A multi-beam three-dimensional acoustic detection system based on self-learning

Abstract

The present application relates to the field of underwater acoustic detection and oceanographic surveying technology, and particularly relates to a multi-beam three-dimensional acoustic detection system based on self-learning, which learns historical data through a training unit, establishes a mapping relationship between regional information and detection parameters; uses an information input unit to input environmental parameters of a region to be detected, automatically generates an optimized deployment scheme and acoustic wave parameters by a command generation unit, drives a transducer array to perform wide emission and narrow reception acoustic detection; after preprocessing and three-dimensional modeling, the results are compared with the mapping model by a verification unit, and the error causes are diagnosed by an analysis unit, and finally the system parameters are adaptively adjusted by a processing unit. Through self-learning and closed-loop feedback regulation technology, the present application realizes adaptive optimization of the detection system to different underwater environments and effectively improves the three-dimensional reconstruction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of underwater acoustic detection and oceanographic surveying, and particularly relates to a multi-beam three-dimensional acoustic detection system based on self-learning. BACKGROUND

[0002] Multi-beam acoustic detection is a key technology for obtaining underwater three-dimensional terrain and topography information, and is widely used in the fields of marine resource investigation, underwater engineering survey, navigation safety guarantee, etc. The detection effect of the traditional multi-beam system is highly dependent on the experience of the operator, and the layout of the transducer array, the acoustic emission parameters (such as frequency, pulse width, power) and the signal processing parameters need to be manually set according to the water depth, bottom material, hydrological environment (such as temperature, salinity, flow rate) and other factors of the operation area. This process is tedious and inefficient, and it is difficult to achieve optimal configuration, especially when facing complex and variable or unknown waters, improper parameter configuration can easily lead to a decrease in the quality of detection data, distortion of the three-dimensional model or the existence of a large number of blind areas.

[0003] The prior art lacks a mechanism that can intelligently and automatically optimize the overall configuration of the system according to the detection environment. Although some systems have parameter adjustment functions, they are usually limited to a single link (such as signal filtering) or rely on fixed rules, and cannot form a complete closed loop from environmental perception, scheme generation, array construction, data acquisition to effect verification and parameter re-optimization. This leads to poor system adaptability and makes it difficult to maintain high-precision and high-reliability three-dimensional reconstruction capability in different scenarios.

[0004] Chinese patent application No. CN120009864B discloses an AI-based dredging engineering multi-beam data rapid processing method, belonging to the field of dredging engineering; when extracting the feature matrix Xα, the method directly extracts the data information of the same angle beam to form the feature matrix Xα, and forms multiple feature matrices Xα through the data information of different angle beams; the RNN neural network is trained through the multiple feature matrices Xα, and the output result can intuitively reflect the water depth and geological information of the same angle reflection point; the difference between the independent variables is increased by the matrices of different angles, which speeds up the training speed of the RNN network, so that the RNN network model can be trained faster; at the same time, the setting of each element parameter in the output matrix facilitates the mapping of the reflection points to the GPS in batches in the later stage, and the production of seabed geological distribution map through three-dimensional software.

[0005] However, the prior art still has the following problems:

[0006] When facing complex waters, improper parameters often lead to poor data quality and model distortion, and there is a lack of intelligent overall configuration optimization mechanism, and the system lacks universality and precision guarantee. SUMMARY

[0007] To this end, the application provides a multi-beam three-dimensional acoustic detection system based on self-learning, so as to overcome the problems in the prior art that when facing complex water areas, the data quality is poor and the model is distorted due to improper parameters, the overall configuration optimization mechanism lacks intelligence, and the system lacks universality and precision guarantee.

[0008] To achieve the above-mentioned purpose, the application provides a multi-beam three-dimensional acoustic detection system based on self-learning. It comprises:

[0009] A training unit is used to determine the mapping relationship between the region information and the detection information based on historical data;

[0010] A detection unit comprises a plurality of transducers, each of which can emit a wide sector acoustic wave and receive a narrow beam acoustic wave;

[0011] An information input unit is used to receive the region information of the area to be measured, which at least includes one or more of volume, maximum depth, terrain complexity, temperature, pressure and flow rate;

[0012] An instruction generation unit is connected to the information input unit and is used to generate corresponding detection information based on the region information, and issue instructions based on the detection information, wherein the detection information includes the position parameters and the acoustic wave parameters of each transducer;

[0013] An array construction unit is connected to the detection unit and the instruction generation unit respectively, and is used to set each transducer at a corresponding position in the area to be measured to form a transducer array through instructions;

[0014] A preprocessing unit is connected to the detection unit and is used to preprocess the narrow beam acoustic wave collected by the detection unit;

[0015] A three-dimensional construction unit is connected to the preprocessing unit and is used to construct a three-dimensional model of the area to be measured based on the preprocessed narrow beam acoustic wave;

[0016] A verification unit is connected to the training unit, the information input unit and the three-dimensional construction unit respectively, and is used to obtain a mapping model corresponding to the region information, and verify the three-dimensional model with the mapping model, and determine whether the training is completed based on the verification result;

[0017] An analysis unit is connected to the verification unit and is used to analyze the reason for incomplete training based on the verification result when it is determined that the training is not completed, and generate corresponding processing instructions based on the reason for incomplete training;

[0018] A processing unit is connected to the preprocessing unit, the instruction generation unit, the detection unit and the analysis unit respectively, and is used to adjust the system parameters based on the processing instructions.

[0019] Specifically, the specific structure of the analysis unit is not limited, which can be composed of a logic component including a field programmable processor, a computer and a microprocessor in the computer.

[0020] Further, the verification unit is used to verify the three-dimensional model with a mapping model, comprising:

[0021] The three-dimensional model is divided into several sub-regions to obtain several sub-models,

[0022] The mapping model is divided into several sub-regions in the same way to obtain several mapping sub-models;

[0023] The calibration points of the sub-models and the mapping sub-models are determined, and the calibration points of the sub-models and the mapping sub-models are overlapped,

[0024] The area ratio of the overlapped surface to the total area of the mapping sub-model is calculated to obtain an area ratio;

[0025] The average value of the area ratio of each sub-model and the mapping sub-model is calculated,

[0026] Based on the environmental factors and the transducer factors, a correction coefficient is determined, and the average value is corrected by using the correction coefficient to obtain a comprehensive matching degree.

[0027] Further, the verification unit is used to determine whether the training is completed based on the verification result, comprising:

[0028] If the comprehensive matching degree is greater than or equal to a preset comprehensive matching degree, it is determined that the training is completed;

[0029] If the comprehensive matching degree is less than the preset comprehensive matching degree, it is determined that the training is not completed, and the analysis unit analyzes the reason for the incomplete training based on the area ratio.

[0030] Further, the analysis unit analyzes the reason for the incomplete training based on the area ratio, comprising:

[0031] The variance of the area ratio of each sub-model and the mapping sub-model is calculated,

[0032] If the variance is less than or equal to a preset variance, it is determined that the reason for the incomplete training is that the pretreatment process is unqualified;

[0033] If the variance is greater than the preset variance, the corresponding transducer is marked, and the analysis unit analyzes the reason for the incomplete training based on the distribution of the marked transducer.

[0034] Furthermore, if the analysis unit determines that the preprocessing process is unqualified, it calculates the difference between the preset comprehensive matching degree and the comprehensive matching degree to obtain the comprehensive matching degree difference, adjusts the signal-to-noise ratio based on the comprehensive matching degree difference, and sends an instruction to the processing unit, wherein the increase in signal-to-noise ratio is positively correlated with the comprehensive matching degree difference.

[0035] Furthermore, the analysis unit is also used to correct the signal-to-noise ratio based on the angular width of the narrow beam of the transducer when adjusting the signal-to-noise ratio, and send instructions to the processing unit, wherein the increase in signal-to-noise ratio is negatively correlated with the angular width.

[0036] Furthermore, the verification unit is also used to verify the re-acquired 3D model and the mapping model when the processing unit completes the adjustment of the signal-to-noise ratio, and determine whether training is completed based on the verification result. If it is determined that training is not completed, the analysis unit analyzes the reason for the failure to complete training based on the distribution of the marked transducers.

[0037] Furthermore, the analysis unit analyzes the reasons for incomplete training based on the distribution of the labeled transducers, including:

[0038] Calculate the physical distance between any two marked transducers;

[0039] The distribution dispersion is calculated based on the physical distance between all marked transducers;

[0040] If the distribution dispersion is greater than or equal to the preset dispersion threshold, the reason for the failure to complete the training is determined to be the influence of terrain, and the angle width is adjusted based on the complexity of the 3D model.

[0041] If the distribution dispersion is less than the preset dispersion threshold, the reason for the failure to complete the training is determined to be the influence of local environmental factors.

[0042] Furthermore, the analysis unit adjusts the angle width based on the complexity of the 3D model, including:

[0043] Calculate the average ratio of the area of ​​each plane in the 3D model to the standard area to obtain the average area ratio.

[0044] The complexity is obtained by multiplying the reciprocal of the average area ratio, the correction factor, and the number of faces in the 3D model.

[0045] The angle width is adjusted based on the complexity of the 3D model, and instructions are sent to the processing unit, wherein the reduction in angle width is positively correlated with the complexity.

[0046] Further, the analysis unit is configured to, in a condition that the reason for the incomplete training is determined to be affected by a local environmental factor, acquire a temperature gradient corresponding to a position where the local marked transducer is located in the region information, and adjust an acoustic wave acquisition parameter of the corresponding transducer based on the temperature gradient, wherein the acoustic wave acquisition parameter includes an acoustic wave emission power and / or a signal sampling frequency, and an increase amplitude of the acoustic wave emission power and an increase amplitude of the signal sampling frequency are positively correlated with the temperature gradient.

[0047] Compared with the prior art, the beneficial effects of the present application are that the present application realizes quantitative and refined evaluation of the training effect of the self-learning system through the above-mentioned verification method, can effectively locate the spatial distribution of modeling errors through local comparison by dividing sub-regions, and makes the evaluation standard more suitable for actual operation conditions by introducing a dynamic correction coefficient based on the actual situation of the environment and the device, thereby avoiding unfair judgments that may be caused by using a fixed threshold in different environments, so that whether the system has completed sufficient training can be more scientifically and accurately judged to reach the standard of being put into actual use.

[0048] Further, the present application introduces a preset comprehensive matching degree as an explicit and quantitative training completion criterion, converts the judgment of the system training state from subjective experience to objective data comparison, and realizes automatic decision of the termination of the training process. This not only avoids performance not meeting the standard caused by insufficient training or resource waste caused by overtraining, but also ensures that each substandard detection can automatically trigger the diagnosis and optimization process through the closed-loop mechanism of "substandard analysis", thereby driving the system to converge to a usable state stably and efficiently through iterative self-learning.

[0049] Further, the present application introduces "variance" as a key diagnostic index, skillfully decomposes the complex modeling failure problem into two categories of "globality" and "locality", and realizes rapid and automatic preliminary screening of the fault reason by using the preset variance as a threshold. This avoids lengthy investigation of all possible reasons, so that the system can quickly locate the main cause of the problem, whether it is a common data processing problem or a specific acquisition unit problem, thereby greatly improving the efficiency and pertinence of the system self-diagnosis and self-optimization process, and is one of the core decision logics for the system to realize efficient self-learning.

[0050] Further, the present application realizes intelligent and adaptive adjustment of the pretreatment link parameters through the above-mentioned mechanism. The core innovation is that the performance gap (comprehensive matching difference) of the whole system is directly and proportionally mapped to the specific adjustment amount of the core pretreatment parameter (signal-to-noise ratio). This method avoids the adjustment method of fixed step or blind trial by experience, so that each parameter adjustment has a clear target and measurement. The system can automatically prescribe the corresponding "dose" (signal-to-noise ratio improvement amplitude) according to the severity of the "disease" (performance gap), so as to quickly correct the modeling failure caused by data quality problems with the most efficient iteration number, and significantly improve the speed and stability of system self-learning convergence.

[0051] Further, the present application introduces the key acoustic parameter of angular width to finely correct the signal-to-noise ratio adjustment amount. By making the signal-to-noise ratio improvement amplitude negatively related to the angular width, the system can provide stronger signal enhancement and noise suppression support for high-precision narrow beams, thereby ensuring that the most sensitive and demanding units in the whole array are prioritized and fully optimized. This differentiated parameter adjustment strategy makes the overall performance improvement more balanced and efficient, and is an important intelligent guarantee for improving the cooperative performance of complex acoustic array systems.

[0052] Further, the present application sets up an iterative closed loop of adjustment-retest-reanalysis, which greatly enhances the depth and reliability of system self-learning and self-repair. It overcomes the shortcomings of one-time and single-path optimization that may fall into local optimum or misjudgment. When the general parameter optimization effect is not good, the system can automatically switch the diagnosis dimension to use the distribution information of the transducer to explore the potential physical environmental root cause. This multi-stage and multi-angle progressive problem diagnosis mechanism enables the system to cope with more complex fault scenarios, significantly improving the success rate and robustness of achieving stable working state through autonomous learning in an undesirable or unknown environment.

[0053] Further, the present application introduces the spatial statistical feature of distribution dispersion as a diagnostic key to realize accurate spatial attribution of modeling failure root cause. It converts the abstract distribution into a calculable quantitative index and presets a scientific threshold, so that the system can automatically and reliably distinguish whether the problem is caused by global topographic challenges or local environmental disturbances. This diagnostic method directly relates to the spatial characteristics of the physical world, providing a decisive basis for adopting completely different optimization strategies (such as adjusting the angular width to cope with complex topography or optimizing local acquisition parameters to resist environmental disturbances), greatly improving the system's ability to intelligently cope with complex and variable underwater environments, and is the core step for the system to realize advanced situational awareness and adaptive decision-making.

[0054] Further, the present application quantifies the abstract three-dimensional model geometry into a specific "complexity" index through the above method, and uses it as the direct basis for dynamically optimizing the acoustic detection core parameter (angular width), and its core lies in the realization that: in the face of complex terrain, high resolution (narrow angular width) detection is crucial for depicting details, and the present application combines the average area ratio and the total surface number to make the complexity calculation not only perceive the fragmentation degree of the terrain, but also perceive the fine degree of the model, so as to more comprehensively evaluate the level of terrain challenge, and based on the complexity, the beam is narrowed on demand, so that the system can intelligently optimize the trade-off between detection coverage and resolution, and the more complex the terrain is, the more inclined to sacrifice a certain coverage to obtain higher local precision. This realizes the adaptive matching between the acoustic detection parameters and the real-time perceived terrain features, significantly improves the accuracy and detail restoration ability of three-dimensional reconstruction of complex underwater terrain, and is a key adaptive strategy for the system to intelligently cope with unknown complex environment.

[0055] Further, the present application realizes intelligent diagnosis and accurate compensation of detection failure caused by local environmental disturbance (typified by temperature gradient) through the above method, and its core lies in: the quantitative information of the environmental physical quantity (temperature gradient) is directly and proportionally mapped to the adjustment instruction of the acoustic acquisition hardware core parameter (transmitting power, sampling frequency), and this "measurement-quantization-compensation" closed loop enables the system to actively adapt to the complex marine hydrological environment, effectively offsets the negative effects caused by the inhomogeneity of the medium by enhancing the signal strength and improving the sampling accuracy, and significantly improves the robustness and quality of data acquisition in complex local environments such as temperature stratification and vortex, which marks the key evolution of the system from passive reception of environmental influence to active perception and compensation of environmental disturbance, greatly improving the reliability and data accuracy of the multi-beam acoustic detection system in actual complex marine scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0056] Fig. 1 It is a structural schematic diagram of the multi-beam three-dimensional acoustic detection system based on self-learning of the present application.

[0057] Fig. 2 It is a working flowchart of the verification unit in the multi-beam three-dimensional acoustic detection system based on self-learning of the present application. DETAILED DESCRIPTION

[0058] In order to make the purpose and advantages of the present application more clear and explicit, the present application will be further described below in combination with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the present application.

[0059] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art will appreciate that the embodiments are only used to explain the technical principles of the present application, and are not intended to limit the protection scope of the present application.

[0060] It should be noted that, in the description of the present application, the terms of direction or position relationship such as "upper", "lower", "left", "right", "inner", "outer" and the like are based on the direction or position relationship shown in the drawings, which is only for the convenience of description, and does not indicate or imply that the device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0061] In addition, it should be noted that, in the description of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connection" should be understood in a broad sense, for example, can be fixedly connected, or can be detachably connected, or integrally connected; can be mechanically connected, or can be electrically connected; can be directly connected, or can be indirectly connected through an intermediate medium; can be the communication inside two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0062] Please refer to Figs. 1-2 shown, Fig. 1 which is a structural schematic diagram of a self-learning-based multi-beam three-dimensional acoustic detection system of the present application; Fig. 2 which is a working flowchart of a checking unit in the self-learning-based multi-beam three-dimensional acoustic detection system of the present application.

[0063] The self-learning-based multi-beam three-dimensional acoustic detection system of the present application comprises:

[0064] a training unit, configured to determine a mapping relationship between regional information and detection information based on historical data;

[0065] a detection unit, comprising a plurality of transducers, each transducer being capable of emitting a wide-sector acoustic wave and receiving a narrow-beam acoustic wave;

[0066] an information input unit, configured to receive regional information of a region to be detected, the regional information comprising at least one or more of volume, maximum depth, terrain complexity, temperature, pressure and flow rate;

[0067] an instruction generation unit, connected to the information input unit, configured to generate corresponding detection information based on the regional information, and issue an instruction based on the detection information, wherein the detection information comprises position parameters of each transducer and acoustic wave parameters;

[0068] an array construction unit connected to the detection unit and the instruction generation unit respectively, configured to set each transducer at a corresponding position in the region to be detected to form a transducer array through instructions;

[0069] a preprocessing unit connected to the detection unit, configured to preprocess the narrow-beam sound waves collected by the detection unit;

[0070] a three-dimensional construction unit connected to the preprocessing unit, configured to construct a three-dimensional model of the region to be detected based on the preprocessed narrow-beam sound waves;

[0071] a verification unit connected to the training unit, the information input unit and the three-dimensional construction unit respectively, configured to obtain a mapping model corresponding to the region information, and verify the three-dimensional model with the mapping model, and determine whether the training is completed based on the verification result;

[0072] an analysis unit connected to the verification unit, configured to analyze the reason for incomplete training based on the verification result when it is determined that the training is not completed, and generate corresponding processing instructions based on the reason for incomplete training;

[0073] a processing unit connected to the preprocessing unit, the instruction generation unit, the detection unit and the analysis unit respectively, configured to adjust system parameters based on the processing instructions.

[0074] Specifically, in the embodiment, the system has a self-learning mode and a working mode; in the self-learning mode, the system iteratively optimizes according to the feedback of the analysis unit and the processing unit; when the verification result indicates that the real-time three-dimensional model reaches the preset accuracy, the system switches to the working mode.

[0075] In the present application, the historical data refers to a data set collected in past detection tasks, verified, containing region information, corresponding detection system parameters and final high-precision three-dimensional model; the mapping relationship refers to a mathematical model from region information to optimal detection system parameters (including array layout and acoustic parameters) induced from historical data through machine learning algorithms (such as neural networks, regression models); the standard three-dimensional mapping model refers to a three-dimensional model corresponding to the case most matched with the current region information in historical data, or a theoretically optimal three-dimensional model generated under the guidance of the mapping relationship, serving as a benchmark for the verification link.

[0076] Specifically, the specific structure of the analysis unit is not limited, which can be composed of logical components including field programmable processors, computers and microprocessors in computers.

[0077] Specifically, the verification unit is configured to verify the three-dimensional model with the mapping model, including:

[0078] dividing the three-dimensional model into several sub-regions to obtain several sub-models,

[0079] dividing the mapping model into several sub-regions in the same way to obtain several mapping sub-models;

[0080] determining calibration points of the sub-models and the mapping sub-models, and coinciding the calibration points of the sub-models and the mapping sub-models,

[0081] calculating a ratio of an area of the coinciding surface to a total area of the mapping sub-models to obtain an area ratio;

[0082] calculating an average value of the area ratios of the sub-models and the mapping sub-models,

[0083] determining a correction coefficient based on environmental factors and transducer factors, and correcting the average value by using the correction coefficient to obtain a comprehensive matching degree.

[0084] In the embodiment of the present application, firstly, the three-dimensional model of the to-be-detected area constructed by the system is uniformly divided into a plurality of sub-regions according to spatial positions, and a sub-model is generated for each sub-region. Meanwhile, the mapping model corresponding to the current area information obtained from the training unit is also divided in the same way to generate the same number of mapping sub-models. Then, model calibration is performed to determine a calibration point for each sub-model and its corresponding mapping sub-model. The calibration point is usually selected as a feature obvious reference point in the sub-region, such as a terrain turning point or a pre-set landmark. The calibration points of the sub-model and the mapping sub-model are aligned. Then, the coincidence degree of each sub-region is calculated. After the calibration point coincidence is completed, the area of the coincident part of the surface of the sub-model and the surface of the mapping sub-model is calculated. The coincident area is divided by the total surface area of the mapping sub-model to obtain an area ratio value. The closer the area ratio value is to 1, the higher the modeling accuracy of the sub-region is. Subsequently, a preliminary overall matching degree is calculated. The area ratios of all sub-regions are arithmetically averaged to obtain an average value. The average value preliminarily reflects the overall similarity between the three-dimensional model and the mapping model. However, interference factors in the actual detection environment will affect the fairness of the matching degree. Therefore, it is necessary to introduce a correction coefficient to correct the above average value to obtain a more scientific comprehensive matching degree. The correction coefficient is determined according to the deviation degree of the actual environment from the ideal reference environment. The correction coefficient mainly includes two factors, an environmental factor correction, for example, if the actual seawater flow rate during detection is greater than the reference flow rate during training, the disturbance of the water flow to the sound wave is enhanced, and the matching degree should be allowed to decrease. Therefore, the factor correction value is the "reference flow rate" divided by the "actual flow rate". The ratio is less than 1, thereby downwardly correcting the average value (i.e., multiplying a number less than 1), so that the judgment standard is relatively relaxed when the actual environment is worse. A transducer factor correction: for example, if the actual transducer array spacing is greater than the reference spacing during training, the spatial sampling density is reduced, and the matching degree is expected to decrease. Therefore, the factor correction value is the "reference spacing" divided by the "actual spacing", and the average value is also downwardly corrected. The comprehensive matching degree is the average value adjusted by the above one or more correction coefficients. The "preset comprehensive matching degree" is the threshold value for determining whether the system training is qualified. The determination method is as follows: during the system debugging stage, a plurality of typical historical detection cases with modeling quality artificially judged as "qualified" are selected, the comprehensive matching degrees of the cases are calculated, and the minimum value of the comprehensive matching degrees of the cases is taken as the preset comprehensive matching degree.

[0085] Specifically, in the embodiment, the correction coefficient is obtained by multiplying each factor sub-coefficient. Each factor sub-coefficient is defined as an ideal reference value / actual measured value. The ideal reference value of the environmental factor is taken from the average value of the corresponding environmental parameter in the historical data set. The ideal reference value of the transducer factor is taken from the system nominal design value.

[0086] The present application realizes quantitative and refined evaluation of the training effect of the self-learning system through the above-mentioned verification method. By dividing the sub-regions for local comparison, the spatial distribution of modeling errors can be effectively located. By introducing a dynamic correction coefficient based on the actual situation of the environment and equipment, the evaluation standard is more in line with the actual working conditions, avoiding unfair judgments that may be caused by using fixed thresholds in different environments, so that it is more scientific and accurate to judge whether the system has completed sufficient training and reached the standard of being put into actual use.

[0087] Specifically, the verification unit is configured to determine whether the training is completed based on the verification result, including:

[0088] If the comprehensive matching degree is greater than or equal to a preset comprehensive matching degree, it is determined that the training is completed.

[0089] If the comprehensive matching degree is less than the preset comprehensive matching degree, it is determined that the training is not completed, and the analysis unit analyzes the reason for the incomplete training based on the area ratio.

[0090] In the embodiment of the present application, after a complete detection and modeling process is performed, the verification unit calculates a comprehensive matching degree representing the overall accuracy of the current modeling. The value is compared with a pre-set threshold, i.e., a "pre-set comprehensive matching degree". If the calculated comprehensive matching degree is greater than or equal to the pre-set threshold, it indicates that the overall matching degree of the three-dimensional model constructed this time and the mapping model as the standard answer reaches or exceeds the qualified requirement. Therefore, the verification unit determines that the system has completed the training, and the parameter configuration and processing capacity of the system are sufficient to cope with the current regional information detection task. The system can end the training mode and be ready for actual use. On the contrary, if the calculated comprehensive matching degree is less than the pre-set threshold, it indicates that there is a significant gap between the modeling result and the standard answer, and the overall accuracy is unqualified. The verification unit determines that the system "has not completed the training". At this time, the verification unit triggers the analysis unit to intervene, and provides the area ratio data calculated by each sub-region to the analysis unit. The analysis unit further digs into the deep reasons for the low overall matching degree based on the area ratio data reflecting the local accuracy, such as whether it is a general problem in the data preprocessing link or a specific problem caused by a specific position transducer or local environment, so as to provide a precise direction for subsequent parameter adjustment. The determination method of the "pre-set comprehensive matching degree" threshold is as follows: during the system development and debugging stage, a large number of historical detection cases in different typical scenes are collected, which are artificially evaluated by field experts as "modeling success" or "acceptable accuracy". The comprehensive matching degrees of these successful cases are calculated to form a qualified matching degree data set. The minimum value of all values in the data set, or a lower quantile (such as the fifth percentile) according to the statistical distribution, is taken as the final pre-set comprehensive matching degree. This method ensures that the threshold represents the lowest acceptable performance of the system. Only when the system performance reaches the minimum level of the historical successful cases, it is considered to be qualified for training.

[0091] The present application introduces the pre-set comprehensive matching degree as an explicit and quantitative training completion criterion, which converts the judgment of the system training state from subjective experience to objective data comparison, and realizes the automatic decision of the termination of the training process. This not only avoids the performance not meeting the standard caused by insufficient training or the waste of resources caused by overtraining, but also ensures that each unqualified detection can automatically trigger the diagnosis and optimization process through the closed-loop mechanism of "unqualified analysis", so as to drive the system to converge to a usable state stably and efficiently through iterative self-learning.

[0092] Specifically, the analysis unit analyzes the reasons for not completing the training based on the area ratio, including:

[0093] calculating the variance of the area ratio of each sub-model and the mapping sub-model,

[0094] If the variance is less than or equal to the preset variance, it is determined that the reason for the incomplete training is that the pretreatment process is unqualified;

[0095] If the variance is greater than the preset variance, the corresponding transducer is marked, and the analysis unit analyzes the reason for the incomplete training based on the distribution of the marked transducer.

[0096] In the embodiment of the application, when the checking unit determines that the comprehensive matching degree does not meet the standard (i.e., the training is not completed), the analysis unit first calculates the variance of the "area ratio" data of all sub-regions. The variance is a statistical quantity for measuring the dispersion degree of a group of data. In this scenario, the size of the variance directly reflects the uniformity of the modeling accuracy of each sub-region. The analysis unit compares the calculated actual variance with a reference threshold value called "preset variance", and makes a preliminary cause judgment accordingly. Case one: if the actual variance is less than or equal to the preset variance, it indicates that the area ratio values of each sub-region are not significantly different from each other, and the dispersion degree is low. Combined with the premise that the "comprehensive matching degree is low", it can be inferred that the modeling accuracy of all sub-regions is generally and uniformly low. This global accuracy decline is most likely caused by problems in the common pre-processing link of the original sound wave data, so the analysis unit determines that the main reason for the incomplete training this time is "the pretreatment process is unqualified", for example, improper parameter setting of general algorithms such as filtering, gain control or noise suppression, resulting in damage to the basic quality of all detection data. Case two: if the actual variance is greater than the preset variance. This indicates that the area ratio values of each sub-region are highly uneven, and the dispersion degree is high. That is, there are some sub-regions with acceptable modeling accuracy, but there are also some sub-regions with very poor accuracy. This local accuracy drop is usually directly related to the specific transducer responsible for data acquisition in that sub-region or the local environment in which it is located. At this time, the analysis unit will start the marking mechanism, marking the transducers corresponding to the sub-regions whose area ratio is lower than a certain qualified threshold (for example, lower than a certain percentage of the median of all sub-region area ratios) as "suspected abnormal transducers". Then, the analysis will shift to a deeper root cause analysis based on the spatial distribution of these marked transducers, for example, to determine whether it is a general difficulty caused by complex terrain or a specific failure caused by local hotspots (such as strong temperature gradient, turbulence).

[0097] In the embodiment of the application, the determination method of the preset variance threshold is as follows: during the system debugging phase, simulate or collect a batch of case data that leads to overall failure of modeling due to "improper pretreatment parameter setting", calculate the variance of the area ratio in these cases, and take the maximum value of these variance values as the reference benchmark of the "preset variance". This means that when the actual observed variance does not exceed the maximum dispersion degree that can be caused by pretreatment problems in history, the cause is preferentially attributed to pretreatment problems.

[0098] The application ingeniously decomposes the complex modeling failure problem into two categories of "globality" and "locality" by introducing variance as a key diagnostic index, and realizes rapid and automatic preliminary screening of fault causes by using the preset variance threshold, which avoids lengthy investigation of all possible causes directly, enables the system to quickly locate the problem main cause direction, and is a common data processing problem or a specific acquisition unit problem, thereby greatly improving the efficiency and pertinence of the system self-diagnosis and self-optimization process, and is one of the core decision logics for realizing efficient self-learning of the system.

[0099] Specifically, the analysis unit calculates the difference between the preset comprehensive matching degree and the comprehensive matching degree to obtain a comprehensive matching degree difference under the condition that the pretreatment process is unqualified, adjusts the signal-to-noise ratio based on the comprehensive matching degree difference, and sends an instruction to the processing unit, wherein the increase amplitude of the signal-to-noise ratio is positively correlated with the comprehensive matching degree difference.

[0100] In the embodiment of the application, the analysis unit first performs quantitative evaluation: calculates the numerical difference between the "preset comprehensive matching degree" and the current measured "comprehensive matching degree" to obtain a "comprehensive matching degree difference", which directly and quantitatively reflects the size of the gap between the current system performance and the qualified standard. Subsequently, the analysis unit generates an optimization instruction for the preprocessing unit based on this difference, and the core of the instruction is to improve the signal-to-noise ratio of the acoustic data. The specific adjustment logic is that the increase amplitude of the signal-to-noise ratio is positively correlated with the calculated comprehensive matching degree difference, that is, the greater the performance gap (the greater the difference), the more serious the noise interference in the original data or the weaker the effective signal, so a larger amplitude is required to enhance the signal and suppress the noise, and therefore the instruction requires a larger amplitude of signal-to-noise ratio improvement. On the contrary, if the gap is small, only a small amplitude of optimization adjustment is required. The instruction is sent to the processing unit, and the processing unit adjusts the related algorithm parameters of the preprocessing unit according to the instruction after receiving the instruction. For example, the gain of the signal amplification circuit can be increased, the strength of the noise suppression link in the digital filtering algorithm can be enhanced, or the number of signal averaging processing can be adjusted, and the common purpose is to make the collected acoustic data have a higher signal-to-noise ratio in the subsequent detection cycle, thereby providing purer and more reliable input data for three-dimensional modeling.

[0101] The application realizes intelligent and adaptive adjustment of the pretreatment link parameters through the above mechanism. The core innovation is that the performance gap (comprehensive matching difference) of the whole system is directly and proportionally mapped to the specific adjustment amount of the core pretreatment parameter (signal-to-noise ratio). This method avoids the adjustment mode of fixed step or blind trial by experience, so that each parameter adjustment has a clear target and measurement. The system can automatically provide the corresponding "dose" (signal-to-noise ratio improvement amplitude) according to the severity of the "disease" (performance gap), so as to quickly correct the modeling failure caused by data quality problems with the most efficient iteration number, and significantly improve the speed and stability of system self-learning convergence.

[0102] Specifically, the analysis unit is also used to correct the signal-to-noise ratio based on the angle width of the narrow beam of the transducer when the adjustment of the signal-to-noise ratio is completed, and send instructions to the processing unit, wherein the increase amplitude of the signal-to-noise ratio is negatively correlated with the angle width.

[0103] In the embodiment of the application, the analysis unit will further consider the physical characteristic angle width of the narrow beam sound waves emitted by each transducer. The angle width refers to the opening angle of the sound wave beam. The smaller the angle width, the sharper the beam and the stronger the directivity, and the higher the spatial resolution, but at the same time, its coverage and signal strength may be relatively limited. Based on this physical characteristic, the analysis unit introduces a correction principle: the final increase amplitude of the signal-to-noise ratio needs to be inversely adjusted according to the angle width of the narrow beam of the corresponding transducer. That is, the increase amplitude of the signal-to-noise ratio is negatively correlated with the angle width. Specifically, for narrow beams with small angle width, due to their high inherent directivity accuracy, they are more sensitive to background noise and the signal strength may be weak, so a larger signal-to-noise ratio needs to be increased to ensure that their high-precision advantages can be played and the signal weakness can be overcome. Therefore, a larger additional increase of the signal-to-noise ratio is given. Conversely, for beams with large angle width, they have wide coverage and strong signal, so the demand for signal-to-noise ratio improvement is relatively moderate. Therefore, the increase amplitude of the signal-to-noise ratio is correspondingly reduced. The analysis unit combines the basic increase amplitude determined by the "comprehensive matching difference" and the correction amplitude determined by the "angle width" to calculate the final and individual signal-to-noise ratio adjustment amount for each or each type of transducer, and generates new instructions to send to the processing unit.

[0104] By introducing the key acoustic parameter of angle width to fine-tune the signal-to-noise ratio adjustment amount, the system can provide stronger signal enhancement and noise suppression support for high-precision narrow beams by making the signal-to-noise ratio increase amplitude negatively correlated with the angle width, so as to ensure that the most sensitive and demanding units in the whole array are given priority and sufficient optimization. This differentiated parameter adjustment strategy makes the overall performance improvement of the system more balanced and efficient, and is an important intelligent guarantee for improving the cooperative performance of complex acoustic array systems.

[0105] Specifically, the checking unit is further configured to check the re-acquired three-dimensional model against the mapping model when the processing unit completes the adjustment of the signal-to-noise ratio, determine whether the training is completed based on a checking result, and if it is determined that the training is not completed, analyze a reason for the training not being completed based on a distribution of the labeled transducers.

[0106] In the embodiment of the present application, when the processing unit completes the adjustment of the signal-to-noise ratio parameter of the preprocessing unit according to the instruction of the analysis unit, the system does not immediately make a final determination. The checking unit starts a new detection and modeling cycle: the detection unit reacquires the acoustic data of the region to be detected using the optimized parameter, the preprocessing unit processes the data at the new signal-to-noise ratio level, and the three-dimensional construction unit generates a new three-dimensional model accordingly. The checking unit then acquires the newly generated "optimized three-dimensional model" and checks it again against the "mapping model" as the standard answer, and calculates a new comprehensive matching degree. The checking unit re-determines whether the system has completed the training according to the same preset comprehensive matching degree standard. If the new comprehensive matching degree meets the standard, it is determined that the system has met the requirements after this round of optimization and the training is completed. If the new comprehensive matching degree still does not meet the standard, the system determines that the previous round of optimization for the unqualified preprocessing has not completely solved the problem. At this time, the checking unit will keep or recalculate the area ratio of each sub-region and identify the sub-region with insufficient accuracy and the corresponding transducer again, and mark them. Then, the analysis unit is activated again. However, the focus of this analysis is shifted to the spatial distribution of these marked transducers. This is because, after a round of optimization aimed at improving the overall data quality, if the problem still exists, the root cause is likely to be no longer a global data-based problem, but a local factor closely related to a specific spatial position. The analysis unit will further diagnose the deeper reasons by analyzing whether these "problem transducers" are scattered throughout the region or concentrated in a certain specific local area, such as whether the complex terrain causes acoustic scattering or the local strong flow, thermocline and other environmental factors cause signal distortion.

[0107] The present application greatly enhances the depth and reliability of system self-learning and self-repair by setting an iterative closed loop of adjustment-re-measurement-re-analysis, which overcomes the shortcomings of one-time and single-path optimization that may fall into local optimum or misjudgment. When the general parameter optimization effect is not good, the system can automatically switch the diagnosis dimension and use the distribution information of the transducers to explore the potential physical environmental root cause. This multi-stage and multi-angle progressive problem diagnosis mechanism enables the system to cope with more complex fault scenarios, significantly improving the success rate and robustness of the system in achieving a stable working state through autonomous learning in an undesirable or unknown environment.

[0108] Specifically, the analysis unit analyzes the reason for the training not being completed based on the distribution of the labeled transducers, including:

[0109] calculating the physical distance between any two marked transducers;

[0110] calculating the distribution dispersion based on the physical distances between all marked transducers;

[0111] if the distribution dispersion is greater than or equal to a preset dispersion threshold, determining that the reason for incomplete training is terrain influence, and adjusting the angular width based on the complexity of the three-dimensional model;

[0112] if the distribution dispersion is less than the preset dispersion threshold, determining that the reason for incomplete training is local environmental factor influence.

[0113] In the embodiment of the present application, the analysis unit first obtains the accurate layout position coordinates of all transducers from the array construction unit. For any two marked transducers, the straight-line physical distance between them is calculated. Then, based on the set of physical distances between all marked transducers, a statistical quantity that can represent the overall distribution density is calculated, called distribution dispersion. In the present application, the distribution dispersion can be represented by calculating the standard deviation or average distance of these distances. The larger the value, the farther apart the marked transducers are from each other, and the more dispersed the distribution is. The smaller the value, the more concentrated the positions of the transducers are. The analysis unit compares the calculated actual distribution dispersion with a reference value called "preset dispersion threshold". If the distribution dispersion is greater than or equal to the preset dispersion threshold: this indicates that the performance abnormal transducers are widely and dispersedly present in the detection area, and there is no obvious spatial aggregation. This pattern is usually closely related to the overall terrain complexity of the detection area. For example, there may be a large number of steep slopes, gullies or complex reef structures in the entire area, which cause abnormal reflection, scattering or shielding of sound waves in multiple different positions, so that the transducers dispersedly distributed in the detection area are all affected. Therefore, the analysis unit determines that the main reason for incomplete training is "terrain influence". If the distribution dispersion is less than the preset dispersion threshold: this indicates that the performance abnormal transducers are concentratedly distributed in one or a few relatively small local areas in space. This strong aggregation strongly suggests that the problem is not caused by the general terrain, but by the environmental abnormalities specific to these local areas. Common examples are local existence of strong temperature gradient layer (thermocline), salinity mutation, or eddy, internal wave and other hydrological phenomena, which can significantly change the sound wave propagation speed and path, causing serious distortion of local data. Therefore, the analysis unit determines that the main reason for incomplete training is local environmental factor influence.

[0114] In the embodiment of the present application, the method for determining the preset discrete degree threshold value is as follows: in the system debugging stage, two typical scenarios are analyzed through simulation or historical data, one is a scenario caused by known complex terrain, in which the marker transducers are widely distributed, the discrete degree of which is calculated and the minimum value is taken as a reference; the other is a scenario caused by known local hot spots (such as simulated hot spring mouth), in which the marker transducers are concentratedly distributed, the discrete degree of which is calculated and the maximum value is taken as a reference, and the preset threshold value is usually set between the two reference values to effectively distinguish the two distribution modes.

[0115] The present application realizes accurate spatial attribution of the root cause of modeling failure by introducing the spatial statistical feature of distribution discrete degree as a diagnostic key, which converts the abstract distribution into a calculable quantitative index and presets a scientific threshold value, so that the system can automatically and reliably distinguish whether the problem is caused by global terrain challenges or local environmental disturbances. This diagnostic method directly relates to the spatial characteristics of the physical world, and provides a decisive basis for taking completely different optimization strategies (such as adjusting the angle width to cope with complex terrain, or optimizing local acquisition parameters to resist environmental disturbance), greatly improving the system's ability to intelligently cope with complex and variable underwater environments, and is a core step for the system to realize advanced situation awareness and adaptive decision-making.

[0116] Specifically, the analysis unit adjusts the angle width based on the complexity of the three-dimensional model, including:

[0117] calculating the average value of the ratio of the area of each plane in the three-dimensional model to the standard area to obtain an average area ratio,

[0118] calculating the product of the reciprocal of the average area ratio, the correction multiple, and the number of surfaces of the three-dimensional model to obtain the complexity,

[0119] adjusting the angle width based on the complexity of the three-dimensional model, and sending instructions to the processing unit, wherein the reduction amplitude of the angle width is positively correlated with the complexity.

[0120] Specifically, the processing unit drives the detection unit to perform acoustic detection and three-dimensional modeling again according to the adjusted angle width; the verification unit verifies the newly obtained three-dimensional model and the mapping model again, and if the comprehensive matching degree is still lower than the preset value, the analysis unit repeats the step of analyzing based on the distribution of the marker transducer or triggers manual intervention.

[0121] In the embodiment of the present application, first, the analysis unit obtains the three-dimensional model data of the current construction from the three-dimensional construction unit. The model is composed of a large number of facets (usually triangular facets). The analysis unit performs the following calculation steps: calculate the average area ratio: traverse all the planes in the three-dimensional model, calculate the actual area of each plane. Compare these areas with a "standard plane area" (this area is the median of all plane areas in this detection task), and obtain the "area ratio" of each plane (i.e. actual area / standard area). Then, calculate the arithmetic mean of all plane area ratios to obtain the "average area ratio", which reflects the average deviation of the model plane from the ideal size. The smaller the value, the more the model is composed of a large number of broken facets; the larger the value, the more regular or larger the model plane. Calculate the model complexity: the analysis unit then calculates the model complexity;

[0122] The calculation method is: multiply the reciprocal of the average area ratio (i.e. 1 divided by the average area ratio) by the total number of surfaces of the three-dimensional model, and then multiply by a "correction multiple" less than 1. The "correction multiple" is an empirical coefficient (for example, 0.2) used to balance the order of magnitude and prevent the complexity value from being too large. This coefficient is determined according to typical scenarios during system debugging. The "reciprocal of the average area ratio" will be large when the model plane is generally small (i.e. high fragmentation); the "total number of surfaces of the three-dimensional model" directly reflects the precision or data volume of the model. The product of the two represents the characteristics of the model in terms of precision and fragmentation. The larger the product, the more rugged the model surface, the more details and the more complex the structure. After obtaining the quantitative "model complexity" value, the analysis unit determines the adjustment amount of the angle width according to the preset adjustment rule. The rule is: the reduction amplitude of the angle width is positively correlated with the calculated model complexity. Determine the adjustment amount: the system has a pre-set "complexity-angle width adjustment amount" correspondence (for example, calibrated by experiment). The analysis unit calculates the corresponding angle width reduction value (for example, in degrees) according to the calculated complexity by looking up the table or through a linear relationship. The higher the complexity, the larger the recommended angle width reduction amplitude. Generate and execute instructions: the analysis unit converts this adjustment amount into specific control instructions, such as "set the transmission narrow beam angle width of all marked transducers to Z degrees" or "reduce the angle width by W degrees based on the current basis". The instruction is sent to the processing unit. After the processing unit analyzes the instruction, it adjusts the driving signal parameters of the corresponding transducer (such as changing the phase distribution of the transmission array) to accurately control the narrowing of the angle width of the transmitted beam.

[0123] The application quantifies the abstract three-dimensional model geometry into a specific "complexity" index by the above method, and uses it as a direct basis for dynamically optimizing the acoustic detection core parameter (angular width). The core is to realize that: in the face of complex terrain, high resolution (narrow angular width) detection is essential to depict details. The application combines "average area ratio" and "total surface number" to make the complexity calculation not only perceive the fragmentation of the terrain, but also perceive the fineness of the model, so as to more comprehensively evaluate the level of terrain challenge. Based on this complexity, the beam is narrowed "on demand", so that the system can intelligently optimize the trade-off between detection coverage and resolution - the more complex the terrain, the more inclined to sacrifice a certain coverage to gain higher local precision. This realizes the adaptive matching between acoustic detection parameters and the real-time perceived terrain features, significantly improves the accuracy and detail restoration ability of three-dimensional reconstruction of complex underwater terrain, and is the key adaptive strategy for the system to intelligently cope with unknown complex environment.

[0124] Specifically, the analysis unit is configured to, under the condition that it is determined that the reason for the unfinished training is the influence of local environmental factors, obtain a temperature gradient corresponding to a position where a local marked transducer is located in the region information, and adjust an acoustic wave acquisition parameter of the corresponding transducer based on the temperature gradient, wherein the acoustic wave acquisition parameter includes an acoustic wave emission power and / or a signal sampling frequency, and an increase amplitude of the acoustic wave emission power and an increase amplitude of the signal sampling frequency are positively correlated with the temperature gradient.

[0125] Specifically, the processing unit drives the corresponding marked transducer to re-collect data according to the adjusted acoustic wave acquisition parameter; after the system updates the three-dimensional model with the new data, the verification unit re-verifies, and if the comprehensive matching degree meets the standard, the training is completed; if not, the analysis unit can further analyze other environmental parameters (such as salinity gradient, flow rate) or expand the range of marked transducers for deep diagnosis.

[0126] In the embodiment of the present application, the analysis unit first circumscribes one or more "problematic local areas" in the measurement region according to the spatial coordinates of the tagged transducers. Subsequently, it obtains detailed environmental parameters within these "problematic local areas" from the real-time or historical regional information provided by the information input unit, focusing on the temperature field data. By calculating the degree of temperature variation within the region, i.e., the "temperature gradient" (e.g., the temperature change value per unit horizontal or vertical distance), the intensity of local environmental disturbance is quantified. The greater the temperature gradient, the stronger the vertical or horizontal inhomogeneity of the seawater medium at that location, and the more significant the distortion of sound wave propagation speed and path. Based on the calculated "temperature gradient" value, the analysis unit determines the adjustment strategy for the sound wave collection parameters of the tagged transducers located in the region. The goal of adjustment is to compensate for signal attenuation and distortion caused by medium inhomogeneity. Determine the adjustment reference: the system internally presets a "reference temperature gradient" reference value and a corresponding "reference parameter set". When the measured temperature gradient is greater than the reference value, it is considered necessary to start compensation adjustment. Calculate the adjustment amplitude: the adjustment amplitude (increase) of the parameters is positively correlated with the amount of temperature gradient exceeding the reference value. That is, the greater the temperature gradient, the greater the increase in sound wave transmission power, and the greater the increase in signal sampling frequency. Transmission power adjustment: increase the transmission power, aiming to enhance the intensity of the initial sound wave signal to counteract the additional energy loss and signal attenuation that may be caused by medium inhomogeneity, ensuring that there is enough sound wave energy to reach the target and be reflected back. Sampling frequency adjustment: increase the signal sampling frequency, aiming to improve the time resolution of the system to the received echo signal. Since temperature changes will change the local sound speed, causing the echo signal waveform to spread or distort in the time domain, a higher sampling frequency helps to capture these change details more finely, providing a richer data basis for subsequent signal processing and speed correction. The analysis unit will generate explicit control instructions (such as "increase the transmission power of transducers numbered X to Y in region A by P%" and "set the sampling frequency of transducers in this region to FHz") based on the specific increase in transmission power and / or sampling frequency calculated for each "problematic local area". After receiving the instructions, the processing unit accurately reconfigures the parameters of the driving circuit (controls power) and analog-to-digital converter (controls sampling rate) of the specified transducers.

[0127] The application realizes intelligent diagnosis and accurate compensation of detection failure caused by local environmental disturbance (typified by temperature gradient) by the above method, and the core is that the quantitative information of the environmental physical quantity (temperature gradient) is directly and proportionally mapped to the adjustment instruction of the acoustic acquisition hardware core parameter (transmitting power, sampling frequency). The closed loop of "measurement-quantization-compensation" enables the system to actively adapt to the complex marine hydrological environment, and through enhancing the signal strength and improving the sampling accuracy, the system effectively offsets the negative effects caused by the medium inhomogeneity, significantly improves the robustness and quality of data acquisition in the complex local environment such as temperature stratification and vortex, which marks the key evolution of the system from passive receiving environmental influence to active sensing and compensating environmental disturbance, greatly improves the reliability and data accuracy of the multi-beam acoustic detection system in the actual complex marine scene.

[0128] So far, the technical solutions of the application have been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the application, and the technical solutions after the changes or replacements will fall within the protection scope of the application.

[0129] The above is only the preferred embodiment of the application and is not used to limit the application; for those skilled in the art, the application can have various changes and variations, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A self-learning based multi-beam three-dimensional acoustic probing system, characterized in that, The application relates to a system and method for training a three-dimensional model of a region. The system comprises: a training unit configured to determine a mapping relationship between region information and detection information based on historical data; a detection unit comprising a plurality of transducers, each transducer being capable of emitting a wide sector acoustic wave and receiving a narrow beam acoustic wave; an information input unit configured to receive region information of a region to be detected, the region information comprising at least one or more of volume, maximum depth, terrain complexity, temperature, pressure and flow rate; an instruction generation unit connected to the information input unit and configured to generate corresponding detection information based on the region information, and to issue an instruction based on the detection information, wherein the detection information comprises position parameters of each transducer and acoustic wave parameters; an array construction unit connected to the detection unit and the instruction generation unit respectively, and configured to cause each transducer to be arranged at a corresponding position in the region to be detected by the instruction to form a transducer array; a preprocessing unit connected to the detection unit and configured to preprocess the narrow beam acoustic wave collected by the detection unit; a three-dimensional construction unit connected to the preprocessing unit and configured to construct a three-dimensional model of the region to be detected based on the preprocessed narrow beam acoustic wave; a verification unit connected to the training unit, the information input unit and the three-dimensional construction unit respectively, and configured to obtain a mapping model corresponding to the region information, to verify the three-dimensional model and the mapping model, and to determine whether the training is completed based on a verification result; an analysis unit connected to the verification unit and configured to analyze a reason for incomplete training based on the verification result when it is determined that the training is not completed, and to generate a corresponding processing instruction based on the reason for incomplete training; a processing unit connected to the preprocessing unit, the instruction generation unit, the detection unit and the analysis unit respectively, and configured to adjust system parameters based on the processing instruction; the verification unit is configured to verify the three-dimensional model and the mapping model, comprising: dividing the three-dimensional model into a plurality of sub-regions to obtain a plurality of sub-models, dividing the mapping model into a plurality of sub-regions in the same manner to obtain a plurality of mapping sub-models; determining calibration points of the sub-models and the mapping sub-models, and making the calibration points of the sub-models and the mapping sub-models coincide, calculating a ratio of an area of a coincident surface to a total area of the mapping sub-models to obtain an area ratio, calculating an average value of the area ratio of each sub-model and the mapping sub-model, determining a correction coefficient based on environmental factors and transducer factors, and correcting the average value by using the correction coefficient to obtain a comprehensive matching degree; the verification unit is configured to determine whether the training is completed based on the verification result, comprising: if the comprehensive matching degree is greater than or equal to a preset comprehensive matching degree, it is determined that the training is completed; 2. The self-learning based multi-beam three-dimensional acoustic probing system according to claim 1, characterized in that, if the comprehensive matching degree is less than the preset comprehensive matching degree, it is determined that the training is not completed, and the analysis unit analyzes the reason for incomplete training based on the area ratio. the analysis unit analyzes the reason for incomplete training based on the area ratio, comprising: calculating a variance of the area ratio of each sub-model and the mapping sub-model, if the variance is less than or equal to a preset variance, it is determined that the reason for incomplete training is that a preprocessing process is unqualified. If the variance is greater than the preset variance, the corresponding transducer is marked, and the analysis unit analyzes the reason for incomplete training based on the distribution of the marked transducers.

3. The self-learning based multi-beam three-dimensional acoustic probing system according to claim 2, characterized in that, The analysis unit calculates a difference between the preset comprehensive matching degree and the comprehensive matching degree to obtain a comprehensive matching degree difference under the condition that the pretreatment process is determined to be unqualified, adjusts the signal-to-noise ratio based on the comprehensive matching degree difference, and sends an instruction to the processing unit, wherein the increase amplitude of the signal-to-noise ratio is positively correlated with the comprehensive matching degree difference.

4. The self-learning based multi-beam three-dimensional acoustic probing system according to claim 3, characterized in that, The analysis unit is also used to correct the signal-to-noise ratio based on the angular width of the transducer narrow beam when the adjustment of the signal-to-noise ratio is completed, and sends an instruction to the processing unit, wherein the increase amplitude of the signal-to-noise ratio is negatively correlated with the angular width.

5. The self-learning based multi-beam three-dimensional acoustic probing system according to claim 3, wherein, The verification unit is also used to verify the re-acquired three-dimensional model with the mapping model when the processing unit completes the adjustment of the signal-to-noise ratio, and determines whether the training is completed based on the verification result, and if it is determined that the training is not completed, the analysis unit analyzes the reason for incomplete training based on the distribution of the marked transducers.

6. The self-learning based multi-beam three-dimensional acoustic probing system according to claim 5, characterized in that, The analysis unit analyzes the reason for incomplete training based on the distribution of the marked transducers, including: calculating the physical distance between any two marked transducers; calculating the distribution dispersion based on the physical distance between all marked transducers; if the distribution dispersion is greater than or equal to a preset dispersion threshold, determining that the reason for incomplete training is terrain influence, and adjusting the angular width based on the complexity of the three-dimensional model; if the distribution dispersion is less than the preset dispersion threshold, determining that the reason for incomplete training is local environmental factor influence.

7. The self-learning based multi-beam three-dimensional acoustic probing system according to claim 6, characterized in that, The analysis unit adjusts the angular width based on the complexity of the three-dimensional model, including: calculating the average of the ratio of the area of each plane in the three-dimensional model to the standard area to obtain an average area ratio, calculating the product of the reciprocal of the average area ratio, the correction multiplier, and the number of surfaces of the three-dimensional model to obtain the complexity, adjusting the angular width based on the complexity of the three-dimensional model and sending an instruction to the processing unit, wherein the decrease amplitude of the angular width is positively correlated with the complexity.

8. The self-learning based multi-beam three-dimensional acoustic probing system according to claim 7, characterized in that, The analysis unit is used to obtain the temperature gradient corresponding to the position of the local marked transducer in the area information under the condition that the reason for incomplete training is determined to be local environmental factor influence, and adjust the sound wave acquisition parameter of the corresponding transducer based on the temperature gradient, wherein the sound wave acquisition parameter includes sound wave emission power and / or signal sampling frequency, and the increase amplitude of the sound wave emission power and the increase amplitude of the signal sampling frequency are positively correlated with the temperature gradient.

Images