Trash rack blockage detection system based on multi-band optical acoustics and differential pressure fusion

Abstract

The invention discloses a trash rack blockage detection system and method based on multi-band optical acoustics and differential pressure fusion. The system comprises a multi-band acoustic imaging module, a distributed differential pressure sensing network and an edge fusion processing system. Acoustic image data are acquired through a multi-band acoustic transducer array, a high-precision differential pressure sensor network is combined to monitor the water body differential pressure change, and an adaptive weight convolutional neural network is utilized to realize multi-source data intelligent fusion. The network has a weight coefficient dynamic adjustment mechanism, can dynamically adjust a data fusion weight according to environmental parameters such as real-time water turbidity and flow velocity, and adaptively optimizes a fusion coefficient based on signal quality, thereby improving detection precision and environmental adaptability. The system also supports an integrated optical imaging module, and realizes three-mode fusion detection of acoustics, differential pressure and optics. The method effectively solves the problems that a traditional detection method is insufficient in precision and weak in anti-interference capacity in a complex underwater environment, and has the advantages of being high in real-time performance, high in reliability and suitable for long-term unattended monitoring.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering equipment testing technology, specifically to a trash rack blockage detection system based on multi-band optical acoustics and pressure difference fusion. Background Technology

[0002] Trash grates, as crucial facilities in water conservancy projects, are primarily used to intercept floating debris and other contaminants in water bodies, preventing them from entering critical equipment such as water turbines and causing damage. With the continuous expansion of water conservancy projects and the increasing complexity of operating environments, the importance of trash rack blockage detection technology has become increasingly prominent. Traditional manual inspection methods are not only inefficient but also pose safety hazards, making them unsuitable for the operational needs of modern water conservancy projects.

[0003] In recent years, the technology for detecting debris rack blockages based on acoustic imaging and differential pressure monitoring has received widespread attention and application. Chinese patent CN119395708B discloses a high-resolution imaging method and system for a hydrophone array. This technology improves imaging clarity under different depths and environmental conditions by plotting sound field distribution maps, analyzing imaging quality indicators, and weighted fusion of acoustic signals from different frequency bands. Chinese patent CN120046815B discloses a method and system for assessing fishery resources across the entire water column using fisheries acoustic methods. It uses a multi-band sonar system to pre-detect target water bodies, acquire sonar echo data, analyze detection quality, and determine the optimal sonar frequency range. Chinese patent CN120871148A proposes a fish deterrent method and system based on multi-frequency acoustic wave coordination and dynamic environmental perception. This method constructs a multi-band acoustic wave emission parameter set based on aquatic environmental parameters and controls the transducer array to emit multi-band composite acoustic wave signals.

[0004] Regarding multimodal data fusion, Chinese patent CN119828148B discloses an underwater multimodal identification method and system for hidden defects in high dams. This technology acquires displacement, stress, and temperature monitoring data, extracts abnormal features using an adaptive weighting method, and combines acoustic and optical data for feature fusion. Chinese patent CN121167371A provides an underwater target identification method and system that uses multi-source sensor data fusion to dynamically weight and fuse acoustic wave reflection signals and optical image data through an adaptive fusion module.

[0005] However, existing trash rack blockage detection technologies still have many shortcomings: First, the design and placement of traditional hydrophone arrays have limitations, resulting in limited imaging capabilities under different water depths, flows, and water qualities, reduced spatial resolution, insufficient ability to identify small blockages, and a tendency to create imaging blind spots. Second, single-band acoustic imaging often employs passive imaging methods, making it difficult to flexibly adjust transmission parameters. Traditional sonar uses a single frequency or a broadband signal with a limited frequency range, which has biased penetration and reflection capabilities for blockages of different particle sizes, materials, and densities. Third, existing technologies perform poorly in handling complex environmental interference, especially in areas with high turbidity or strong currents, where sound wave signal attenuation is severe, echo signal quality is poor, and it is difficult to accurately identify the blockage status. In addition, the sensitivity and accuracy of hardware configurations are insufficient, sensor accuracy is not high enough, response speed is slow, and power consumption management and battery life cannot meet the needs of long-term continuous monitoring. Finally, existing technologies lack effective multi-source information fusion mechanisms, often processing acoustic data and differential pressure data independently, failing to comprehensively judge the blockage status from both macroscopic and microscopic levels, thus affecting the accuracy and reliability of detection results. Summary of the Invention

[0006] To address the limitations of existing underwater trash rack clogging detection technologies, such as the design and placement of traditional hydrophone arrays, the limitations of single-band acoustic imaging, insufficient environmental interference handling capabilities, inadequate hardware sensitivity and accuracy, and lack of multi-source data fusion, this paper proposes a trash rack clogging detection system based on multi-band acoustic and differential pressure fusion. This system aims to achieve high-precision multi-band fusion imaging, high-density distributed differential pressure monitoring, intelligent adaptive weighted convolutional neural network fusion, edge computing and real-time processing, as well as system collaboration and remote monitoring.

[0007] The technical solution adopted by this invention to solve its technical problem is as follows: a trash rack blockage detection system based on multi-band acoustic and differential pressure fusion is provided, including a multi-band acoustic imaging module for acquiring acoustic image data of the trash rack; a distributed differential pressure sensing network for acquiring differential pressure change data of the water body surrounding the trash rack; and an edge fusion processing system connected to the multi-band acoustic imaging module and the distributed differential pressure sensing network respectively, for receiving and processing the acoustic image data and the differential pressure change data.

[0008] Preferably, the edge fusion processing system is configured to use an adaptive weighted convolutional neural network to fuse the acoustic image data and the pressure difference change data. The adaptive weighted convolutional neural network includes a dynamic adjustment mechanism for weight coefficients and a feature fusion layer.

[0009] Preferably, the dynamic adjustment mechanism of the weighting coefficient is used to dynamically adjust the weighting coefficients of the acoustic image data and the pressure difference change data in the fusion process according to real-time water environment parameters; the feature fusion layer is used to dynamically adjust the fusion coefficients according to the signal quality of the acoustic image data and the pressure difference change data, so as to generate fusion features and output blockage detection results.

[0010] Preferably, the multi-band acoustic imaging module includes a multi-band acoustic transducer array, which includes at least two acoustic transmitters and receivers with different center frequencies, specifically acoustic transmitters and receivers with center frequencies of 0.9MHz, 2.5MHz and 4MHz respectively.

[0011] Preferably, the at least two acoustic transmitters and receivers with different center frequencies adopt a time-division multiplexing working mechanism to avoid crosstalk between different frequency bands; the multi-band acoustic imaging module uses frequency modulation continuous wave technology to transmit linear frequency modulated signals and uses a phase coherent accumulation algorithm to process the echo signals.

[0012] Preferably, the distributed differential pressure sensing network includes multiple high-precision differential pressure sensors, specifically 12 differential pressure sensors with an accuracy of ±0.01 kPa. The differential pressure sensors are symmetrically arranged on the inlet and outlet sides of the trash rack and are distributed in layers in the vertical direction.

[0013] Preferably, the dynamic adjustment mechanism for the weighting coefficient adjusts the weighting coefficient based on real-time water turbidity and real-time water flow velocity, and the weighting coefficient is dynamically adjusted using the following formula: ,in, These are the adjusted weighting coefficients. The base weighting coefficients are α and β, which are adjustment coefficients, and NTU represents the real-time water turbidity. For reference turbidity, v is the real-time water flow velocity. For reference flow rate.

[0014] Preferably, the feature fusion layer dynamically adjusts the fusion coefficients based on the signal-to-noise ratio of the acoustic image data and the signal-to-noise ratio of the pressure difference change data, and the fused features are generated using the following formula: Where F_fused is the fusion feature, Acoustic image features, λ represents the pressure difference characteristic, and λ is the fusion coefficient dynamically calculated based on the signal-to-noise ratio of the acoustic image data and the signal-to-noise ratio of the pressure difference change data.

[0015] Preferably, the system further includes a control system, which is connected to the multi-band acoustic imaging module, the distributed differential pressure sensing network, and the edge fusion processing system respectively, for coordinating the data acquisition and processing process. The control system includes a wireless communication module for data interaction with a remote monitoring center.

[0016] The beneficial effects of this invention are as follows: By employing a multi-band acoustic transducer array composed of acoustic transmitters and receivers at three different frequency bands (0.9MHz, 2.5MHz, and 4MHz), combined with a time-division multiplexing working mechanism and frequency modulation continuous wave technology, the optimal frequency band can be selected for different water turbidity and detection distances, achieving a spatial resolution of 0.5cm. It also exhibits good penetration capability even in environments with water turbidity as high as 80 NTU, effectively solving the limitations of traditional hydrophone array design and placement. A distributed differential pressure sensing network composed of 12 high-precision differential pressure sensors with an accuracy of ±0.01 kPa is symmetrically arranged in layers on the inlet and outlet sides of the trash rack, enabling real-time and accurate monitoring of differential pressure changes in the water surrounding the trash rack, providing precise macroscopic hydraulic data. An innovative adaptive weighted convolutional neural network is introduced, integrating a dynamic adjustment mechanism for weight coefficients. This allows for intelligent adjustment of the weights of acoustic and differential pressure data based on real-time environmental parameters such as water turbidity and flow velocity. The fusion coefficients are dynamically optimized through a signal-to-noise ratio weighted feature fusion strategy, achieving deep data fusion and robustness. The invention is implemented using NVIDIA Jetson Xavier. The NX platform enables real-time data processing within 50ms, with power consumption controlled within 15W, supporting 5000 hours of continuous operation, meeting the requirements for real-time performance, reliability, and battery life; it also enables remote monitoring, fault diagnosis, and parameter adjustment via 4G or 5G networks, improving the system's intelligent management level and operational efficiency. Attached Figure Description

[0017] Figure 1 This is the overall system architecture diagram for this solution; Figure 2 This is a flowchart of the data processing process of the edge fusion processing system in this solution; Figure 3 This is a diagram of the actual interface for acoustic detection of contaminant blockage in this solution; Figure 4 This is a diagram of the actual interface for optical detection and contamination blocking in this solution. Detailed implementation method. The present invention will now be described in detail with reference to the accompanying drawings and embodiments. Example

[0018] This invention provides a multi-band acoustic and differential pressure fusion trash rack blockage detection system, which achieves comprehensive, accurate and real-time monitoring of the trash rack blockage status by coordinating multiple sensing technologies.

[0019] like Figure 1As shown, the detection system mainly consists of four core parts: a multi-band acoustic imaging module, a distributed differential pressure sensing network, an edge fusion processing system, and a control system. These parts work in coordination to overcome the limitations of single sensors or single technologies, providing a reliable technical solution for monitoring trash racks in hydraulic and marine engineering environments.

[0020] The multi-band acoustic imaging module is responsible for achieving acoustic imaging of the debris barrier structure using acoustic principles. This module can penetrate water with a certain turbidity to acquire information on the morphology, location, and density of the blockage. The module includes a multi-band acoustic transducer array, which consists of at least two acoustic transmitters and receivers with different center frequencies. Specifically, the multi-band acoustic transducer array includes acoustic transmitters and receivers with center frequencies of 0.9MHz, 2.5MHz, and 4MHz. This multi-band design allows the system to balance penetration and resolution. The 0.9MHz band has good underwater penetration capability, suitable for high turbidity or long-distance detection; the 2.5MHz band provides medium resolution and penetration depth, suitable for conventional detection; and the 4MHz band provides extremely high spatial resolution, suitable for fine imaging of fine structures and close-range blockages.

[0021] To optimize the efficiency of multi-band transducers and reduce mutual interference, at least two acoustic transmitters and receivers with different center frequencies employ a time-division multiplexing mechanism. This means they do not operate simultaneously, but rather alternately transmit and receive according to a preset time sequence. For example, they operate at 0.9MHz for a period, then switch to 2.5MHz, then to 4MHz, completing one cycle before repeating. This mechanism effectively avoids crosstalk between different frequency bands and ensures data quality.

[0022] To improve imaging quality and anti-interference capabilities, the multi-band acoustic imaging module employs frequency-modulated continuous wave (FM) technology to transmit linear frequency-modulated (LFM) signals. This technology transmits a continuous wave signal whose frequency varies linearly with time and determines the target distance by receiving the difference frequency between the transmitted and received echo signals. Compared to pulse sonar, this technology offers a higher signal-to-noise ratio (SNR) and range resolution. Furthermore, the module uses a phase-coherent accumulation algorithm to process the echo signals. By coherently superimposing the echo signals from multiple transmissions and receptions, the SNR of the echo signals is significantly improved, effectively suppressing random noise and enabling clear acoustic images even in low SNR environments.

[0023] A distributed differential pressure sensing network provides macroscopic hydraulic changes in the water surrounding trash racks, playing a crucial role in determining the severity and extent of blockages. The network comprises multiple high-precision differential pressure sensors, specifically 12 sensors with an accuracy of ±0.01 kPa. These high-precision sensors can detect extremely minute pressure changes, which is essential for early detection and assessment of blockage severity.

[0024] The differential pressure sensors employ an optimized distributed arrangement, symmetrically positioned on both the inlet and outlet sides of the trash rack, and layered vertically. Specifically, six differential pressure sensors are symmetrically arranged on each of the inlet and outlet sides of the trash rack, forming a symmetrical distribution. Vertically, these sensors are divided into three layers, with four sensors in each layer, ensuring coverage of pressure changes at different water depths. This three-dimensional, symmetrical arrangement comprehensively monitors the hydraulic load in different areas of the trash rack, accurately capturing changes in water level differences and flow velocity distribution caused by blockages, thereby constructing a refined hydraulic model.

[0025] The edge fusion processing system is the core processing unit of the system. It is connected to the multi-band acoustic imaging module and the distributed differential pressure sensing network, respectively. It is responsible for receiving the raw data collected by the two modules, and using advanced algorithms to perform deep fusion, analysis and decision-making on the data, and finally output the blockage detection results.

[0026] The edge fusion processing system is configured to use an adaptive weighted convolutional neural network. This network has a dynamic weight adjustment mechanism to adjust the weight coefficients based on real-time aquatic environmental parameters. To adapt to the complex and ever-changing underwater environment, the network's weight coefficients are not fixed but adjusted according to real-time turbidity and flow velocity. The weight coefficients are dynamically adjusted using the following formula: ,in These are the adjusted weighting coefficients. The base weights are α and β, which are adjustment coefficients, and NTU is the real-time monitored turbidity value of the water body. The reference turbidity value is given, and v is the real-time monitored water flow velocity. The reference flow velocity value is used. Through this mechanism, when the turbidity of the water increases, the weight of the acoustic image data may decrease relatively, while the weight of the pressure difference data may increase relatively, and vice versa, thereby ensuring the reliability of the fusion results under different environments.

[0027] The edge fusion processing system also includes a feature fusion layer, which dynamically adjusts the fusion coefficients based on signal quality, generates fused features, and outputs the blockage detection results. The feature fusion layer dynamically adjusts the fusion coefficients based on the signal-to-noise ratio of the acoustic image data and the signal-to-noise ratio of the pressure difference change data. The fused features are generated using the following formula: ,in The characteristics after fusion Acoustic characteristics, For differential pressure characteristics, the fusion coefficient λ is dynamically adjusted based on the real-time calculated acoustic signal-to-noise ratio and differential pressure signal-to-noise ratio. When the acoustic signal quality is high, λ approaches 1, and the acoustic characteristics dominate the fusion; when the differential pressure signal quality is high, λ approaches 0, and the differential pressure characteristics dominate. This dynamic adjustment mechanism ensures that the system always uses the highest quality data for decision-making, improving the system's robustness.

[0028] The system also includes a control system, which is connected to the multi-band acoustic imaging module, the distributed differential pressure sensing network, and the edge fusion processing system to coordinate data acquisition and processing. As the command center of the entire system, the control system is responsible for coordinating the work between various modules, enabling automated operation, data management, status monitoring, and communication with external systems. The control system sends commands to the acoustic imaging module to control its frequency band selection, transmission power, transmission mode, and other parameters; it sends commands to the distributed differential pressure sensing network to control its sampling frequency, sampling period, and other parameters; and it is also responsible for managing the task scheduling and status monitoring of the edge fusion processing system.

[0029] The control system includes a wireless communication module for data interaction with a remote monitoring center. This module communicates with the center via a 4G or 5G wireless network for data exchange and control command transmission. This allows operators to remotely configure the trash rack blockage detection system, check its status, download data, and receive real-time alarm information from the remote monitoring center, even when not on-site. The wireless communication module ensures the system's manageability and data feedback capabilities in remote or unattended environments.

[0030] The system's overall workflow follows a closed-loop model of data acquisition, preprocessing, intelligent fusion, decision output, and remote interaction. After power-on, the control system initiates self-test procedures for each module and acquires current water environment parameters based on preset parameters. The control system coordinates the acoustic imaging module and the distributed differential pressure sensing network to synchronously acquire data. Based on environmental parameters, the acoustic imaging module uses a time-division multiplexing mechanism and frequency modulation continuous wave technology to sequentially transmit linear frequency modulated signals using acoustic transmitters and receivers at different frequency bands and receive echoes. Simultaneously, a phase coherence accumulation algorithm is used to process the echo signals to improve the signal-to-noise ratio, generating acoustic image data. The distributed differential pressure sensing network's 12 high-precision differential pressure sensors synchronously acquire pressure data from both the inlet and outlet sides and calculate instantaneous differential pressure values.

[0031] like Figure 2As shown, the edge fusion processing system preprocesses the real-time received acoustic image data and raw pressure difference data, including denoising, filtering, compensation, and format conversion, to improve data quality. The preprocessed data is input into an adaptive weighted convolutional neural network. A dynamic adjustment mechanism for the weight coefficients dynamically adjusts the neural network's weight coefficients based on real-time environmental parameters, enabling the network to better adapt to the current underwater environment. The feature fusion layer dynamically adjusts the fusion coefficients based on the real-time signal-to-noise ratio of the acoustic and pressure difference data, achieving optimized fusion of the two types of features and generating fused features containing richer and more reliable information. The classifier analyzes the fused features and outputs a classification result for the debris barrier's clogging status, such as "no clogging," "light clogging," "moderate clogging," or "severe clogging," along with the corresponding confidence level.

[0032] By intelligently fusing multi-band acoustic and differential pressure data, this system can effectively integrate and make decisions based on multi-source heterogeneous data, thereby providing a high-precision and robust solution for detecting trash rack blockages. The system possesses efficient edge computing and real-time processing capabilities, enabling rapid response to blockage conditions and timely early warnings, meeting the high requirements of real-time performance and reliability in industrial settings. Example 2 This embodiment describes in detail a detection system that integrates acoustic, optical, and differential pressure modes. Based on a trash rack blockage detection system that integrates multi-band acoustic and differential pressure modes, this system further integrates an optical imaging module to form a three-mode fusion detection capability, which significantly improves detection accuracy and environmental adaptability.

[0033] The trimodal fusion detection system comprises: a trash rack clogging detection system based on multi-band acoustic and differential pressure fusion, an optical imaging module, and an edge fusion processing system. The multi-band acoustic and differential pressure fusion-based trash rack clogging detection system provides acoustic image data and differential pressure change data; the optical imaging module acquires optical image data of the trash rack; and the edge fusion processing system, connected to the optical imaging module, performs trimodal data fusion processing on the acoustic image data, differential pressure change data, and optical image data.

[0034] The optical imaging module comprises an underwater camera module and a controllable light source module. The underwater camera module uses a high-resolution CCD sensor, enabling it to acquire clear image information in complex underwater environments. The controllable light source module provides illumination for the underwater camera module, ensuring high-quality optical images even in low-light or turbid water conditions. The controllable light source module includes a white LED light source and a specific wavelength LED light source. The white LED light source provides full-spectrum illumination, suitable for conventional image acquisition; the specific wavelength LED light source emits light of a specific wavelength, capable of penetrating water with specific turbidity levels, enhancing imaging performance in high-turbidity environments.

[0035] The optical imaging module comprises fixed optical observation points and mobile underwater inspection units. The fixed optical observation points are installed at key locations on the debris barrier, providing continuous optical monitoring; the mobile underwater inspection units can be moved to designated areas for detailed inspection as needed, achieving comprehensive optical coverage. This combination of fixed and mobile deployment ensures both comprehensiveness and flexibility in optical monitoring.

[0036] The edge fusion processing system adopts a layered-stage fusion strategy, which consists of three stages: In the first stage, an early warning is triggered based on acoustic image data and differential pressure change data. When the acoustic imaging detects an abnormal structure or the differential pressure sensor detects an abnormal pressure change, the system automatically enters an early warning state. In the second stage, in response to the early warning, the optical imaging module is controlled to conduct a detailed investigation of the warning area. By precisely controlling the controllable light source module and the underwater camera module, high-resolution optical images of the warning area are acquired. In the third stage, a decision-level fusion is performed based on the acoustic image data, differential pressure change data, and the optical image data obtained from the detailed investigation, and the final blockage detection result is output.

[0037] The dynamic adjustment mechanism for weighting coefficients is further used to adjust the weighting coefficients based on the optical visibility factor (VF). The optical visibility factor (VF) comprehensively considers factors affecting optical imaging quality, such as water transparency, illumination conditions, and suspended solids concentration. When the VF value is high, it indicates good optical imaging conditions, and the system will increase the weight of optical data in the fusion process. When the VF value is low, the system will correspondingly reduce the weight of optical data and rely more on acoustic and differential pressure data.

[0038] The adjusted weighting coefficients are calculated using the following formula: ,in The adjusted weighting coefficients are: W_base (base weighting coefficients), α, β, and γ (adjustment parameters), and NTU (real-time turbidity value). Here is the reference turbidity value, and v is the real-time flow rate. For reference flow rate, The optical visibility factor (VF) is used. This formula enables dynamic adjustment of weights based on multiple environmental parameters. The weights are adjusted accordingly when turbidity or flow rate increases. The introduction of the optical visibility factor (VF) allows the system to adaptively adjust the weights of optical data based on the quality of optical imaging conditions.

[0039] like Figure 3-4As shown, the working principle of this three-modal fusion detection system is as follows: First, the acoustic imaging module acquires acoustic images of the trash rack structure through a multi-band acoustic transducer array, while a distributed differential pressure sensing network monitors differential pressure changes in real time. When an anomaly is detected, the system triggers the optical imaging module to perform a detailed optical inspection of the abnormal area. Finally, the edge fusion processing system employs a hierarchical-stage fusion strategy, combined with a dynamic adjustment mechanism for weight coefficients, to intelligently fuse the three modal data and output accurate blockage detection results. Through the collaborative work of acoustic, optical, and differential pressure three-modal data, the system can achieve high-precision and high-reliability trash rack blockage detection in various complex underwater environments. Example

[0040] This invention provides a system and method for monitoring the blockage status of trash racks in pumping stations. The method achieves accurate detection of the blockage status of trash racks through multi-source data fusion and adaptive weight adjustment mechanism.

[0041] S1: Acquire multi-band acoustic image data First, acoustic image data of the debris barrier is acquired from the multi-band acoustic imaging module. The multi-band acoustic imaging module uses acoustic transmitters and receivers in three different frequency bands: 0.9MHz, 2.5MHz, and 4MHz, performing acoustic imaging sequentially through a time-division multiplexing mechanism. The 0.9MHz band has good underwater penetration capability and is suitable for high turbidity or long-distance detection; the 2.5MHz band provides medium resolution and penetration depth, suitable for conventional detection; and the 4MHz band provides extremely high spatial resolution, suitable for fine imaging of fine structures and close-range blockages.

[0042] The acoustic imaging module employs frequency-modulated continuous wave (FM) technology to transmit a linear frequency-modulated (LFM) signal. The target distance is determined by the frequency difference between the received echo signal and the transmitted signal. Simultaneously, a phase-coherent accumulation algorithm is used to process the echo signal. By coherently superimposing multiple transmitted and received echo signals, the signal-to-noise ratio of the echo signal is significantly improved. This multi-band collaborative imaging method yields high-quality acoustic image data containing information on the structure of the debris barrier and the morphology, location, and density of attached blockages.

[0043] S2: Obtain distributed differential pressure change data Acquire differential pressure change data of the water body surrounding the trash rack from a distributed differential pressure sensing network. The distributed differential pressure sensing network contains 12 high-precision differential pressure sensors with an accuracy of ±0.01 kPa. Six differential pressure sensors are symmetrically arranged on the inlet and outlet sides of the trash rack, and four sensors are arranged in each of the three vertical layers.

[0044] Each differential pressure sensor calculates the pressure difference at its corresponding location in real time using the following formula: Where ΔP(i,j) represents the pressure difference at a specific location. This indicates the pressure value on the inlet side of the trash rack at that location. This indicates the pressure value on the outlet side of the trash rack at the corresponding location. This three-dimensional, symmetrical arrangement allows for comprehensive monitoring of the hydraulic load in different areas of the trash rack, accurately capturing changes in water level difference and flow velocity distribution caused by blockages, and constructing a refined hydraulic model.

[0045] S3: Multi-source data fusion processing Acoustic image data and pressure difference change data are input into an adaptive weighted convolutional neural network for fusion processing. The fusion process includes two key stages: dynamic adjustment of weight coefficients and feature fusion. In the dynamic weight coefficient adjustment stage, based on real-time water environment parameters, the weight coefficients of the acoustic image data and pressure difference change data are dynamically adjusted during the fusion process using a dynamic weight coefficient adjustment mechanism. The weight adjustment formula is: Where W_adj is the adjusted weight coefficient, The base weights are α and β, which are adjustment coefficients, and T is the real-time monitored turbidity value of the water body. The reference turbidity value is V, and the real-time monitored water flow velocity is V. The reference flow velocity value is used. As water turbidity increases, the weight of acoustic image data decreases relatively, while the weight of pressure difference data increases relatively, ensuring the reliability of the fusion results under different environments.

[0046] In the feature fusion stage, the fusion coefficients are dynamically adjusted by the feature fusion layer based on the signal quality of the acoustic image data and the pressure difference change data. The feature fusion layer employs a signal-to-noise ratio weighted fusion strategy, and the fusion formula is as follows: ,in The fusion coefficient λ is the fused feature, dynamically adjusted based on the real-time calculated acoustic signal-to-noise ratio and pressure difference signal-to-noise ratio. The calculation formula is as follows: When the acoustic signal quality is high, λ approaches 1, and acoustic features dominate the fusion process; when the pressure difference signal quality is high, λ approaches 0, and pressure difference features dominate, ensuring that the system always uses the best quality data for decision-making.

[0047] S4: Output blockage detection results Based on fused features, a classifier analyzes and processes these features to output the trash rack clogging detection results. The classifier categorizes the trash rack status into four levels—"no clogging," "slightly clogging," "moderately clogging," or "severely clogging"—based on the feature vectors of the fused features, and provides corresponding confidence levels. The entire process is completed within 50ms, achieving real-time clogging status monitoring.

[0048] In another preferred embodiment, when a blockage is detected to reach a preset alarm threshold, the system sends alarm information to a remote monitoring center via a wireless communication module, and simultaneously triggers a local audible and visual alarm to achieve timely warning and response.

[0049] Through the above methods and steps, intelligent deep fusion of acoustic imaging data and differential pressure data is achieved. The fusion strategy is dynamically adjusted according to real-time environmental parameters and signal quality, which significantly improves the accuracy, environmental adaptability and system robustness of debris barrier blockage detection. It solves the technical problem of insufficient detection accuracy of single sensor or fixed weight fusion methods in complex underwater environments. Example

[0050] This invention provides a method for detecting blockage of trash racks based on the fusion of acoustic, optical and differential pressure modes. This method further introduces optical image data on the basis of the dual-modal fusion described in claim 21, so as to realize the synergistic fusion of three different modal data, which significantly improves the detection accuracy and environmental adaptability.

[0051] S1: Perform multi-band acoustic and differential pressure fusion detection The process involves acquiring acoustic image data of the trash rack from the multi-band acoustic imaging module, acquiring differential pressure change data of the water body surrounding the trash rack from the distributed differential pressure sensing network, and inputting the acoustic image data and differential pressure change data into an adaptive weighted convolutional neural network for preliminary fusion processing.

[0052] In this step, the multi-band acoustic imaging module uses acoustic transmitters and receivers in three frequency bands (0.9MHz, 2.5MHz, and 4MHz) for time-division multiplexing, and obtains high-quality acoustic image data through frequency modulation continuous wave technology and phase coherence accumulation algorithm. Twelve high-precision differential pressure sensors in the distributed differential pressure sensing network monitor the pressure difference changes at the inlet and outlet of the trash rack in real time, constructing a refined hydraulic model. An adaptive weighted convolutional neural network dynamically adjusts the fusion weights of the acoustic and differential pressure data based on real-time water environment parameters, generating preliminary dual-modal fusion features.

[0053] S2: Acquire optical image data Optical image data of the debris barrier was acquired. This data was collected using an underwater optical imaging device deployed near the barrier, equipped with a high-resolution CCD sensor and an LED lighting system. The device utilizes a 1920×1080 resolution industrial-grade underwater camera and is equipped with six high-power white LEDs to provide uniform illumination, with an effective illumination distance of up to 5 meters.

[0054] Optical image data can intuitively reflect the visual characteristics of blockages on the surface of trash racks, such as color, texture, and shape—information that acoustic and differential pressure data cannot provide. Especially for transparent or translucent blockages, such as plastic film and aquatic plants, optical images can provide more accurate identification. The optical imaging system acquires 30 frames per second, and uses image preprocessing algorithms for denoising, enhancement, and normalization to extract optical feature data containing information on blockage distribution, density, and type.

[0055] S3: Trimodal Fusion Processing Acoustic image data, pressure difference change data, and optical image data are fused into three modalities to output the final blockage detection result. The three-modal fusion processing adopts an improved adaptive weighted convolutional neural network architecture, which extends the optical feature processing branch on the basis of the original two-modal fusion.

[0056] The trimodal fusion network comprises three parallel feature extraction branches: acoustic feature extraction, pressure differential feature extraction, and optical feature extraction. Each branch employs specially designed convolutional and pooling layers to extract deep features corresponding to its modality. The acoustic branch primarily extracts the spatial distribution and density features of the blockage, the pressure differential branch primarily extracts hydraulic blockage features, and the optical branch primarily extracts the visual appearance features of the blockage. In the feature fusion stage, a ternary adaptive weight adjustment mechanism is used, with the fusion weights dynamically allocated based on the real-time quality of the three modalities. The weight adjustment formula is: , ,in , and These represent the signal-to-noise ratios of acoustic, differential pressure, and optical data, respectively.

[0057] The three-modal fusion features are generated through a weighted summation method: ,in and These are feature vectors extracted from the three branches. The fused features contain more comprehensive and complementary congestion information, effectively overcoming the limitations of a single modality in specific environments.

[0058] Finally, the fused features are input into a multilayer perceptron classifier to determine the clogging status, outputting four levels of detection results: "no clogging," "mild clogging," "moderate clogging," or "severe clogging," along with corresponding confidence scores. The entire three-modal fusion processing is completed within 80ms, meeting real-time detection requirements.

[0059] By incorporating optical image data, this three-modal fusion method can provide more accurate blockage detection results in relatively clear water environments, especially demonstrating stronger identification capabilities for blockages with complex shapes and diverse materials. When the data quality of one modality deteriorates, the other two modalities can provide effective compensation, significantly improving the robustness and detection accuracy of the system.

[0060] The above description is merely a preferred embodiment of the present invention, and the present invention is not limited to the above embodiments. It is understood that other improvements and variations that are directly derived or conceived by those skilled in the art without departing from the spirit and concept of the present invention should be considered to be included within the protection scope of the present invention.

Claims

1. A trash rack clogging detection system, characterized in that, include: A multi-band acoustic imaging module is used to acquire acoustic image data of the trash rack; A distributed differential pressure sensing network is used to acquire differential pressure change data of the water body surrounding the trash rack; The edge fusion processing system is connected to the multi-band acoustic imaging module and the distributed differential pressure sensing network, respectively. The edge fusion processing system is configured to use an adaptive weighted convolutional neural network to fuse the acoustic image data and the pressure difference change data. The adaptive weighted convolutional neural network includes: A dynamic weighting coefficient adjustment mechanism is used to dynamically adjust the weighting coefficients of the acoustic image data and the pressure difference change data during the fusion process based on real-time water environment parameters. The feature fusion layer is used to dynamically adjust the fusion coefficients based on the signal quality of the acoustic image data and the differential pressure change data, so as to generate fusion features and output the blockage detection results.

2. The system according to claim 1, characterized in that, The dynamic adjustment mechanism for the weighting coefficient adjusts the weighting coefficient based on the real-time turbidity and real-time flow velocity of the water body.

3. The system according to claim 2, characterized in that, The weighting coefficients are dynamically adjusted using the following formula: ， in, These are the adjusted weighting coefficients. The base weighting coefficients are α and β, which are adjustment coefficients, and NTU represents the real-time water turbidity. For reference turbidity, v is the real-time water flow velocity. For reference flow rate.

4. The system according to claim 1, characterized in that, The feature fusion layer dynamically adjusts the fusion coefficients based on the signal-to-noise ratio of the acoustic image data and the signal-to-noise ratio of the pressure difference change data.

5. The system according to claim 4, characterized in that, The fusion feature is generated using the following formula: ， in, As a feature of fusion, Acoustic image features, λ represents the pressure difference characteristic, and λ is the fusion coefficient dynamically calculated based on the signal-to-noise ratio of the acoustic image data and the signal-to-noise ratio of the pressure difference change data.

6. A trash rack clogging detection system, characterized in that, include: The trash rack clogging detection system as described in any one of claims 1-5; An optical imaging module is used to acquire optical image data of the trash rack; The edge fusion processing system is further connected to the optical imaging module and configured to perform three-modal data fusion processing on the acoustic image data, the pressure difference change data, and the optical image data.

7. The system according to claim 6, characterized in that, The dynamic adjustment mechanism for the weighting coefficients is further used to adjust the weighting coefficients according to the optical visibility factor (VF). The adjusted weighting coefficients are calculated using the following formula: ， Wherein, VF is the optical visibility factor, and γ is the corresponding adjustment coefficient.

8. A method for detecting blockage in a trash rack, characterized in that, include: Acquire acoustic image data of the trash rack; Acquire data on the pressure difference changes in the water body surrounding the trash rack; The acoustic image data and the pressure difference change data are input into an adaptive weighted convolutional neural network for fusion processing. The fusion process includes: dynamically adjusting the weighting coefficients of the acoustic image data and the pressure difference change data in the fusion process according to real-time water environment parameters, and dynamically adjusting the fusion coefficients according to the signal quality of the acoustic image data and the pressure difference change data to generate fusion features; Based on the fusion features, the blockage detection results of the trash rack are output.

9. The method according to claim 8, characterized in that, The weighting coefficients are dynamically adjusted based on real-time water turbidity and real-time water flow velocity, and are achieved through the following formula: ， The fusion features are dynamically generated using the following formula: ， Wherein, λ is a fusion coefficient dynamically calculated based on the signal-to-noise ratio of the acoustic image data and the signal-to-noise ratio of the pressure difference change data.

10. A method for detecting blockage in a trash rack, characterized in that, include: Perform the trash rack blockage detection method as described in claim 8 or 9; Acquire optical image data of the trash rack; The acoustic image data, the pressure difference change data, and the optical image data are fused in three modes to output the blockage detection result.

Images