Error compensation method and system for underwater stripe structured light three-dimensional reconstruction technology

Abstract

The invention relates to the technical field of image data processing, in particular to an error compensation method and system for an underwater stripe structured light three-dimensional reconstruction technology. Determining an actual working area sensitive factor corresponding to the actual projection path of the underwater stripe structured light based on the seawater environment parameters; determining a predicted seawater refractive index corresponding to the monitoring area based on the seawater environment parameters, the actual working area sensitive factors and a trained preset error compensation model; and obtaining an initial region image corresponding to the monitoring region, and correcting the initial region image based on the predicted seawater refractive index to obtain a target region image. The accuracy of the seawater refractive index is improved, so that the image correction effect is improved, and the image distortion probability is reduced.

Description

Technical Field

[0001] This application relates to the field of image data processing technology, and in particular to an error compensation method and system for underwater stripe structured light three-dimensional reconstruction technology. Background Technology

[0002] As a vital component of Earth's life support system, the ocean not only harbors abundant mineral and biological resources but also serves as a crucial vehicle for regulating global climate and maintaining ecological balance. With ocean exploration expanding into the deep and open seas, structured light 3D reconstruction technology, with its advantages of being "intuitive, efficient, and non-destructive," has become a core tool for scenarios such as deep-sea exploration, seabed resource discovery, and underwater engineering monitoring.

[0003] However, the unique characteristics of the marine environment present significant challenges to structured light 3D reconstruction. Unlike terrestrial optical imaging, light propagating in seawater is directly affected by the properties of the seawater medium, leading to increased 3D reconstruction errors. Therefore, it is necessary to correct the acquired image based on the seawater refractive index to avoid severe image distortion. However, conventional methods typically use a fixed seawater refractive index for image correction. Since the seawater refractive index is not a fixed value, it dynamically changes with seawater environmental parameters. Therefore, even after image correction, there is still a probability of image distortion. Summary of the Invention

[0004] To improve the accuracy of seawater refractive index, thereby enhancing image correction and reducing image distortion probability, this application provides an error compensation method and system for underwater stripe structured light 3D reconstruction technology.

[0005] Firstly, this application provides an error compensation method for underwater stripe structured light three-dimensional reconstruction technology, employing the following technical solution: An error compensation method for underwater stripe structured light 3D reconstruction technology includes: Obtain the seawater environment parameters corresponding to the monitoring area, and determine the actual working area sensitivity factor corresponding to the actual projection path of the underwater striped structured light based on the seawater environment parameters. The predicted seawater refractive index of the monitoring area is determined based on the seawater environmental parameters, the sensitivity factors of the actual working area, and the trained preset error compensation model. An initial region image corresponding to the monitoring area is obtained, and the initial region image is corrected based on the predicted seawater refractive index to obtain the target region image.

[0006] By adopting the above technical solution, the sensitivity factors of the actual working area corresponding to the actual projection path of the underwater striped structured light are determined by analyzing the seawater environmental parameters. This facilitates the analysis of the influencing factors that the projection path may encounter in seawater. By using the seawater environmental parameters and the sensitivity factors of the actual working area as input, and then using a pre-trained error compensation model that conforms to the characteristics of the deep sea for calculation, the adaptability between the analysis process of predicting the seawater refractive index and the actual scene is improved, thereby improving the accuracy of determining the predicted seawater refractive index. Based on the highly accurate predicted seawater refractive index, the initial area image is corrected, which improves the effectiveness of the target area image and thus reduces the distortion probability of the target area image.

[0007] In one possible implementation, the training process of the preset error compensation model includes: Obtain a preset error compensation model and a sample environment parameter set, wherein the sample environment parameter set contains multiple sample environment parameters and the sample collection location and sample seawater refractive index corresponding to each sample environment parameter; Based on the sample collection location of each sample environmental parameter, the sensitivity factor of the sample working area corresponding to each sample environmental parameter is determined. The initial loss function is optimized based on the mean of the sensitivity factor of the sample working area sensitivity factor to obtain the target loss function. The target loss function is Loss=MSE(n_pred,n_true)+λα*MSE(n_pred,n_physical), where n_pred is the model predicted seawater refractive index, n_true is the sample seawater refractive index, n_physical is the empirical seawater refractive index calculated based on the preset empirical formula for seawater refractive index, λ is the preset balance weight, and α is the mean of the sensitivity factor. Based on each sample environment parameter and the corresponding sample working area sensitivity factor in the sample environment parameter set, determine the sample neuron discarding strategy corresponding to each sample environment parameter; Based on the sample environment parameter set, the target loss function, and the sample neuron discarding strategy corresponding to each sample environment parameter, the preset error compensation model is iteratively trained. Training stops when the output value of the preset error compensation model is within the preset error range, thus obtaining the trained preset error compensation model.

[0008] By adopting the above technical solution, the training process of the preset error compensation model is constrained based on the optimized target loss function. This facilitates the strengthening of model training constraints from three dimensions: data fitting accuracy, consistency with physical laws, and adaptability to deep-sea scenarios. It solves the problems of the preset error compensation model being prone to overfitting noise, deviating from physical essence, and inaccurate prediction of highly sensitive working areas in deep-sea scenarios. This improves the accuracy and robustness of the model's output in predicting seawater refractive index. In addition, by determining the corresponding sample neuron discarding strategy according to the actual situation of each sample's environmental parameters, the rationality of neuron discarding is improved, thereby enhancing the accuracy and robustness of the preset error compensation model.

[0009] In one possible implementation, the sensitivity factor of the sample working area corresponding to the sample environmental parameters is determined based on the sample acquisition location, including: Based on the sample environment parameters and the preset weight allocation mapping relationship, the sensitivity factor weights corresponding to the sample environment parameters are determined; Based on the sample acquisition location of the sample environment parameters, the sample projection path corresponding to the sample environment parameters is obtained. Based on the sample projection path, the working area temperature sensitivity factor and absolute depth sensitivity factor are identified. According to the sensitivity factor weight, the working area temperature sensitivity factor, the absolute depth sensitivity factor, and a preset sensitivity factor calculation formula, the sample working area sensitivity factor corresponding to the sample environment parameters is determined. The preset sensitivity factor calculation formula is as follows: β = sigmoid(k1*T + k2*D), where β is the sensitivity factor of the working area of ​​the sample, k1 and k2 are the sensitivity factor weights, T is the temperature sensitivity factor of the working area, and D is the absolute depth sensitivity factor.

[0010] By adopting the above technical solution, after determining the sensitivity factor weights by analyzing the sample environmental parameters, the sensitivity factor weights are used in the subsequent calculation process of the sensitivity factors of the sample working area, instead of using fixed values ​​for calculation. This facilitates the improvement of the accuracy of the sensitivity factors of the sample working area. In addition, by extracting the working area temperature sensitivity factor and absolute depth sensitivity factor from the sample projection path, and quantifying and normalizing the two with the sensitivity factor weights, an interpretable and standardized range can be obtained. This ensures that the sensitivity factors of the sample working area can directly serve the subsequent model optimization process, thereby facilitating the assurance of the model training results' adaptability to deep-sea features from the parameter source.

[0011] In one possible implementation, based on sample environment parameters and corresponding sample working area sensitivity factors, a sample neuron discarding strategy corresponding to the sample environment parameters is determined, including: The corresponding sample coupling degree is determined based on the sample environment parameters; Based on the sample working area sensitivity factor and the sample coupling degree, the sample neuron discarding strategy corresponding to the sample environment parameter is determined from the preset discarding strategy library, and the sample neuron discarding strategy is the discarding rate of each neural layer.

[0012] By adopting the above technical solution, the sample coupling degree is determined based on the sample environmental parameters, which facilitates the quantification of the interrelationship between different sample environmental parameters. This allows the model to better understand the complex relationships between environmental parameters, enabling more accurate learning and prediction based on the specific environmental parameter coupling when dealing with different sample environmental parameters. This improves the model's adaptability to various real deep-sea environments. In addition, determining the sample neuron discarding strategy based on the sample working area sensitivity factor and sample coupling degree allows the model to discard neurons more reasonably during training, rather than randomly. This helps prevent the model from over-relying on certain neurons or combinations of neurons and avoids co-adaptation among neurons, while enhancing the model's robustness.

[0013] In one possible implementation, determining the predicted seawater refractive index corresponding to the monitoring area based on the seawater environmental parameters, the sensitivity factor of the actual working area, and a trained preset error compensation model includes: Based on the actual working area sensitivity factors corresponding to the seawater environmental parameters, determine the actual neuron discarding strategy corresponding to the actual working area sensitivity factors; Based on the actual neuron dropout strategy, the Dropout layer in the trained preset error compensation model is activated to obtain the activated error compensation model, and the loop steps are executed until the preset conditions are met. The initial confidence level that meets the preset conditions is determined as the final prediction confidence level, and the initial seawater refractive index corresponding to the final prediction confidence level is determined as the final prediction seawater refractive index. The loop steps include: The seawater environmental parameters are input into the activation error compensation model, and multiple initial predicted seawater refractive indices are obtained after multiple forward propagations. The average of the multiple predicted seawater refractive indices is calculated to obtain the corresponding initial predicted seawater refractive index. The standard deviation of the multiple predicted seawater refractive indices is calculated to obtain the corresponding initial confidence levels. Determine whether the initial confidence level is greater than a preset confidence threshold; The preset conditions include: The initial confidence level is greater than the preset confidence threshold.

[0014] By adopting the above technical solution, by identifying the sensitive factors of the actual working area corresponding to the seawater environmental parameters, and determining the actual neuron drop-off strategy based on them, it is convenient to accurately consider the influence of seawater environmental factors on the prediction of seawater refractive index. Applying this strategy to the Dropout layer of the preset error compensation model, and by multiple forward propagation and averaging of multiple predicted seawater refractive indices, the error of the model prediction can be effectively reduced, thereby improving the accuracy of the prediction of seawater refractive index.

[0015] In one possible implementation, the seawater environmental parameters are input into the activation error compensation model, and an initial predicted seawater refractive index is obtained after forward propagation, including: The actual discard rate corresponding to each actual neural layer is determined according to the actual neuron discard strategy, and the corresponding layer scaling rate is determined according to the actual discard rate corresponding to each actual neural layer. The actual neural layer includes the bottom neural layer, the middle neural layer and the top neural layer. Obtain the primary layer output result of the bottom neural layer, and perform scaling processing on the primary layer output result according to the layer scaling ratio corresponding to the bottom neural layer to obtain the first optimized input result; The first optimized input result is input into the middle neural layer to obtain the secondary layer output result. The secondary layer output result is scaled according to the layer scaling ratio corresponding to the middle neural layer to obtain the second optimized input result. The second optimized input result is input into the top neural layer to obtain the initial predicted seawater refractive index corresponding to the seawater environmental parameters.

[0016] By adopting the above technical solution, and by setting corresponding layer scaling rates for the bottom, middle, and top neural layers respectively, instead of using a uniform parameter, it is easier to improve the matching degree between the feature learning intensity of each neural layer and the actual computing function. By scaling processing based on the corresponding layer scaling rate after the output of each neural layer, instead of uniformly scaling only at the final output of the model, it is easier to solve the distribution offset of the Dropout mechanism from the root, thereby improving the accuracy of determining the initial predicted seawater refractive index.

[0017] Secondly, this application provides an error compensation system, which adopts the following technical solution: An error compensation system, the error compensation system comprising: At least one processor; Memory; At least one application, wherein the at least one application is stored in memory and configured to be executed by at least one processor, the at least one application being configured to: perform the error compensation method described above for underwater striped structured light 3D reconstruction technology.

[0018] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium includes: a computer program that can be loaded by a processor and executed by the above-described error compensation method for underwater stripe structured light 3D reconstruction technology.

[0019] Fourthly, this application provides a computer program product, which adopts the following technical solution: A computer program product includes a computer program that, when executed by a processor, implements the above-described error compensation method for underwater stripe structured light three-dimensional reconstruction technology.

[0020] In summary, this application includes at least one of the following beneficial technical effects: By analyzing seawater environmental parameters and determining the sensitivity factors of the actual working area corresponding to the actual projection path of underwater fringe structured light, it is easy to analyze the influencing factors that the projection path may encounter in seawater. By using seawater environmental parameters and the sensitivity factors of the actual working area as input, and then using a pre-trained error compensation model that conforms to the characteristics of deep sea, it is easy to improve the adaptability between the analysis process of predicting seawater refractive index and the actual scene, thereby improving the accuracy of determining the predicted seawater refractive index. Based on the highly accurate predicted seawater refractive index, the initial area image is corrected, which improves the effectiveness of the target area image and thus reduces the distortion probability of the target area image.

[0021] By setting corresponding layer scaling rates for the bottom, middle, and top neural layers respectively, instead of using a uniform parameter, it is easier to improve the matching degree between the feature learning intensity of each neural layer and the actual computing function. By scaling the output of each neural layer based on the corresponding layer scaling rate, instead of scaling uniformly only at the final output of the model, it is easier to solve the distribution offset of the Dropout mechanism from the root, thereby improving the accuracy of determining the initial predicted seawater refractive index. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an error compensation method for underwater striped structured light 3D reconstruction technology in an embodiment of this application. Figure 2 This is a schematic diagram of a process for determining the initial predicted seawater refractive index in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an error compensation system according to an embodiment of this application. Detailed Implementation

[0023] The following is in conjunction with the appendix Figures 1 to 3This application will be described in further detail.

[0024] After reading this specification, those skilled in the art may make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] It should be noted that, in the optional embodiments of this application, the data related to object information, when applied to specific products or technologies, requires the permission or consent of the object. Furthermore, the collection, use, and processing of this data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. In other words, if the embodiments of this application involve data related to an object, it must be obtained with the object's authorization and consent, the authorization and consent of relevant departments, and in accordance with the relevant laws, regulations, and standards of the country and region. If the embodiments involve personal information, the acquisition of all personal information requires the individual's consent. If sensitive information is involved, the separate consent of the information subject is required. The embodiments also need to be implemented with the object's authorization and consent.

[0027] Specifically, this application provides an error compensation method for underwater striped structured light 3D reconstruction technology, executed by an error compensation system. This error compensation system can be a server or a terminal device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, and this application does not impose any limitations on this.

[0028] refer to Figure 1 , Figure 1 This is a flowchart illustrating an error compensation method for underwater striped structured light 3D reconstruction technology according to an embodiment of this application. The method includes steps S110-S130, wherein: Step S110: Obtain the seawater environment parameters corresponding to the monitoring area, and determine the sensitivity factor of the actual working area corresponding to the actual projection path of the underwater striped structured light based on the seawater environment parameters.

[0029] Specifically, the monitoring area refers to a specific sea area or underwater space range where underwater striped structured light imaging, seawater refractive index prediction, and image correction are required. The specific range can be set by relevant personnel based on real-time needs. The corresponding seawater environmental parameters for the monitoring area include, but are not limited to, salinity, temperature, and pressure. These parameters are key variables directly determining the physical properties of seawater refractive index. Furthermore, these parameters may influence the seawater's ability to refract light through the coupling effect of molecular density and motion. Seawater's main components are water molecules and dissolved salts. Higher salinity results in a greater total number of molecules per unit volume and a denser molecular arrangement. The essence of seawater refractive index is the degree to which the direction of light propagation changes due to the scattering and absorption of light by molecules in the medium as it passes through. Therefore, the denser the molecules, the stronger their effect on light, and the higher the refractive index. If salinity is ignored, the predicted seawater refractive index... Significant deviations may occur. Temperature affects the refractive index of seawater by altering the intensity of the thermal motion of water molecules. Lower temperatures result in slower water molecule movement, more stable relative positions between molecules, and a more regular overall arrangement, leading to stronger refraction of light and a higher refractive index. Conversely, higher temperatures cause more vigorous water molecule movement, resulting in a looser molecular arrangement, weaker refraction, and a lower refractive index. Without considering temperature, the effectiveness of image correction results may be difficult to guarantee. Pressure is directly related to seawater depth; the deeper the seawater, the greater the pressure. By compressing the volume of seawater, the molecular density per unit volume further increases, leading to an increase in refractive index. Without considering pressure, a highly accurate prediction of the seawater refractive index may not be obtained. Preset integrated sensors can be used to collect seawater environmental parameters of the monitoring area. These sensors can be integrated devices for salinity, temperature, and pressure. They can be deployed using a preset submersible. Specific preset integrated sensors are not limited in this embodiment. After collecting the seawater environmental parameters corresponding to the monitoring area, the preset integrated sensors will upload them to the error compensation system.

[0030] The actual acquisition location corresponding to the seawater environmental parameters can be obtained, and the actual projection path of the underwater striped structured light can be determined based on the actual acquisition location. Then, the actual working area sensitive factors corresponding to the actual projection path can be identified according to the preset feature recognition algorithm. The actual working area sensitive factors include the actual working area temperature and actual absolute depth of the underwater striped structured light actual projection path. The specific preset feature recognition algorithm is not specifically limited in this application embodiment.

[0031] Step S120: Determine the predicted seawater refractive index corresponding to the monitoring area based on seawater environmental parameters, actual working area sensitivity factors, and trained preset error compensation model.

[0032] Specifically, the system performs calculations using a pre-trained error compensation model, outputting the predicted seawater refractive index corresponding to the seawater environmental parameters and the sensitivity factors of the actual working area. Since the training process of the pre-trained error compensation model is well-suited to the deep-sea environment, it facilitates improving the accuracy of the predicted seawater refractive index. The training process of the pre-trained error compensation model may specifically include: A preset error compensation model and a sample environment parameter set are obtained. The sample environment parameter set contains multiple sample environment parameters and the corresponding sample collection location and seawater refractive index for each sample environment parameter. Based on the sample collection location of each sample environment parameter, the sensitivity factor of the sample working area corresponding to each sample environment parameter is determined. The initial loss function is optimized based on the mean of the sensitivity factors of the sample working area sensitivity factors to obtain the target loss function. The target loss function is Loss=MSE(n_pred,n_true)+λα*MSE(n_pred,n_physical), where n_pred is the seawater refractive index predicted by the model, n... _true represents the refractive index of the sample seawater, n_physical represents the empirical refractive index of seawater calculated based on the preset empirical formula for seawater refractive index, λ represents the preset balance weight, and α represents the mean value of the sensitivity factor. Based on each sample environmental parameter in the sample environmental parameter set and the corresponding sample working area sensitivity factor, the sample neuron discarding strategy corresponding to each sample environmental parameter is determined. Based on the sample environmental parameter set, the target loss function, and the sample neuron discarding strategy corresponding to each sample environmental parameter, the preset error compensation model is iteratively trained. Training stops when the output value of the preset error compensation model is within the preset error range, thus obtaining the trained preset error compensation model.

[0033] Specifically, the preset error compensation model is the basic error compensation model and may not possess deep-sea characteristics. Directly performing calculations based on this model could reduce the accuracy of predicting seawater refractive index. The preset error compensation model can be uploaded to the error compensation system in advance by relevant personnel. Similarly, the sample environmental parameter set can also be uploaded to the error compensation system in advance by relevant personnel. This set contains multiple sample environmental parameters and the corresponding seawater refractive index for each parameter. Training the preset error compensation model based on this set improves its adaptability to deep-sea characteristics. The process involves first identifying the sample collection location for each environmental parameter, then obtaining the corresponding sample projection path from the error compensation system based on the sample collection location. The sample projection paths for each environmental parameter can be uploaded to the error compensation system in advance by relevant personnel. The sample projection path is the sample projection path when underwater stripe structured light is projected based on the corresponding environmental parameter. After identifying the corresponding sensitive factor of the working area of ​​each sample projection path based on the preset feature recognition algorithm, the mean value of the sensitive factor of the working area of ​​each environmental parameter is calculated to obtain the mean value of the sensitive factor corresponding to the environmental parameter. The initial loss function used for supervised model training is then optimized based on the mean value of the sensitive factor.

[0034] For any given sample environmental parameter, the sensitivity factor of the sample working area corresponding to the sample environmental parameter is determined based on the sample collection location. Specifically, this may include: Based on the mapping relationship between sample environment parameters and preset weight allocation, the sensitivity factor weights corresponding to the sample environment parameters are determined; based on the sample acquisition location of the sample environment parameters, the sample projection path corresponding to the sample environment parameters is obtained, and the working area temperature sensitivity factor and absolute depth sensitivity factor are identified from the sample projection path; according to the sensitivity factor weights, working area temperature sensitivity factor, absolute depth sensitivity factor, and preset sensitivity factor calculation formula, the sample working area sensitivity factor corresponding to the sample environment parameters is determined, where the preset sensitivity factor calculation formula is: β=sigmoid(k1*T+k2*D), where β is the sample working area sensitivity factor, k1 and k2 are the sensitivity factor weights, T is the working area temperature sensitivity factor, and D is the absolute depth sensitivity factor.

[0035] Specifically, different sample environmental parameters correspond to different weight allocation results, that is, different sensitivity factor weights. The sensitivity factor weights corresponding to the sample environmental parameters can be determined based on a preset weight allocation mapping relationship. The preset weight allocation mapping relationship is the correspondence between sample environmental parameters and sensitivity factor weights. That is, the preset weight allocation mapping relationship is the correspondence between the parameter combination of salinity, temperature and pressure and the sensitivity factor weights. The specific content of this mapping relationship is not specifically limited in this application embodiment. It can be determined by relevant personnel based on historical experimental data and then uploaded to the error compensation system. In the error compensation system, the sample acquisition location of each sample environmental parameter corresponds one-to-one with the corresponding sample projection path. The corresponding sample projection path can be uniquely obtained based on the storage location of the sample acquisition location. The working area temperature sensitivity factor is used to characterize the average temperature of each temperature acquisition point in the sample projection path, and the absolute depth sensitivity factor is used to characterize the vertical depth between the projection start point and the projection end point of the sample projection path. By substituting the sensitivity factor weight, working area temperature sensitivity factor, and absolute depth sensitivity factor into the preset sensitivity factor calculation formula, the sample working area sensitivity factor corresponding to the sample environmental parameter can be calculated. The preset sensitivity factor calculation formula can be: β=sigmoid(k1*T+k2*D), where β is the sample working area sensitivity factor, k1 and k2 are the sensitivity factor weights, T is the working area temperature sensitivity factor, and D is the absolute depth sensitivity factor. By using sigmoid to calculate k1*T+k2*D, it is convenient to normalize and quantize the working area temperature sensitivity factor and the absolute depth sensitivity factor into an interpretable and standardized range between [0,1]. For example, 0 represents extremely low sensitivity, and 1 represents extremely high sensitivity.

[0036] Because the physical formula used in the constraint part of the target loss function regarding the empirical seawater refractive index is an empirical formula fitted after obtaining data points through simulation or sampling, and the deep-sea environment has extreme depth and temperature conditions, the current empirical formula has relatively sparse data points in these areas. Therefore, the error between the empirical seawater refractive index calculated according to the preset empirical formula and the actual situation may be large. Therefore, this application introduces a dynamic sensitivity factor when determining the target loss function. That is, by preset a sensitivity factor formula in the sample working area, in the deep-sea environment, the weight of the existing empirical formula supervision signal in the loss is reduced, that is, the weight of the empirical seawater refractive index in the target loss function is reduced. In the shallow-sea environment, the weight of the empirical formula supervision signal in the loss is increased, that is, the weight of the empirical seawater refractive index in the target loss function is increased. Therefore, combined with the actual deep-sea environment, the β value should be negatively correlated with D, k2 should be negative, the β value should be positively correlated with T, and k1 should be positive.

[0037] The core logic of this application using the sigmoid function lies in addressing the quantification requirements of temperature-sensitive and absolute depth-sensitive factors in the working area, tailored to the characteristics of deep-sea scenarios. On one hand, the sigmoid function's properties strictly constrain the values ​​of both factors within a fixed range, ensuring a unified and comparable quantitative benchmark for parameters under different temperature and absolute depth conditions. On the other hand, leveraging the nonlinear mapping capability of the sigmoid function, it significantly distinguishes subtle differences between the two factors near critical thresholds, facilitating accurate adaptation to the nonlinear effects of temperature and absolute depth changes in deep-sea scenarios. Based on these methods, the working area sensitivity factors corresponding to the environmental parameters of each sample can be determined.

[0038] In addition, the refractive index of the classical physical model corresponding to the sample environmental parameters can be determined according to the preset empirical formula for seawater refractive index, and the model output can be constrained based on this. This helps to provide a physical reasonable boundary for the model output and avoid the predicted seawater refractive index value of the model output from violating the basic optical laws. Specifically, the empirical seawater refractive index can be calculated using the preset empirical formula for seawater refractive index, and then the initial loss function can be optimized using the empirical seawater refractive index. The preset empirical formula for seawater refractive index can be the Millad-Seaver algorithm, and the specific formula is not specifically limited in this embodiment. The initial loss function can be Loss=MSE(n_pred,n_true), where n_pred is the predicted seawater refractive index output after driving the model based on the corresponding sample environmental parameters, and n_true is the sample seawater refractive index corresponding to the corresponding sample environmental parameters. After optimizing the initial loss function based on the mean of the sensitivity factors corresponding to the sample environmental parameter set and the classical physical model refractive index corresponding to each sample environmental parameter, the target loss function is obtained as follows: Loss = MSE(n_pred, n_true) + λα*MSE(n_pred, n_physical), where λ is the preset balancing weight, which can be determined by relevant personnel based on historical experimental data and uploaded to the error compensation system; α is the mean of the sensitivity factor, which can be obtained by averaging the sensitivity factors of the corresponding sample working area for each sample environmental parameter; and n_physical is the empirical seawater refractive index corresponding to each sample environmental parameter, which can be calculated based on the preset empirical formula for seawater refractive index. The optimized target loss function constrains the training process of the preset error compensation model, facilitating the strengthening of model training constraints from three dimensions: data fitting accuracy, consistency with physical laws, and adaptability to deep-sea scenarios. This addresses the problems of the preset error compensation model being prone to overfitting noise, deviating from physical essence, and inaccurate prediction of highly sensitive working areas in deep-sea scenarios, thereby improving the accuracy and robustness of the model's output prediction of seawater refractive index.

[0039] By analyzing the coupling between various sample environment parameters in the sample environment parameter set, it is easier to determine the neuron dropout strategy of the preset error compensation model during the model training phase. Dropping neurons in the Dropout layer of the preset error compensation model during model training helps to reduce overfitting during model training. To improve the accuracy of the sample neuron dropout strategy corresponding to the sample environment parameters, this application provides a specific process for determining the sample neuron dropout strategy based on the sample environment parameters and the corresponding sample working region sensitivity factor: The corresponding sample coupling degree is determined based on the sample environment parameters; based on the sample working area sensitivity factor and sample coupling degree, the sample neuron discarding strategy corresponding to the sample environment parameters is determined from the preset discarding strategy library, and the sample neuron discarding strategy is the discarding rate of each neural layer.

[0040] Specifically, for any sample environmental parameter, the linear correlation strength between each parameter in the sample environmental parameter can be calculated based on a preset feature correlation algorithm to obtain the correlation coefficient between any two parameters. Then, the absolute value of each correlation coefficient or the mean value is calculated. Finally, the mean value is normalized to the [0,1] interval to obtain the final sample coupling degree. The preset feature correlation algorithm can be the Pearson correlation coefficient, and the specific algorithm is not specifically limited in this embodiment. For example, the sample environmental parameters include salinity samples, temperature samples, and pressure samples. First, the correlation coefficients r_ST between salinity samples and temperature samples, r_SP between salinity samples and pressure samples, and r_TP between temperature samples and pressure samples are obtained based on the Pearson correlation coefficient. Then, the absolute values ​​of each correlation coefficient are taken to obtain |r_ST|, |r_SP|, and |r_TP|. The absolute values ​​of the three sets of correlation coefficients are averaged and normalized to the [0,1] interval, where 0 indicates that the parameters are completely independent and 1 indicates that the parameters are completely linearly related, thus obtaining the final sample coupling degree.

[0041] Different combinations of parameters related to sensitivity factors and sample coupling in different sample working regions correspond to different neuron dropout strategies. The neuron dropout strategy corresponding to the parameter combinations of sensitivity factors and sample coupling in the sample working region can be determined from a pre-set dropout strategy library. This library contains sample neuron dropout strategies corresponding to various combinations of sensitivity factors and sample coupling in different working regions. These strategies can be determined by relevant personnel based on historical experimental data and uploaded to the error compensation system in advance. The sample neuron dropout strategy is the dropout rate for each neuron in the Dropout layer; for example, the dropout rate for the top layer is 1%; the dropout rate for the middle layer is 15%; and the dropout rate for the bottom layer is 25%. By employing differentiated dropout rates, shallow layers retain general features while deeper layers reduce dependence on specific patterns and minimize co-adaptation of deep neurons. By introducing sample coupling degree and working region sensitivity factors to jointly determine differentiated dropout strategies, this scheme achieves dynamic adaptation to nonlinear relationships between parameters during model training. Environmental parameter coupling degree reflects the intrinsic correlation strength between different physical quantities; high coupling often means the model is more susceptible to collinearity interference during fitting, thus exacerbating the risk of overfitting. Therefore, the dropout rates of the middle and bottom layers can be appropriately increased. The working region sensitivity factor identifies specific marine environments; in deep-sea environments, the dropout rates of the middle and bottom layers can be appropriately increased to improve model prediction stability. Based on the synergistic adjustment of dropout rates at different network layers, while preserving general marine features at the bottom layer, it effectively constrains the excessive dependence of deep networks on local complex correlations, thereby improving the model's generalization ability. Based on the above method, the sample neuron dropout strategy corresponding to each sample's environmental parameters can be determined. Determining the sample neuron dropout strategy based on the sample working region sensitivity factor and sample coupling degree facilitates more reasonable neuron dropout during model training. After determining the target loss function and the neuron drop-off strategy corresponding to each sample environmental parameter, the sample environmental parameter set can be sequentially input into the preset error compensation model. The drop-off strategy corresponding to each sample environmental parameter is used to drop off neurons in the Dropout layer of the preset error compensation model. The target loss function is used to constrain the output of the preset error compensation model. When the difference between the output value of a sample environmental parameter and the corresponding sample seawater refractive index is within the preset error range, the iterative training stops, indicating that the preset error compensation model has completed training. In order to reduce the influence of random data, the iterative training of the preset error compensation model can be stopped when the difference between the output values ​​of multiple sample environmental parameters and the corresponding sample seawater refractive index is within the preset error range, thus obtaining the trained preset error compensation model.The training process of the preset error compensation model is constrained by the optimized target loss function, which facilitates the strengthening of model training constraints from three dimensions: data fitting accuracy, consistency of physical laws, and adaptability to deep-sea scenarios. At the same time, by determining the corresponding sample neuron discarding strategy according to the actual situation of each sample's environmental parameters, the rationality of the neuron discarding process can be improved.

[0042] Step S130: Obtain the initial area image corresponding to the monitoring area, and correct the initial area image based on the predicted seawater refractive index to obtain the target area image.

[0043] Specifically, the initial area image is the image that needs to be corrected. The initial area image of the monitoring area may be severely distorted due to the characteristics of the seawater environment and cannot directly reflect the true shape and location of the seabed scene. If it is not corrected, subsequent image-based analysis may be completely invalid. Therefore, it is necessary to correct the initial area image according to the determined seawater refractive index with high accuracy. The initial area image can be acquired by relevant image acquisition equipment and uploaded to the error compensation system, or it can be selectively uploaded by relevant personnel according to actual needs. The specific method of acquiring the initial area image is not specifically limited in this application embodiment.

[0044] When correcting the initial region image based on the predicted seawater refractive index, the initial region image can first be mapped onto the air, and then the image can be reconstructed in three dimensions. The specific process is as follows: First, set the focal length of the camera of the preset image acquisition device to f. The preset image acquisition device can be an underwater robot camera. The distance from the camera center to the lens on the surface of the waterproof outer shell is d. The coordinates of a pixel P1 on the edge of a pipe in the initial region image are (u1, v1), and the coordinates of the image center P0 of the initial region image are (u0, v0). Calculate the physical distance di between pixel P1 and the image center P0 according to the two-point distance formula: lateral distance xi = (u1 - u0) * f / dx, where dx is the horizontal pixel size of the camera; vertical distance yi = (v1 - v0) * f / dy, where dy is the vertical pixel size of the camera. Based on Snell's law n_water×sinθwater=n_air×sinθair, and the predicted seawater refractive index n_water obtained through model calculation, the refraction angle θair in air is calculated, and then the incident angle θwater in water is derived.

[0045] The physical coordinates (xb, yb) of the virtual imaging point P1' of the underwater pixel in the air satisfy: xb=xi*(n_water / n_air)* / Based on the above formula, we can obtain xb, and similarly, we can obtain yb.

[0046] Transform the physical coordinates (xb, yb) back to pixel coordinates (u1', v1'), where: u1'=u0+xb*dx / f; v1'=v0+yb*dy / f.

[0047] The above method generates a pixel correction mapping table corresponding to all pixels in the initial region image. Based on this pixel correction mapping table, each pixel in the initial region image is moved to its corresponding correction position to obtain the target region image. The method of correcting the initial region image based on the predicted seawater refractive index to obtain the target region image is not specifically limited in this embodiment. It is sufficient that the actual propagation path of underwater fringe structured light imaging rays can be derived from the predicted seawater refractive index, thereby correcting the object position shifts and shape distortions in the initial region image caused by the dynamic changes in seawater refractive index to be consistent with the actual scene of the monitoring area.

[0048] In this embodiment of the application, by analyzing the seawater environmental parameters and determining the actual working area sensitivity factor corresponding to the actual projection path of the underwater striped structured light, it is convenient to analyze the influencing factors that the projection path may face in seawater. By using the seawater environmental parameters and the actual working area sensitivity factor as input, and then using a pre-trained preset error compensation model that conforms to the characteristics of the deep sea for calculation, it is convenient to improve the adaptability between the analysis process of predicting the seawater refractive index and the actual scene, thereby improving the accuracy of determining the predicted seawater refractive index. Based on the highly accurate predicted seawater refractive index, the initial area image is corrected, which is convenient to improve the effectiveness of the target area image and thus facilitates the reduction of the distortion probability of the target area image.

[0049] Furthermore, this application provides a specific process for determining the predicted seawater refractive index of the monitoring area based on seawater environmental parameters, actual working area sensitivity factors, and a pre-trained preset error compensation model: Based on the actual working area sensitivity factors corresponding to the seawater environmental parameters, the actual neuron dropout strategy corresponding to the actual working area sensitivity factors is determined; the Dropout layer in the pre-trained preset error compensation model is activated based on the actual neuron dropout strategy to obtain the activated error compensation model, and the loop steps are executed until the preset conditions are met. The initial confidence level that meets the preset conditions is determined as the final prediction confidence level, and the initial seawater refractive index corresponding to the final prediction confidence level is determined as the final prediction seawater refractive index. The iterative steps include: inputting seawater environmental parameters into the activation error compensation model, obtaining multiple initial predicted seawater refractive indices after multiple forward propagations; calculating the mean of the multiple predicted seawater refractive indices to obtain the corresponding initial predicted seawater refractive indices; calculating the standard deviation of the multiple predicted seawater refractive indices to obtain the corresponding initial confidence level; and determining whether the initial confidence level is greater than a preset confidence threshold. The preset conditions include: the initial confidence level is greater than the preset confidence threshold.

[0050] Specifically, the method for identifying sensitive factors of the actual working area from actual seawater environmental parameters can refer to the method for identifying sensitive factors of the sample working area based on sample seawater environmental parameters, and will not be elaborated here. The specific method for determining the corresponding actual coupling degree based on actual seawater environmental parameters can refer to the method for determining the corresponding sample coupling degree based on sample environmental parameters, and will not be elaborated here. Finally, based on the actual working area sensitive factors and actual coupling degree, the actual neuron discarding strategy corresponding to the actual seawater environmental parameters is determined from the preset discarding strategy library. The preset discarding strategy library contains actual neuron discarding strategies corresponding to various combinations of actual working area sensitive factors and actual coupling degrees. This library can be determined by relevant personnel based on historical experimental data and uploaded to the error compensation system in advance.

[0051] The Dropout layer in the pre-trained error compensation model is activated based on the actual neuron dropout strategy. Specifically, neurons in each layer of the Dropout layer are awakened according to the actual neuron dropout strategy, resulting in an activation error compensation model corresponding to the actual neuron dropout strategy. To improve the accuracy of seawater refractive index prediction, this application also provides a temporal smoothing preprocessing step. Specifically, the real-time collected seawater environmental parameters are averaged using a sliding window. Only seawater environmental parameters that have undergone sliding window averaging are input into the activation error compensation model. The seawater environmental parameters input into the activation error compensation model are randomly forward-propagated N times to obtain N predicted seawater refractive indices. The mean of the N predicted seawater refractive indices is calculated to obtain the initial predicted seawater refractive index. Simultaneously, the standard deviation of the N predicted seawater refractive indices is calculated to obtain the initial confidence level corresponding to the initial predicted seawater refractive index, which is then used to determine... If the initial confidence level of the first output is not greater than a preset confidence threshold, it indicates that the reliability of the predicted seawater refractive index of the first output is low. In this case, the seawater environmental parameters input into the activation error compensation model need to be randomly forward-propagated N times to obtain a new initial predicted seawater refractive index and a new initial confidence level. Then, it is determined whether the preset conditions are met again. If the preset conditions are met, the initial confidence level that meets the preset conditions is determined as the final predicted confidence level, and the initial seawater refractive index corresponding to the final predicted confidence level is determined as the final predicted seawater refractive index. The specific value of N can be 20 or 25. The specific value of N and the preset confidence threshold are not specifically limited in this embodiment of the application and can be set by relevant personnel according to actual needs.

[0052] Furthermore, this application provides a specific implementation process for inputting seawater environmental parameters into the activation error compensation model and obtaining the initial predicted seawater refractive index after forward propagation, including steps S210-S240, as follows: Figure 2 As shown: Step S210: Determine the actual discard rate for each actual neural layer according to the actual neuron discard strategy, and determine the corresponding layer scaling rate according to the actual discard rate for each actual neural layer, wherein the actual neural layers include the bottom neural layer, the middle neural layer and the top neural layer.

[0053] Specifically, the Dropout layer of the activation error compensation model contains multiple actual neural layers, including bottom, middle, and top neural layers. The bottom layer is located closer to the input, and the top layer is closer to the output. The bottom layer is significantly affected by the predicted seawater refractive index, therefore it typically exhibits a higher neuron dropout rate. The top neural layer is less affected by the predicted seawater refractive index, and therefore generally has a lower neuron dropout rate. The actual dropout rate for each actual neural layer can be identified from the actual neuron dropout strategy using a preset feature recognition algorithm.

[0054] Step S220: Obtain the primary layer output of the bottom neural layer, and scale the primary layer output according to the scaling ratio of the corresponding layer of the bottom neural layer to obtain the first optimized input result.

[0055] Specifically, to address the output distribution shift caused by the random dropping of neurons in the Dropout layer, ensure consistency of model output between training and usage phases, and ultimately improve the accuracy and stability of predicting seawater refractive index, it is necessary to scale the output of each actual neural layer based on its actual dropout rate during the output phase. The output of the bottom layer is the primary layer output; scaling the primary layer output according to the scaling rate of the bottom layer yields the first optimized input.

[0056] Step S230: Input the first optimized input result into the middle neural layer to obtain the secondary layer output result, and scale the secondary layer output result according to the scaling ratio of the middle neural layer to obtain the second optimized input result.

[0057] Specifically, the first optimized input result obtained after scaling is used as the input object of the middle neural layer. After processing, the middle neural layer will output the output result of the secondary layer. The second optimized input result can be obtained by scaling the output result of the secondary layer according to the scaling rate of the middle neural layer.

[0058] Step S240: Input the second optimized input result into the top neural layer to obtain the initial predicted seawater refractive index corresponding to the seawater environmental parameters.

[0059] Specifically, the second optimized input result obtained after scaling is used as the input object of the top neural layer. After processing, the top neural layer will output the final initial predicted seawater refractive index.

[0060] By scaling down the output of each neural layer based on the corresponding layer scaling ratio, rather than scaling down uniformly only at the final output of the model, it is easier to address the distribution offset of the Dropout mechanism at its root, thereby improving the accuracy of determining the initial predicted seawater refractive index.

[0061] This application provides an error compensation system, such as... Figure 3 As shown, Figure 3The error compensation system 300 shown includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the error compensation system 300 may also include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one, and the structure of this error compensation system 300 does not constitute a limitation on the embodiments of this application.

[0062] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0063] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by only one line, but this does not mean that there is only one bus or one type of bus.

[0064] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0065] The memory 303 is used to store application code that executes the solution of this application, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0066] Error compensation systems include, but are not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Tablet PCs), PMPs (Portable Multimedia Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. They can also be used for servers, etc. Figure 3 The error compensation system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0067] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.

[0068] This application provides a computer program product including a computer program that, when executed by a processor, implements the methods described in any of the above embodiments.

[0069] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0070] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An error compensation method for underwater striped structured light 3D reconstruction technology, characterized in that, include: Obtain the seawater environment parameters corresponding to the monitoring area, and determine the actual working area sensitivity factor corresponding to the actual projection path of the underwater striped structured light based on the seawater environment parameters. The predicted seawater refractive index of the monitoring area is determined based on the seawater environmental parameters, the sensitivity factors of the actual working area, and the trained preset error compensation model. An initial region image corresponding to the monitoring area is obtained, and the initial region image is corrected based on the predicted seawater refractive index to obtain the target region image.

2. The error compensation method for underwater striped structured light three-dimensional reconstruction technology according to claim 1, characterized in that, The training process of the preset error compensation model includes: Obtain a preset error compensation model and a sample environment parameter set, wherein the sample environment parameter set contains multiple sample environment parameters and the sample collection location and sample seawater refractive index corresponding to each sample environment parameter; Based on the sample collection location of each sample environmental parameter, the sensitivity factor of the sample working area corresponding to each sample environmental parameter is determined. The initial loss function is optimized based on the mean of the sensitivity factor of the sample working area sensitivity factor to obtain the target loss function. The target loss function is Loss=MSE(n_pred,n_true)+λα*MSE(n_pred,n_physical), where n_pred is the model predicted seawater refractive index, n_true is the sample seawater refractive index, n_physical is the empirical seawater refractive index calculated based on the preset empirical formula for seawater refractive index, λ is the preset balance weight, and α is the mean of the sensitivity factor. Based on each sample environment parameter and the corresponding sample working area sensitivity factor in the sample environment parameter set, determine the sample neuron discarding strategy corresponding to each sample environment parameter; Based on the sample environment parameter set, the target loss function, and the sample neuron discarding strategy corresponding to each sample environment parameter, the preset error compensation model is iteratively trained. Training stops when the output value of the preset error compensation model is within the preset error range, thus obtaining the trained preset error compensation model.

3. The error compensation method for underwater striped structured light three-dimensional reconstruction technology according to claim 2, characterized in that, Based on the sample collection location using sample environmental parameters, the sensitivity factors of the sample working area corresponding to the sample environmental parameters are determined, including: Based on the sample environment parameters and the preset weight allocation mapping relationship, the sensitivity factor weights corresponding to the sample environment parameters are determined; Based on the sample collection location of the sample environment parameters, obtain the sample projection path corresponding to the sample environment parameters, and identify the temperature sensitive factor and absolute depth sensitive factor of the working area based on the sample projection path. Based on the sensitivity factor weights, the working area temperature sensitivity factor, the absolute depth sensitivity factor, and the preset sensitivity factor calculation formula, the sample working area sensitivity factor corresponding to the sample environment parameters is determined, wherein the preset sensitivity factor calculation formula is: β = sigmoid(k1*T + k2*D), where β is the sensitivity factor of the working area of ​​the sample, k1 and k2 are the sensitivity factor weights, T is the temperature sensitivity factor of the working area, and D is the absolute depth sensitivity factor.

4. The error compensation method for underwater striped structured light three-dimensional reconstruction technology according to claim 2, characterized in that, Based on sample environment parameters and corresponding sample working area sensitivity factors, determine the sample neuron discarding strategy corresponding to the sample environment parameters, including: The corresponding sample coupling degree is determined based on the sample environment parameters; Based on the sample working area sensitivity factor and the sample coupling degree, the sample neuron discarding strategy corresponding to the sample environment parameter is determined from the preset discarding strategy library, and the sample neuron discarding strategy is the discarding rate of each neural layer.

5. The error compensation method for underwater striped structured light three-dimensional reconstruction technology according to claim 2, characterized in that, The step of determining the predicted seawater refractive index corresponding to the monitoring area based on the seawater environmental parameters, the sensitivity factor of the actual working area, and the trained preset error compensation model includes: Based on the actual working area sensitivity factors corresponding to the seawater environmental parameters, determine the actual neuron discarding strategy corresponding to the actual working area sensitivity factors; Based on the actual neuron dropout strategy, the Dropout layer in the trained preset error compensation model is activated to obtain the activated error compensation model, and the loop steps are executed until the preset conditions are met. The initial confidence level that meets the preset conditions is determined as the final prediction confidence level, and the initial seawater refractive index corresponding to the final prediction confidence level is determined as the final prediction seawater refractive index. The loop steps include: The seawater environmental parameters are input into the activation error compensation model, and multiple initial predicted seawater refractive indices are obtained after multiple forward propagations. The average of the multiple predicted seawater refractive indices is calculated to obtain the corresponding initial predicted seawater refractive index. The standard deviation of the multiple predicted seawater refractive indices is calculated to obtain the corresponding initial confidence levels. Determine whether the initial confidence level is greater than a preset confidence threshold; The preset conditions include: The initial confidence level is greater than the preset confidence threshold.

6. The error compensation method for underwater striped structured light three-dimensional reconstruction technology according to claim 5, characterized in that, The seawater environmental parameters are input into the activation error compensation model, and after forward propagation, the initial predicted seawater refractive index is obtained, including: The actual discard rate corresponding to each actual neural layer is determined according to the actual neuron discard strategy, and the corresponding layer scaling rate is determined according to the actual discard rate corresponding to each actual neural layer. The actual neural layer includes the bottom neural layer, the middle neural layer and the top neural layer. Obtain the primary layer output result of the bottom neural layer, and perform scaling processing on the primary layer output result according to the layer scaling ratio corresponding to the bottom neural layer to obtain the first optimized input result; The first optimized input result is input into the middle neural layer to obtain the secondary layer output result. The secondary layer output result is scaled according to the layer scaling ratio corresponding to the middle neural layer to obtain the second optimized input result. The second optimized input result is input into the top neural layer to obtain the initial predicted seawater refractive index corresponding to the seawater environmental parameters.

7. An error compensation system, characterized in that, The error compensation system includes: At least one processor; Memory; At least one application, wherein the at least one application is stored in memory and configured to be executed by at least one processor, the at least one application being configured to: perform an error compensation method for underwater stripe structured light three-dimensional reconstruction technology according to any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, include: The computer program is stored and can be loaded by a processor and executed as described in any one of claims 1-6, which is a method for error compensation in underwater striped structured light 3D reconstruction technology.

9. A computer program product, characterized in that, The system includes a computer program that, when executed by a processor, implements the steps of an error compensation method for underwater stripe structured light three-dimensional reconstruction technology as described in any one of claims 1-6.

Images