point and assign the attribute value of the point to the target point. For each point, take a weighted average based on the spatial distance and brightness difference between it and its neighboring points, and calculate the 3D coordinates based on the distance value obtained by the camera intrinsic parameter matrix and phase unwrapping. S630, surface detail preservation: uses local feature enhancement technology to highlight the key structure and texture information of the object surface, combines filters of different scales or transform domain methods to process point cloud data at different resolution levels.

9. A high-precision underwater structured light three-dimensional point cloud imaging system, characterized in that: The method for executing the steps in the high-precision underwater structured light three-dimensional point cloud imaging method as claimed in any one of claims 1 to 8 comprises: Environmental monitoring and parameter initialization module (101): deploy multiple sensor nodes to monitor the key parameters of water temperature, salinity and turbidity in real time, select the appropriate light source wavelength and power based on the collected data to ensure the best signal-to-noise ratio under the current conditions, and initialize the camera exposure time and gain settings; Dynamic light source adjustment and multi-band emission module (102): based on real-time feedback of water conditions, dynamically adjust light source characteristics to optimize imaging effects, and use lights of different wavelengths to alternately emit structured light patterns; Structured light pattern generation and optimization module (103): predicts the optimal structured light pattern through a pre-trained machine learning model, taking into account scattering effects and absorption characteristics, and maximizing signal-to-noise ratio and contrast; Image acquisition and preprocessing module (104): using a camera array equipped with a fast shutter mechanism to synchronously capture the structured light pattern emitted and reflected by the multi-band light source, applying a deep learning algorithm to remove scattered noise, and improving image quality through super-resolution reconstruction technology; Data fusion and phase unwrapping module (105): aligns and fuses image data of different frequency bands to generate a comprehensive structured light pattern, extracts complete phase information from the fused image, and then calculates the three-dimensional coordinates of the object surface; Anti-interference 3D reconstruction module (106): considering possible non-ideal factors, applying a geometric correction algorithm to further improve the reconstruction accuracy, and filling the gaps in the point cloud through an interpolation or filtering algorithm; Performance evaluation and feedback adjustment module (107): performs a comprehensive evaluation on the reconstructed three-dimensional point cloud, including point cloud density, geometric error, and surface detail retention, and automatically adjusts system parameters based on the evaluation results to form a closed-loop control mechanism.

10. A storage medium, characterized in that: Non-temporary computer-readable instructions are stored for executing the steps in a high-precision underwater structured light three-dimensional point cloud imaging method as described in any one of claims 1-8.

Images