AI image enhanced acoustic tomography multi-mode river flow monitoring system

Abstract

The invention discloses an AI image enhanced acoustic tomography multi-modal river flow monitoring system, and relates to the technical field of graphic data sensing and river flow monitoring, and the system comprises an acoustic tomography monitoring module which employs M sequence spread spectrum to measure sound wave data; the image acquisition and AI enhancement module outputs an enhanced image through denoising and super-resolution reconstruction; the multi-modal data synchronization control module is used for realizing millisecond-level alignment of acoustic and visual data; the multi-source data fusion module is used for extracting and normalizing a multi-dimensional feature vector by using a cross-modal attention mechanism; the AI intelligent flow calculation module is used for fitting the mapping relation through a neural network and calculating the flow; the data storage and visual display module is used for storing data by adopting an edge and cloud dual-architecture; and the remote control and parameter configuration module realizes two-way communication through an internet of things interface. According to the invention, sound tomography and AI enhanced image technologies are integrated, monitoring coverage is comprehensive, power consumption is low, operation is stable, and efficient and reliable support is provided for water resource management.

Description

Technical Field

[0001] This invention relates to the field of graphic data sensing and river flow monitoring technology, and in particular to an AI image-enhanced acoustic tomography multimodal river flow monitoring system. Background Technology

[0002] River flow monitoring is a core foundation for water resource management, flood control and disaster reduction, and ecological protection. The accuracy and real-time nature of the data directly affect the scientific basis of related decisions. Currently, the mainstream monitoring methods are mainly single-point intrusive or in-situ observations, using equipment such as acoustic Doppler current profilers and temperature-salinity-depth (TDT) meters. These methods require direct deployment of instruments in the water body, covering only a local flow velocity within the probe's detection range, making it difficult to reflect the average flow of the river cross-section. Furthermore, they often employ intermittent, mobile observations, which cannot meet the needs of real-time monitoring. In complex river environments, instrument deployment is easily affected by fishing activities and shipping traffic, making implementation difficult, and incurring high costs in terms of manpower and resources, as well as a heavy burden of long-term operation and maintenance.

[0003] The emergence of acoustic tomography offers a new approach to solving the aforementioned problems. It collects average temperature and flow data by penetrating the entire water body with sound waves, enabling large-scale monitoring without immersion in the water, and possesses potential advantages in real-time performance and high accuracy. However, existing acoustic tomography monitoring systems still have significant technical shortcomings. Relying solely on acoustic data for flow calculation is susceptible to interference from factors such as water turbulence, water turbidity, and environmental noise, leading to unstable signal transmission and decreased data acquisition accuracy. Furthermore, while some systems incorporate image acquisition equipment, the acquired images are easily affected by environmental factors such as changes in lighting, water flow disturbances, and strong reflections, resulting in problems such as high noise levels, low resolution, and blurred key features, failing to provide effective supplementation for flow calculation. The lack of targeted image enhancement processing techniques makes it difficult for visual data to effectively complement acoustic data, thus failing to fully leverage the advantages of multimodal monitoring.

[0004] Furthermore, existing multimodal monitoring systems generally suffer from poor data synchronization, rigid fusion mechanisms, and insufficient anti-interference capabilities. Inconsistent timestamps across different modalities lead to distorted fusion results; feature fusion fails to consider dynamic changes in data quality, making low-quality data prone to interfering with calculations; and the lack of differentiated anti-interference strategies against complex interference sources such as shipping noise and electronic equipment interference further impacts monitoring accuracy. Simultaneously, challenges in remote river areas, difficulties in quickly diagnosing equipment failures, and insufficient data interaction security also limit the large-scale application and long-term stable operation of existing systems. As water resource management demands increasingly higher accuracy, real-time performance, and continuity in monitoring data, existing technologies can no longer meet the diverse needs of various scenarios. There is an urgent need for an integrated monitoring system that combines AI image enhancement, precise multimodal fusion, strong anti-interference capabilities, and low power consumption to overcome the current technological bottlenecks. Summary of the Invention

[0005] This invention proposes an AI-enhanced image-based acoustic tomography multimodal river flow monitoring system to address the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an AI image-enhanced acoustic tomography multimodal river flow monitoring system, comprising the following modules: The acoustic tomography monitoring module consists of at least four acoustic station nodes deployed on both banks of the river cross section. Each acoustic station node integrates an omnidirectional underwater acoustic transducer, a signal transmitting unit, a signal receiving unit, and a demodulation module. It uses M-sequence spread spectrum modulation and matched filtering algorithm to measure the sound wave transmission time and velocity, and simultaneously collects acoustic propagation data of the entire river cross section. The image acquisition and AI enhancement module is equipped with a high-definition network camera and image preprocessing unit at the river cross section to acquire visual image data. Through adaptive denoising algorithm, super-resolution reconstruction technology and edge enhancement algorithm, the image is optimized and the enhanced image is output. The multimodal data synchronization control module has a built-in millisecond-level timestamp synchronization unit and data alignment engine. It receives acoustic and visual data and timestamps the two types of data through Beidou / GPS timing signals. The multi-source data fusion module uses a cross-modal attention mechanism to extract acoustic and visual features, constructs a multi-dimensional feature vector containing acoustic parameters, image features, and environmental parameters, performs normalization processing, and transmits the fused feature vector to the AI ​​solution module. The AI ​​intelligent flow calculation module has a built-in pre-trained flow prediction model based on deep learning. It takes multimodal feature vectors as input and combines them with river cross-section data. It uses a deep neural network algorithm to fit the mapping relationship between acoustic propagation speed, image flow velocity texture and actual flow, and calculates instantaneous flow and average flow. The data storage and visualization module stores recently collected raw data, enhanced images, and calculation results at the edge, while the cloud database stores monitoring data and historical records long-term. It also features a visualization interface to display traffic data, acoustic propagation curves, enhanced images, and traffic change trends. The remote control and parameter configuration module enables bidirectional communication with the remote control center through an IoT communication interface, receiving remote control commands to complete parameter updates, status queries, and fault alarm processing.

[0007] Furthermore, it also includes an image enhancement quality quantification evaluation module, which constructs a multi-dimensional image quality evaluation model and quantifies the usability of the enhanced image through comprehensive indicators, expressed as follows: ; in A comprehensive score for image enhancement quality. This is the texture sharpness weighting coefficient. To enhance the texture sharpness value of the post-image, The texture sharpness threshold for a standard sharp image. This is the contrast weighting coefficient. To enhance the contrast of the resulting image, The contrast threshold for a standard image. This is the signal-to-noise ratio weighting coefficient. To enhance the signal-to-noise ratio of the resulting image, To preset the signal-to-noise ratio threshold, Only enhanced images that meet the set scoring criteria can participate in multimodal data fusion. It is also equipped with an AI adaptive enhancement parameter adjustment unit to dynamically optimize the denoising intensity, super-resolution reconstruction iteration number, and edge enhancement coefficient based on real-time river environment monitoring data.

[0008] Furthermore, it also includes a multimodal feature fusion weight dynamic adjustment module, which adaptively allocates fusion weights based on the real-time quality status of acoustic and image data to enhance the contribution ratio of the data. The expression is as follows: ; in For the fusion weights of acoustic data, For the fusion weights of visual image data, Score the credibility of acoustic data. For the signal integrity of acoustic data, This is the acoustic weighting adjustment coefficient. For the feature integrity of image data, The visual weight adjustment coefficient is used. The dynamic weight allocation mechanism adjusts the fusion ratio according to the real-time quality of the two types of data. When the quality of a certain type of data decreases, its weight ratio is automatically reduced. A feature quality real-time detection unit is added. Through a triple mechanism of acoustic signal-to-noise ratio analysis, image texture clarity detection, and feature dimension integrity verification, the quality level of the two types of data is determined in real time. At the same time, a weight adjustment buffer mechanism is established and a weight change rate threshold is set.

[0009] Furthermore, it also includes a dynamic correction module for river cross-section geometric parameters. Combining preset cross-section benchmark parameters, it uses image measurement algorithms to calculate geometric parameters in real time, automatically correcting deviations in cross-section parameters caused by water level changes and riverbed siltation. The corrected geometric parameters are synchronized to the AI ​​intelligent flow calculation module in real time. It also integrates a multi-view image fusion unit, which uses high-definition cameras deployed on the river cross-section to collect multi-view images and uses stereo vision algorithms to generate a three-dimensional point cloud model of the cross-section. At the same time, it introduces a historical cross-section data trend correction mechanism, which filters out correction errors caused by short-term abnormal disturbances by analyzing the long-term accumulated changes in cross-section geometric parameters.

[0010] Furthermore, it also includes an acoustic signal anti-interference optimization module, which uses an adaptive bandpass filter to filter environmental interference signals, identifies interference frequency ranges through signal power spectrum analysis, adjusts filtering parameters, optimizes the coding length and modulation method of the M-sequence spread spectrum signal, supports full-scale cross-sectional width measurement, adds an intelligent interference source identification unit, builds an interference signal feature library based on machine learning algorithms, automatically identifies interference source types by analyzing features, formulates differentiated anti-interference strategies for different interference sources, and is equipped with an acoustic signal transmission power dynamic adjustment unit to adjust the transmission power in real time according to the river cross-sectional width and water turbidity.

[0011] Furthermore, it also includes a traffic data anomaly detection and correction module, which has a built-in anomaly data identification model. By analyzing the temporal variation pattern of traffic data and the consistency of acoustic and image features, it identifies the types of abnormal data. It uses an interpolation algorithm based on historical data and a multimodal feature backtracking verification method to correct the abnormal data. The correction record is synchronously stored in the data storage module. A multimodal cross-validation unit is added to perform bidirectional verification by utilizing the complementarity of acoustic data and image data. At the same time, an anomaly cause classification model is constructed to automatically distinguish the causes of anomalies. An anomaly data hierarchical processing mechanism is established, and a correction strategy or alarm is automatically selected according to the severity of the anomaly.

[0012] Furthermore, it also includes a low-power intelligent control module, which adopts a dual power supply mode of solar power and battery backup. It has a built-in power consumption monitoring unit and intelligent sleep mechanism, which controls non-core modules to enter a low-power sleep state during data acquisition intervals. It dynamically adjusts the sampling frequency and data transmission interval according to the priority of monitoring tasks. It also adds a power dynamic allocation unit to allocate power resources according to the importance of modules and real-time task requirements. At the same time, it is equipped with a solar power prediction unit, which combines local weather forecast data and historical solar power supply records to predict the solar power supply capacity in the future and adjust the system operating parameters in advance.

[0013] Furthermore, it also includes a self-diagnostic module for equipment faults, which monitors the working status of core equipment in real time, identifies equipment fault types by detecting signal transmission power, image acquisition success rate, and communication link stability, generates fault diagnosis reports, and pushes them to the control center via remote communication, marking the location of the faulty equipment and maintenance suggestions.

[0014] Furthermore, it also includes a historical data mining and trend prediction module. Based on long-term stored monitoring data, it uses time series analysis algorithms to mine the regular characteristics of flow, and combines meteorological forecast data and watershed hydrological data to build a flow trend prediction model. The prediction results are displayed through a visual interface. It integrates multi-source data into the prediction model, and uses deep learning algorithms to mine the nonlinear correlation between various factors and river flow. At the same time, it adds a personalized prediction configuration unit to support users to customize the prediction time scale and accuracy requirements, generate customized prediction reports, and optimize the model's early warning threshold by combining historical abnormal flow event data.

[0015] Furthermore, it includes a multi-platform data interaction interface module, reserving standardized data communication interfaces to support integration with third-party platforms. It adopts a unified data format to achieve real-time sharing of data reports and supports receiving control commands from third-party platforms. It also adds a data encryption transmission unit, using national cryptographic symmetric encryption algorithms to encrypt transmitted data. In addition, it is equipped with an interface adaptive adaptation unit to automatically identify the data protocols and format requirements of third-party platforms. Through protocol conversion and format adaptation technology, it achieves seamless integration, supports both batch data synchronization and real-time push transmission modes, and records data interaction logs.

[0016] Compared with existing technologies, the beneficial effects of this invention are: Regarding data acquisition quality, the system achieves accurate acquisition of acoustic data across the entire river cross-section through an acoustic tomography monitoring module. Relying on BeiDou / GPS timing to ensure node clock synchronization provides a reliable acoustic foundation for flow calculation. The image acquisition and AI enhancement module effectively improves image quality issues caused by environmental interference through adaptive denoising, super-resolution reconstruction, and edge enhancement technologies, outputting clear and stable enhanced images. It strengthens key features such as water surface velocity texture and riverbank boundaries, enabling visual data to accurately assist in flow calculation. The high-quality acquisition of dual-modal data lays a solid foundation for subsequent fusion analysis, significantly improving the overall accuracy of flow monitoring.

[0017] In multimodal data processing, the multimodal data synchronization control module achieves precise alignment between acoustic and visual data, ensuring consistency over time. The multimodal feature fusion weight dynamic adjustment module flexibly allocates weights based on real-time data quality, enhancing the contribution of high-quality data and avoiding interference from low-quality data, making the fusion results more realistic. The AI ​​intelligent traffic calculation module combines deep learning algorithms to fit the mapping relationship between multimodal features and traffic, automatically correcting calculation deviations caused by various interference factors, further improving the accuracy and reliability of traffic calculation.

[0018] In terms of system adaptability and stability, the acoustic signal anti-interference optimization module effectively resists various environmental interferences through intelligent interference source identification and differentiated anti-interference strategies, ensuring signal transmission quality in complex river environments. The low-power intelligent control module adopts dual power supply modes and an intelligent sleep mechanism, adapting to the long-term monitoring needs of remote river areas and extending equipment runtime. The equipment fault self-diagnosis module can monitor the status of core equipment in real time, quickly identify fault types, and push maintenance suggestions, reducing monitoring downtime and improving system operational stability.

[0019] In terms of practical value and scalability, the dynamic correction module for river cross-section geometric parameters automatically corrects deviations in cross-section parameters, providing a precise geometric basis for flow calculation. The historical data mining and trend prediction module can uncover patterns in flow changes and output short- and medium-term forecasts, providing scientific support for water resource allocation and flood warning. The multi-platform data interaction interface module supports seamless integration with third-party platforms, enabling real-time data sharing and cross-system collaborative scheduling, while ensuring information security through encrypted data transmission. Overall, this system integrates the advantages of multiple technologies, effectively reducing monitoring costs, enhancing data utilization value, and adapting to different river environments and application scenarios, providing efficient and reliable technical support for water resource management. Attached Figure Description

[0020] Figure 1 This is a schematic block diagram of the AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system proposed in this invention; Figure 2 Line graph comparing the signal-to-noise ratio of acoustic signals under different interference intensities; Figure 3 A bar chart comparing system power consumption under different river environments; Figure 4 A horizontal bar chart comparing the correction errors of cross-sectional geometric parameters. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0023] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0024] Reference Figures 1 to 4 A multimodal river flow monitoring system based on AI image enhancement acoustic tomography includes the following modules: The acoustic tomography monitoring module consists of at least four acoustic station nodes deployed on both banks of the river cross section. Each acoustic station node integrates an omnidirectional underwater acoustic transducer, a signal transmitting unit, a signal receiving unit, and a demodulation module. It uses M-sequence spread spectrum modulation technology to transmit coded acoustic signals and uses a matched filtering algorithm to accurately measure the acoustic wave transmission time and propagation speed between nodes. It synchronously collects acoustic propagation data of the entire river cross section. All nodes achieve clock synchronization through BeiDou / GPS time synchronization, and the synchronization error is controlled within 100ns. The image acquisition and AI enhancement module is equipped with a high-definition network camera and an image preprocessing unit at key observation points of the river cross section. It acquires visual image data of the river cross section in real time, removes image noise caused by water flow disturbance and changes in ambient light through an adaptive denoising algorithm based on residual network, improves image spatial resolution by using super-resolution reconstruction technology, and enhances key features such as water surface velocity texture and riverbank boundary by using Laplacian edge enhancement algorithm, outputting high-definition and high-stability enhanced images. The multimodal data synchronization control module has a built-in millisecond-level timestamp synchronization unit and data alignment engine. It receives acoustic data from the acoustic tomography monitoring module and visual data from the image acquisition module. It marks the two types of data with precise timestamps through Beidou / GPS timing signals. Based on timestamp matching, it achieves millisecond-level alignment of acoustic data and image data, ensuring high consistency of multi-source data in the time dimension. The multi-source data fusion module uses a cross-modal attention mechanism to extract flow velocity propagation features from acoustic data and visual texture features from enhanced images, constructs a multi-dimensional feature vector containing acoustic parameters, image features, and environmental parameters, eliminates the dimensional differences between different modal data through feature normalization processing, and transmits the fused feature vector to the AI ​​calculation module to provide comprehensive and complementary data support for flow calculation. The AI ​​intelligent flow calculation module has a built-in pre-trained flow prediction model based on deep learning. It takes the fused multimodal feature vector as input, combines the geometric parameters of the river cross section and historical flow data, and fits the mapping relationship between acoustic propagation speed, image flow velocity texture and actual flow through deep neural network algorithm to accurately calculate the instantaneous flow and average flow of the river cross section, and automatically corrects the calculation deviation caused by water flow turbulence and cross section morphology changes. The data storage and visualization module adopts a dual storage architecture of edge storage and cloud backup. The edge storage stores recently collected raw data, enhanced images and calculation results, while the cloud database stores complete monitoring data and historical records for a long time. It is equipped with a visualization interface to display traffic data, acoustic propagation curves, enhanced images and traffic change trends in real time, and supports multi-dimensional data retrieval and standardized format export. The remote control and parameter configuration module enables bidirectional communication with the remote control center via an IoT communication interface. It supports remote adjustment of the transmission power, sampling frequency, and image acquisition frame rate and resolution of the sound station nodes. It receives remote control commands to update system parameters, query equipment status, and handle fault alarms, adapting to the monitoring needs of different river environments.

[0025] This invention also includes an image enhancement quality quantification and evaluation module, which constructs a multi-dimensional image quality evaluation model. This model quantifies the usability of the enhanced image through comprehensive indicators, providing a quality judgment basis for subsequent visual feature extraction. The expression is as follows: ; in A comprehensive score for image enhancement quality. This is the texture sharpness weighting coefficient. To enhance the texture sharpness value of the post-image, The texture sharpness threshold for a standard sharp image. This is the contrast weighting coefficient. To enhance the contrast of the resulting image, The contrast threshold for a standard image. This is the signal-to-noise ratio weighting coefficient. To enhance the signal-to-noise ratio of the resulting image, To preset the signal-to-noise ratio threshold, The multi-dimensional image quality assessment model quantifies the usability of enhanced images through comprehensive indicators, providing a quality judgment basis for subsequent visual feature extraction. Only enhanced images that meet the set standards can participate in multimodal data fusion. It is also equipped with an AI adaptive enhancement parameter adjustment unit, which dynamically optimizes the denoising intensity, super-resolution reconstruction iteration number and edge enhancement coefficient based on real-time river environment monitoring data. It increases the weight of the denoising algorithm for highly turbid rivers, strengthens image brightness compensation and contrast optimization for low-light environments, and adds polarization filtering algorithm adaptation for strong reflective scenes, ensuring that high-quality enhanced images can be output in different environments.

[0026] This invention also includes a multimodal feature fusion weight dynamic adjustment module, which adaptively allocates fusion weights based on the real-time quality status of acoustic and image data, thereby enhancing the contribution ratio of high-quality data. The expression is as follows: ; in For the fusion weights of acoustic data, For the fusion weights of visual image data, Score the credibility of acoustic data. For the signal integrity of acoustic data, This is the acoustic weighting adjustment coefficient. For the feature integrity of image data, The visual weight adjustment coefficient and dynamic weight allocation mechanism can flexibly adjust the fusion ratio according to the real-time quality of the two types of data. When the quality of a certain type of data declines, its weight ratio is automatically reduced to avoid interference from low-quality data on the fusion result. A feature quality real-time detection unit is added. Through a triple mechanism of acoustic signal-to-noise ratio analysis, image texture clarity detection, and feature dimension integrity verification, the quality level of the two types of data is determined in real time. At the same time, a weight adjustment buffer mechanism is established and a weight change rate threshold is set to avoid frequent weight switching caused by short-term data quality fluctuations, thus ensuring the stability and continuity of the fusion result.

[0027] This invention also includes a dynamic correction module for river cross-section geometric parameters. By enhancing the image, it identifies key geometric features such as the riverbank and riverbed boundary. Combined with preset cross-section benchmark parameters, it uses image measurement algorithms to calculate geometric parameters such as cross-section width and average water depth in real time. This automatically corrects deviations in cross-section parameters caused by water level changes and riverbed siltation, providing accurate geometric basis data for flow calculation. The corrected geometric parameters are synchronized in real time to the AI ​​intelligent flow calculation module. The module also integrates a multi-view image fusion unit, which uses multiple high-definition cameras deployed at different locations along the river cross-section to collect multi-view images. A stereo vision algorithm is used to generate a 3D point cloud model of the cross-section, accurately reproducing the topographic undulations. Simultaneously, a historical cross-section data trend correction mechanism is introduced. By analyzing the long-term accumulated patterns of cross-section geometric parameter changes, it filters correction errors caused by short-term abnormal disturbances. For scenarios such as seasonal siltation and scouring, correction compensation coefficients are preset in advance. Combined with a hydrodynamic model, the parameter calculation logic is optimized to further improve the accuracy and stability of geometric parameter calculation.

[0028] This invention also includes an acoustic signal anti-interference optimization module, which uses an adaptive bandpass filter to filter environmental interference signals such as water flow noise and shipping interference. It identifies interference frequency ranges through signal power spectrum analysis, dynamically adjusts filtering parameters, and optimizes the coding length and modulation method of the M-sequence spread spectrum signal to improve the anti-interference capability and propagation distance of the acoustic signal. This ensures the accuracy of acoustic data acquisition in complex river environments, supports full-scale cross-sectional width measurement, and adds an intelligent interference source identification unit. Based on machine learning algorithms, it constructs an interference signal feature library containing the spectral and temporal characteristics of typical interferences such as shipping noise, water flow turbulence noise, and electronic equipment interference. By analyzing the frequency distribution, amplitude changes, and duration of the acoustic signal, it automatically identifies the type of interference source and formulates differentiated anti-interference strategies for different interference sources. It employs frequency avoidance technology for narrowband interference, enhances spread spectrum coding gain for broadband interference, and uses signal reconstruction and repair for transient pulse interference. Simultaneously, it is equipped with an acoustic signal transmission power dynamic adjustment unit that adjusts the transmission power in real time according to the river cross-sectional width and water turbidity, reducing energy consumption while ensuring signal transmission quality.

[0029] This invention also includes a flow data anomaly detection and correction module, which incorporates an anomaly data identification model. By analyzing the temporal variation patterns of flow data and the consistency of acoustic and image features, it identifies anomalous data types such as sudden changes and continuous deviations. It employs interpolation algorithms based on historical data and multimodal feature backtracking verification to correct and repair the anomalous data, ensuring the continuity and reliability of flow monitoring data. Correction records are synchronously stored in the data storage module. A multimodal cross-validation unit is added, utilizing the complementarity of acoustic and image data for bidirectional verification. When a suspected anomaly appears in one modality of data, the validity of the anomaly is determined by the rationality of another modality. Simultaneously, an anomaly cause classification model is constructed to automatically distinguish between anomalies caused by equipment failure, sudden environmental changes, and natural fluctuations in water flow. For equipment failures, a backup data model is activated for alternative calculations; for sudden environmental changes, real-time neighborhood data weighted correction is used; for natural fluctuations in water flow, data features are preserved and the cause of the fluctuation is labeled. An anomaly data hierarchical processing mechanism is established, automatically selecting a correction strategy or triggering an alarm based on the severity of the anomaly.

[0030] This invention also includes a low-power intelligent control module, which adopts a dual power supply mode of solar power and battery backup. It has a built-in power consumption monitoring unit and intelligent sleep mechanism, which controls non-core modules to enter a low-power sleep state during data acquisition intervals. It dynamically adjusts the sampling frequency and data transmission interval according to the priority of monitoring tasks to reduce the overall power consumption of the system, adapting to the long-term uninterrupted monitoring needs of remote river areas and extending the equipment's battery life. It also adds a power dynamic allocation unit to allocate power resources according to the importance of modules and real-time task requirements. Core modules such as acoustic tomography monitoring module and AI calculation module are given priority to ensure stable power supply, while non-core modules such as image acquisition module and data transmission module adopt an on-demand power supply strategy. At the same time, it is equipped with a solar power prediction unit, which combines local weather forecast data and historical solar power supply records to predict the solar power supply capacity in the future period and adjust the system operating parameters in advance. On sunny days, the sampling frequency and data transmission frequency are increased, and on cloudy days or at night, unnecessary power consumption is automatically reduced. A battery power threshold early warning mechanism is set to avoid monitoring interruption due to insufficient power supply.

[0031] This invention also includes a device fault self-diagnosis module, which monitors the working status of core devices such as sound station nodes, cameras, and communication modules in real time. By detecting key indicators such as signal transmission power, image acquisition success rate, and communication link stability, it identifies the type of device fault, generates a fault diagnosis report, and pushes it to the control center via remote communication, marking the location of the faulty device and maintenance suggestions, thereby improving system maintenance efficiency and reducing monitoring interruption time.

[0032] This invention also includes a historical data mining and trend prediction module. Based on long-term stored monitoring data, it uses time series analysis algorithms to mine the daily variations and seasonal fluctuations of flow, and combines meteorological forecast data and watershed hydrological data to construct a flow trend prediction model. It outputs short-term and medium-term flow prediction results, providing scientific decision support for water resource allocation and flood warning. The prediction results can be displayed through a visual interface. The module integrates a multi-factor correlation analysis unit, incorporating multi-source data such as watershed precipitation, temperature, vegetation cover, and reservoir operation into the prediction model. Deep learning algorithms are used to mine the nonlinear correlation between various factors and river flow, improving prediction accuracy. A personalized prediction configuration unit is also added to optimize the short-term prediction model for flood warning scenarios, improving the accuracy of hourly flow change predictions. For water resource allocation scenarios, it strengthens medium-term prediction capabilities, outputting weekly and monthly flow trends. Users can customize the prediction time scale and accuracy requirements, generating customized prediction reports. The module also optimizes the model's warning threshold by combining historical abnormal flow event data.

[0033] This invention also includes a multi-platform data interaction interface module, reserving standardized data communication interfaces to support integration with third-party platforms such as river basin hydrological monitoring centers, water resources management platforms, and emergency command systems. It uses a unified data format to achieve real-time sharing of monitoring data, image data, and flow reports. Simultaneously, it supports receiving control commands from third-party platforms, enabling cross-system collaborative monitoring and scheduling, and enhancing the comprehensive utilization value of monitoring data. A data encryption transmission unit is added, employing national cryptographic symmetric encryption algorithms to encrypt transmitted data, combined with digital signature technology to ensure data integrity and authentication, preventing data tampering and theft during transmission. It also features an interface adaptive adaptation unit that automatically identifies the data protocols and format requirements of third-party platforms, achieving seamless integration through protocol conversion and format adaptation technologies without requiring manual configuration of interface parameters. It supports both batch data synchronization and real-time push transmission modes to meet the data reception needs of different platforms, while recording data interaction logs for easy tracing of data transmission processes and troubleshooting.

[0034] The following two examples further illustrate specific embodiments of the present invention: Example 1: Monitoring of Navigation Flow in Wide Cross-Sectional Navigation Canals in Plains This embodiment addresses the flow monitoring needs of a wide-section, busy river in a plain area. The river has a large cross-section and a gentle flow, but frequent shipping activity. The water quality is slightly turbid due to shipping disturbances, and the environmental noise is mainly shipping noise and water flow turbulence noise. It is necessary to achieve high-precision, real-time, and interference-resistant flow monitoring, while also being adaptable to long-term uninterrupted operation.

[0035] The acoustic tomography monitoring module deploys eight acoustic station nodes, evenly distributed along both banks of the river cross-section, with four nodes symmetrically arranged on each bank to ensure sound wave coverage of the entire river cross-section. Each acoustic station node integrates an omnidirectional underwater acoustic transducer, a signal transmitting unit, a signal receiving unit, and a demodulation module. It uses M-sequence spread spectrum modulation technology to transmit coded acoustic signals and immediately switches to receiving mode after transmission. All nodes are synchronized via BeiDou / GPS time synchronization, with synchronization errors controlled within 100ns. Matched filtering algorithms are used to accurately measure the sound wave transmission time and propagation speed between nodes, synchronously collecting acoustic propagation data across the entire river cross-section. The data sampling frequency is dynamically adjusted according to the level of shipping activity, with the sampling frequency increased during peak shipping periods to ensure data continuity.

[0036] The image acquisition and AI enhancement module deploys two high-definition network cameras at key observation points on both banks of the river cross-section. The camera installation height is adapted to the river width to ensure complete acquisition of cross-sectional visual images. Real-time image data is transmitted to the image preprocessing unit, where an adaptive denoising algorithm based on residual networks removes image noise caused by water flow disturbances, water splashes from shipping, and changes in ambient lighting. For sunny, highly reflective scenes, a polarization filtering algorithm is added to improve image detail recognition. Super-resolution reconstruction technology is employed to enhance image spatial resolution, and a Laplacian edge enhancement algorithm is used to strengthen key features such as water surface velocity texture and riverbank boundaries, outputting high-definition and highly stable enhanced images. The image enhancement quality quantification and evaluation module scores the enhanced images across multiple dimensions; only images meeting the set standards are allowed to participate in subsequent multimodal data fusion.

[0037] The multimodal data synchronization control module incorporates a millisecond-level timestamp synchronization unit and a data alignment engine. It receives acoustic data from the acoustic tomography monitoring module and visual data from the image acquisition module, and uses BeiDou / GPS timing signals to accurately timestamp both types of data. Based on timestamp matching, it achieves millisecond-level alignment between acoustic and image data, ensuring high consistency of multi-source data in the time dimension and avoiding distortion of fusion results due to data asynchrony.

[0038] The multi-source data fusion module employs a cross-modal attention mechanism to extract flow velocity propagation features from acoustic data and visual texture features from enhanced images, constructing a multi-dimensional feature vector encompassing acoustic parameters, image features, and environmental parameters. Feature normalization eliminates dimensional differences between different modalities, and the multi-modal feature fusion weight dynamic adjustment module adaptively allocates fusion weights based on the real-time quality status of the two types of data. When shipping noise degrades acoustic data quality, the acoustic data weight is automatically reduced, while the visual data weight is increased. Simultaneously, a weight adjustment buffer mechanism prevents frequent weight switching, ensuring the stability and continuity of the fusion results. The fused feature vector is then transmitted to the AI ​​calculation module, providing comprehensive and complementary data support for flow calculation.

[0039] The AI-powered intelligent flow calculation module incorporates a pre-trained flow prediction model based on deep learning. It takes a fused multimodal feature vector as input, combines river cross-section geometric parameters and historical flow data, and uses a deep neural network algorithm to fit the mapping relationship between acoustic propagation velocity, image flow velocity texture, and actual flow. The dynamic correction module for river cross-section geometric parameters identifies key geometric features such as the riverbank and riverbed boundary through enhanced images, combines them with preset cross-section benchmark parameters, and uses image measurement algorithms to calculate geometric parameters such as cross-section width and average water depth in real time. An integrated multi-view image fusion unit generates a 3D point cloud model of the cross-section from multi-view images collected by multiple cameras, accurately reproducing the cross-section topographic undulation features and automatically correcting cross-section parameter deviations caused by water level changes. The corrected geometric parameters are synchronized to the AI-powered intelligent flow calculation module in real time, accurately calculating the instantaneous and average flow of the river cross-section and automatically correcting calculation errors caused by water flow turbulence.

[0040] The acoustic signal anti-interference optimization module employs an adaptive bandpass filter to filter environmental interference signals such as shipping noise and water flow turbulence noise. It identifies interference frequency ranges through signal power spectrum analysis and dynamically adjusts filtering parameters. The added intelligent interference source identification unit constructs an interference signal feature library based on machine learning algorithms, automatically identifying interference source types such as shipping noise and water flow turbulence noise. It uses frequency avoidance technology for narrowband interference generated by shipping and enhances spread spectrum coding gain for broadband interference generated by water flow turbulence. Simultaneously, it is equipped with an acoustic signal transmission power dynamic adjustment unit, which adjusts the transmission power in real time according to the river cross-section width and water turbidity, reducing energy consumption while ensuring signal transmission quality.

[0041] The data storage and visualization module adopts a dual-storage architecture of edge storage and cloud backup. The edge storage stores raw data, enhanced images, and calculation results collected within the past week, while the cloud database stores complete monitoring data and historical records long-term. It features a visualization interface that displays traffic data, acoustic propagation curves, enhanced images, and traffic change trends in real time. It supports multi-dimensional searching by time range and data type and can export monitoring reports in standardized formats.

[0042] The remote control and parameter configuration module achieves bidirectional communication with the remote control center via an IoT communication interface, supporting remote adjustment of the transmission power, sampling frequency, and image acquisition frame rate and resolution of the sound station nodes. It receives remote control commands to update system parameters, query equipment status, and handle fault alarms, adapting to the complex monitoring needs of the navigation canal. The low-power intelligent control module adopts a dual power supply mode of solar power and battery backup, with a built-in power consumption monitoring unit and intelligent sleep mechanism. During data acquisition intervals, it controls non-core modules to enter a low-power sleep state and dynamically adjusts the sampling frequency and data transmission interval according to the priority of monitoring tasks, extending the equipment's battery life.

[0043] Table 1: Comparison of Monitoring Results of Navigation and Canal Flow in Wide Cross-Sectional Plains Evaluation indicators Traditional single-point monitoring methods This invention system monitoring Monitoring coverage Local cross-sectional area Full cross-section and full area coverage Anti-interference capability Susceptible to shipping noise Strong resistance to complex interference Data synchronization accuracy Large timestamp deviation Millisecond-level precise synchronization Flow calculation accuracy Large error High calculation accuracy Maintenance costs High human resource investment Low maintenance costs Data continuity Data interruption is likely. Long-term continuous stability Table 1 clearly demonstrates the significant advantages of the system of this invention in monitoring navigation and canal flow in broad cross-sections of plains. Traditional single-point monitoring methods can only cover local cross-sectional areas, are susceptible to navigation noise interference, have low data synchronization accuracy, large flow calculation errors, and require frequent manual maintenance, making it difficult to guarantee data continuity. The system of this invention combines full-section acoustic tomography monitoring with AI image enhancement technology to achieve full-section, full-area coverage monitoring, has strong anti-complex interference capabilities, and provides millisecond-level accurate synchronization of multi-modal data, significantly improving flow calculation accuracy. Low-power design and fault self-diagnosis function reduce maintenance costs, ensure long-term continuous and stable operation, perfectly adapt to the complex monitoring needs of navigation and canal flow in plains, and provide reliable data support for water resource management.

[0044] Example 2: Flow monitoring of narrow-section turbulent rivers in mountainous areas This embodiment addresses the flow monitoring needs of narrow-section, turbulent rivers in mountainous areas. These rivers have narrow cross-sections, fast and turbulent water flow, significant undulations in the riverbed topography, seasonal siltation and scouring, and marked changes in ambient light. Power supply conditions are also limited in some areas, necessitating high-precision, interference-resistant, and low-power flow monitoring.

[0045] The acoustic tomography monitoring module deploys four acoustic station nodes, two symmetrically arranged on each bank, to meet the monitoring needs of narrow-section rivers. Each acoustic station node integrates an omnidirectional underwater acoustic transducer, a signal transmitting unit, a signal receiving unit, and a demodulation module. It uses M-sequence spread spectrum modulation technology to transmit coded acoustic signals and employs a matched filtering algorithm to accurately measure the acoustic wave transmission time and propagation speed between nodes, simultaneously acquiring acoustic propagation data across the entire river cross-section. All nodes achieve clock synchronization via BeiDou / GPS time synchronization, with synchronization errors controlled within 100ns. To address signal propagation instability caused by turbulent water flow, the signal transmission power and sampling frequency are appropriately increased to ensure the quality of acoustic data acquisition.

[0046] The image acquisition and AI enhancement module is equipped with high-definition network cameras at key observation points upstream and downstream of the river cross-section. These cameras feature a waterproof and dustproof design, making them suitable for the complex mountainous environment. Real-time acquired image data is transmitted to the image preprocessing unit, where an adaptive denoising algorithm based on residual networks removes image noise caused by severe water flow disturbances, splashing water, and changes in ambient lighting. For the low-light environment in mountainous areas, image brightness compensation and contrast optimization are enhanced. Super-resolution reconstruction technology is used to improve image spatial resolution, and the Laplacian edge enhancement algorithm is employed to enhance key features such as water surface velocity texture, riverbank boundaries, and riverbed topography, outputting high-definition and highly stable enhanced images. The image enhancement quality quantification and evaluation module dynamically optimizes denoising intensity, super-resolution reconstruction iterations, and edge enhancement coefficients to ensure high-quality enhanced images are output under different lighting conditions.

[0047] The multimodal data synchronization control module has a built-in millisecond-level timestamp synchronization unit and data alignment engine. It receives acoustic data from the acoustic tomography monitoring module and visual data from the image acquisition module. It marks the two types of data with precise timestamps through Beidou / GPS timing signals. Based on timestamp matching, it achieves millisecond-level alignment of acoustic data and image data, ensuring high consistency of multi-source data in the time dimension and laying the foundation for accurate fusion.

[0048] The multi-source data fusion module employs a cross-modal attention mechanism to extract flow velocity propagation features from acoustic data and visual texture features from enhanced images, constructing a multi-dimensional feature vector encompassing acoustic parameters, image features, and environmental parameters. Feature normalization eliminates dimensional differences between different modalities, and the multi-modal feature fusion weight dynamic adjustment module adaptively allocates fusion weights based on the real-time quality status of the two types of data. When turbulence in the water flow degrades the integrity of the acoustic data signal, the acoustic data weight is automatically reduced, while the visual data weight is increased. A weight adjustment buffer mechanism prevents frequent weight switching due to short-term data quality fluctuations, ensuring the stability and continuity of the fusion results. The fused feature vector is then transmitted to the AI ​​calculation module, providing comprehensive and complementary data support for flow calculation.

[0049] The AI-powered intelligent flow calculation module incorporates a pre-trained flow prediction model based on deep learning. It takes a fused multimodal feature vector as input, combines river cross-section geometric parameters and historical flow data, and uses a deep neural network algorithm to fit the mapping relationship between acoustic propagation speed, image flow velocity texture, and actual flow. The dynamic correction module for river cross-section geometric parameters identifies key geometric features such as the riverbank and riverbed boundary through enhanced images. Combined with preset cross-section benchmark parameters, it uses image measurement algorithms to calculate geometric parameters such as cross-section width and average water depth in real time. An integrated multi-view image fusion unit generates a 3D point cloud model of the cross-section from multi-view images collected by multiple cameras, accurately reproducing the cross-section topographic undulation features. It introduces a historical cross-section data trend correction mechanism, analyzes the long-term accumulated variation patterns of cross-section geometric parameters, and pre-sets correction compensation coefficients for seasonal siltation and scour scenarios to automatically correct cross-section parameter deviations. The corrected geometric parameters are synchronized to the AI-powered intelligent flow calculation module in real time, accurately calculating the instantaneous and average flow of the river cross-section and automatically correcting calculation deviations caused by water turbulence and cross-section morphology changes.

[0050] The acoustic signal anti-interference optimization module employs an adaptive bandpass filter to filter interference signals such as water flow turbulence noise and environmental noise. It identifies interference frequency ranges through signal power spectrum analysis and dynamically adjusts filtering parameters. An additional intelligent interference source identification unit automatically identifies interference source types such as water flow turbulence noise and electronic equipment interference. It enhances the spread spectrum coding gain for broadband interference caused by water flow turbulence and uses signal reconstruction to repair transient pulse interference. A dynamic acoustic signal transmission power adjustment unit is also included, adjusting the transmission power in real time based on the river cross-section width and water turbidity to reduce energy consumption while ensuring signal transmission quality.

[0051] The data storage and visualization module adopts a dual-storage architecture of edge storage and cloud backup. The edge storage stores raw data, enhanced images, and calculation results collected within the past week, while the cloud database provides long-term storage of complete monitoring data and historical records. It features a visualization interface that displays real-time traffic data, acoustic propagation curves, enhanced images, and traffic change trends, supporting multi-dimensional retrieval and standardized format export.

[0052] The remote control and parameter configuration module achieves two-way communication with the remote control center via an IoT communication interface. It supports remote adjustment of the operating parameters of sound station nodes and image acquisition equipment, and receives remote control commands to update system parameters, query equipment status, and handle fault alarms. The low-power intelligent control module adopts a dual power supply mode of solar power and battery backup, and adds a dynamic power allocation unit to prioritize stable power supply to core modules, while non-core modules use an on-demand power supply strategy. Equipped with a solar power prediction unit, it combines local weather forecast data and historical solar power supply records to adjust system operating parameters in advance, avoiding monitoring interruptions due to insufficient power supply, thus adapting to the long-term uninterrupted monitoring needs of remote mountainous areas.

[0053] Table 2: Comparison of Monitoring Results for Narrow-Section Turbulent Rivers in Mountainous Areas Evaluation indicators Traditional in-situ monitoring methods This invention system monitoring Cross-sectional adaptability Difficult to adapt to complex terrain Adaptable to complex cross-sectional terrain Resistance to water flow interference Weak anti-interference ability Strong resistance to water flow interference Geometric parameter accuracy Susceptible to siltation and erosion High parameter calibration accuracy Low power consumption High power consumption and short battery life Low power consumption and long battery life Data reliability The data fluctuates significantly. Stable and reliable data Environmental adaptability Poor adaptability Strong environmental adaptability Table 2 data fully demonstrates the technical advantages of the system of this invention in monitoring narrow-section, turbulent rivers in mountainous areas. Traditional in-situ monitoring methods are difficult to adapt to the complex terrain of mountain rivers, have weak resistance to water flow interference, are easily affected by siltation and erosion of cross-sectional geometric parameters, have high power consumption leading to short endurance, large data fluctuations, and poor environmental adaptability. The system of this invention, through multi-view image fusion and dynamic correction of cross-sectional parameters, accurately adapts to complex cross-sectional terrain, has strong resistance to water flow interference, and high accuracy in geometric parameter correction. The low-power intelligent control design enables long-endurance operation, multi-modal data fusion and AI calculation ensure stable and reliable data, and strong environmental adaptability perfectly matches the complex monitoring scenarios of mountain rivers, effectively solving many pain points of traditional monitoring methods and providing high-precision, continuous, and stable monitoring data for water resource management in mountainous areas.

[0054] Reference Figure 2 The line graph clearly highlights the technological breakthrough of this invention's system in acoustic signal anti-interference performance. Traditional acoustic tomography systems lack targeted interference identification and processing mechanisms. As interference intensity increases, the signal-to-noise ratio (SNR) drops rapidly, reaching only 35.8 dB in complex, high-interference environments, leading to acoustic data distortion and affecting the accuracy of flow rate calculation. This invention's acoustic signal anti-interference optimization module effectively resists various environmental interferences through intelligent interference source identification, differentiated anti-interference strategies, and dynamic power adjustment. Even under level 5 complex high-interference conditions, the SNR remains above 51.6 dB, ensuring stable signal quality. Stable acoustic data provides a reliable foundation for multimodal fusion and flow rate calculation, enabling the system to accurately monitor in complex environments such as busy shipping lanes and turbulent waters, breaking through the anti-interference bottleneck of traditional systems.

[0055] Reference Figure 3 The bar chart visually demonstrates the significant advantages of the present invention's system in low-power design. Traditional monitoring systems lack intelligent power management mechanisms, resulting in continuous high-load operation of the core module. Regardless of the river environment, power consumption remains above 37W. In remote, unpowered areas relying on battery power, the battery life is short, making it difficult to meet long-term monitoring needs. The low-power intelligent control module of this invention adopts a dual-power supply mode of solar energy and battery, combined with an intelligent sleep mechanism and dynamic power allocation strategy, dynamically adjusting operating parameters according to environmental and task requirements. In various river environments, power consumption is controlled below 19W, and as low as 15.8W in remote, unpowered river scenarios. The low-power design significantly extends the equipment's battery life, reduces dependence on external power supply, perfectly adapts to the long-term uninterrupted monitoring needs of remote river areas, and reduces operation and maintenance costs.

[0056] Reference Figure 4The horizontal bar chart fully demonstrates the technical advantages of this invention's system in terms of the accuracy of cross-sectional geometric parameter correction. Traditional monitoring systems often use fixed cross-sectional parameters or manual periodic calibration, which is difficult to cope with dynamic factors such as water level changes, siltation, and erosion. The correction error is generally above 3.8%, and as high as 6.2% in complex river scenarios, leading to significant deviations in flow calculation. The river cross-sectional geometric parameter dynamic correction module of this invention generates a three-dimensional point cloud model through multi-view image fusion and combines it with a historical data trend correction mechanism to automatically adapt to dynamic changes in the cross-section. The correction error is controlled within 1.1% in various river scenarios, accurately restoring the cross-sectional topography and geometric parameters, providing high-precision basic data for flow calculation. Precise geometric parameter correction reduces flow calculation errors from the source, making the monitoring results more consistent with the actual river conditions and improving the scientific validity and practicality of the data.

[0057] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An AI-enhanced image-based acoustic tomography multimodal river flow monitoring system, characterized in that, Includes the following modules: The acoustic tomography monitoring module consists of at least four acoustic station nodes deployed on both banks of the river cross section. Each acoustic station node integrates an omnidirectional underwater acoustic transducer, a signal transmitting unit, a signal receiving unit, and a demodulation module. It uses M-sequence spread spectrum modulation and matched filtering algorithm to measure the sound wave transmission time and velocity, and simultaneously collects acoustic propagation data of the entire river cross section. The image acquisition and AI enhancement module is equipped with a high-definition network camera and image preprocessing unit at the river cross section to acquire visual image data. Through adaptive denoising algorithm, super-resolution reconstruction technology and edge enhancement algorithm, the image is optimized and the enhanced image is output. The multimodal data synchronization control module has a built-in millisecond-level timestamp synchronization unit and data alignment engine. It receives acoustic and visual data and timestamps the two types of data through Beidou / GPS timing signals. The multi-source data fusion module uses a cross-modal attention mechanism to extract acoustic and visual features, constructs a multi-dimensional feature vector containing acoustic parameters, image features, and environmental parameters, performs normalization processing, and transmits the fused feature vector to the AI ​​solution module. The AI ​​intelligent flow calculation module has a built-in pre-trained flow prediction model based on deep learning. It takes multimodal feature vectors as input and combines them with river cross-section data. It uses a deep neural network algorithm to fit the mapping relationship between acoustic propagation speed, image flow velocity texture and actual flow, and calculates instantaneous flow and average flow. The data storage and visualization module stores recently collected raw data, enhanced images, and calculation results at the edge, while the cloud database stores monitoring data and historical records long-term. It also features a visualization interface to display traffic data, acoustic propagation curves, enhanced images, and traffic change trends. The remote control and parameter configuration module enables bidirectional communication with the remote control center through an IoT communication interface, receiving remote control commands to complete parameter updates, status queries, and fault alarm processing.

2. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes an image enhancement quality quantification evaluation module, which constructs a multi-dimensional image quality evaluation model and quantifies the usability of the enhanced image through comprehensive indicators, expressed as follows: ； in A comprehensive score for image enhancement quality. This is the texture sharpness weighting coefficient. To enhance the texture sharpness value of the post-image, The texture sharpness threshold for a standard sharp image. This is the contrast weighting coefficient. To enhance the contrast of the resulting image, The contrast threshold for a standard image. This is the signal-to-noise ratio weighting coefficient. To enhance the signal-to-noise ratio of the resulting image, To preset the signal-to-noise ratio threshold, Only enhanced images that meet the set scoring criteria can participate in multimodal data fusion. It is also equipped with an AI adaptive enhancement parameter adjustment unit to dynamically optimize the denoising intensity, super-resolution reconstruction iteration number, and edge enhancement coefficient based on real-time river environment monitoring data.

3. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes a multimodal feature fusion weight dynamic adjustment module, which adaptively allocates fusion weights based on the real-time quality status of acoustic and image data to enhance the contribution ratio of the data. The expression is as follows: ； in For the fusion weights of acoustic data, For the fusion weights of visual image data, Score the credibility of acoustic data. For the signal integrity of acoustic data, This is the acoustic weighting adjustment coefficient. For the feature integrity of image data, The visual weight adjustment coefficient is used. The dynamic weight allocation mechanism adjusts the fusion ratio according to the real-time quality of the two types of data. When the quality of a certain type of data decreases, its weight ratio is automatically reduced. A feature quality real-time detection unit is added. Through a triple mechanism of acoustic signal-to-noise ratio analysis, image texture clarity detection, and feature dimension integrity verification, the quality level of the two types of data is determined in real time. At the same time, a weight adjustment buffer mechanism is established and a weight change rate threshold is set.

4. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes a dynamic correction module for river cross-section geometric parameters. Combining preset cross-section reference parameters, it uses image measurement algorithms to calculate geometric parameters in real time and automatically corrects deviations in cross-section parameters caused by water level changes and riverbed siltation. The corrected geometric parameters are synchronized to the AI ​​intelligent flow calculation module in real time. It integrates a multi-view image fusion unit, which uses high-definition cameras deployed on the river cross-section to collect multi-view images and uses stereo vision algorithms to generate a three-dimensional point cloud model of the cross-section. At the same time, it introduces a historical cross-section data trend correction mechanism, which filters out correction errors caused by short-term abnormal disturbances by analyzing the long-term accumulated cross-section geometric parameter change patterns.

5. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes an acoustic signal anti-interference optimization module, which uses an adaptive bandpass filter to filter environmental interference signals, identifies interference frequency ranges through signal power spectrum analysis, adjusts filtering parameters, optimizes the coding length and modulation method of the M-sequence spread spectrum signal, supports full-scale cross-sectional width measurement, adds an intelligent interference source identification unit, builds an interference signal feature library based on machine learning algorithms, automatically identifies the type of interference source by analyzing features, formulates differentiated anti-interference strategies for different interference sources, and is equipped with an acoustic signal transmission power dynamic adjustment unit to adjust the transmission power in real time according to the river cross-sectional width and water turbidity.

6. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes a traffic data anomaly detection and correction module, with a built-in anomaly data identification model. By analyzing the temporal variation patterns of traffic data and the consistency of acoustic and image features, it identifies the types of abnormal data. It uses an interpolation algorithm based on historical data and a multimodal feature backtracking verification method to correct the abnormal data. The correction records are synchronously stored in the data storage module. A multimodal cross-validation unit is added to perform bidirectional verification by utilizing the complementarity of acoustic and image data. At the same time, an anomaly cause classification model is constructed to automatically distinguish the causes of anomalies. An anomaly data hierarchical processing mechanism is established, and a correction strategy or alarm is automatically selected based on the severity of the anomaly.

7. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes a low-power intelligent control module, which adopts a dual power supply mode of solar power and battery backup. It has a built-in power consumption monitoring unit and intelligent sleep mechanism, which controls non-core modules to enter a low-power sleep state during data acquisition intervals. It dynamically adjusts the sampling frequency and data transmission interval according to the priority of monitoring tasks. It also adds a power dynamic allocation unit to allocate power resources according to the importance of modules and real-time task requirements. At the same time, it is equipped with a solar power prediction unit, which combines local weather forecast data and historical solar power supply records to predict the solar power supply capacity in the future and adjust the system operating parameters in advance.

8. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes a self-diagnostic module for equipment faults, which monitors the working status of core equipment in real time, identifies equipment fault types by detecting signal transmission power, image acquisition success rate, and communication link stability, generates fault diagnosis reports, and pushes them to the control center via remote communication, marking the location of the faulty equipment and maintenance suggestions.

9. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes a historical data mining and trend prediction module. Based on long-term stored monitoring data, it uses time series analysis algorithms to mine the regular characteristics of flow, and combines meteorological forecast data and watershed hydrological data to build a flow trend prediction model. The prediction results are displayed through a visual interface. It integrates multi-source data into the prediction model, and uses deep learning algorithms to mine the nonlinear correlation between various factors and river flow. At the same time, it adds a personalized prediction configuration unit to support users to customize the prediction time scale and accuracy requirements, generate customized prediction reports, and optimize the model's early warning threshold by combining historical abnormal flow event data.

10. The AI ​​image-enhanced acoustic tomography multimodal river flow monitoring system according to claim 1, characterized in that, It also includes a multi-platform data interaction interface module, which reserves standardized data communication interfaces to support integration with third-party platforms. It uses a unified data format to achieve real-time sharing of data reports and supports receiving control commands from third-party platforms. It adds a data encryption transmission unit, which uses the national cryptographic symmetric encryption algorithm to encrypt transmitted data. It also has an interface adaptive adaptation unit that automatically identifies the data protocol and format requirements of third-party platforms. Through protocol conversion and format adaptation technology, it achieves seamless integration, supports both batch data synchronization and real-time push transmission modes, and records data interaction logs.

Images