A service information processing method and system based on multi-source data fusion

Abstract

The application discloses a kind of based on multi-source data fusion's service information processing method and system, receive multi-source original data frame and carry out time synchronization, data calibration, noise suppression and standardization processing;In distributed computing node, based on data quality score and dynamic weight, local fusion is executed, and the local fusion result is spatio-temporal registration, credibility evaluation and global fusion;Real-time signal enhancement, target identification, target tracking, state determination and real-time feedback, decision suggestion and alarm output are further executed.This application solves the problem that the marine service information processing in the prior art relies on a single data source, has poor noise resistance, insufficient real-time performance and weak collaboration ability, and realizes high-precision perception, stable fusion and rapid response of service information in complex marine environment.

Description

Technical Field

[0001] This invention belongs to the field of marine information processing and intelligent detection technology, specifically relating to a business information processing method and system based on multi-source data fusion, and more particularly to a multi-source collaborative sensing, distributed fusion computing and real-time decision support method and system suitable for business information processing of marine acoustic equipment in complex marine environments. Background Technology

[0002] With the increasing demands for marine exploration, marine monitoring, underwater security, and offshore resource development, marine acoustic equipment is being used more and more widely in scenarios such as underwater detection, environmental perception, target tracking, and mission coordination. In actual operation, marine acoustic equipment not only needs to process underwater acoustic signals collected by acoustic sensors, but also needs to comprehensively consider various information such as water temperature, salinity, current velocity, depth, background noise intensity, and node link status in order to make reliable judgments on target status, environmental changes, and mission execution.

[0003] In existing technologies, marine operational information processing solutions often rely on a single data source or a centralized processing architecture, typically transmitting data collected by various sensors back to a central server for analysis. This approach has the following drawbacks: First, a single data source is susceptible to ocean background noise, current disturbances, equipment drift, and propagation attenuation, resulting in poor result stability. Second, centralized processing models face high communication pressure and processing latency in large-scale, multi-node scenarios, making it difficult to meet real-time task requirements.

[0004] Therefore, there is an urgent need for a business information processing method and system that can comprehensively utilize acoustic data, environmental data, and background noise data in complex marine environments, and combine distributed computing and dynamic fusion mechanisms to achieve high real-time performance, high robustness, and high scalability. Summary of the Invention

[0005] This application provides a business information processing method and system based on multi-source data fusion, aiming to solve the technical problems in the existing technology of marine acoustic equipment information processing that rely on a single data source or centralized architecture, resulting in low data utilization, large noise interference, high processing latency, and insufficient system scalability in complex marine environments. The method aims to achieve high-precision perception, high-reliability fusion, high-real-time analysis, and high-stability decision-making for business information in complex marine environments. The method includes: continuously receiving multi-source raw data frames sent by distributed acquisition nodes; performing time synchronization, data calibration, noise suppression, and standardization processing on the multi-source raw data frames to generate a standardized multi-source data frame sequence; continuously reading the standardized multi-source data frame sequence; performing local data fusion based on multi-source data quality scores and dynamic weights within each distributed computing node to generate a local fusion result sequence; continuously receiving the local fusion results output by each distributed computing node; performing spatiotemporal registration, credibility assessment, and global data fusion on the local fusion results to generate a global fusion result; continuously reading the global fusion result; performing real-time signal analysis, target recognition, target tracking, and status determination to generate business processing results; and outputting real-time feedback information, decision suggestions, and alarm information based on the business processing results.

[0006] A second aspect of this application provides a business information processing system based on multi-source data fusion. The system includes: a multi-source data acquisition and preprocessing module, which continuously receives multi-source raw data frames sent by distributed acquisition nodes, performs time synchronization, data calibration, noise suppression, and standardization processing on the multi-source raw data frames, and generates a standardized multi-source data frame sequence; a distributed local fusion module, which continuously reads the standardized multi-source data frame sequence and performs local data fusion based on multi-source data quality scores and dynamic weights within each distributed computing node, generating a local fusion result sequence; a global fusion module, which continuously receives the local fusion results output by each distributed computing node, performs spatiotemporal registration, credibility assessment, and global data fusion on the local fusion results, and generates a global fusion result; a business analysis module, which continuously reads the global fusion result, performs real-time signal analysis, target recognition, target tracking, and status determination, and generates a business processing result; and a feedback decision module, which outputs real-time feedback information, decision suggestions, and alarm information based on the business processing result.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application provides a business information processing method and system based on multi-source data fusion, relating to the field of marine information processing and intelligent detection technology. By performing local fusion within distributed computing nodes and global fusion at the central fusion layer, it effectively solves the technical problems in existing technologies, such as low recognition accuracy, large noise interference, poor real-time performance, weak fault tolerance, and difficulty in system expansion, caused by reliance on a single data source, centralized computing, and static fusion strategies in marine acoustic equipment information processing. This achieves the technical effects of high accuracy, fast response speed, strong stability, and good scalability in business information processing under complex marine environments. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of the overall process of a business information processing method based on multi-source data fusion provided in this application embodiment; Figure 2 A schematic diagram of a business information processing system based on multi-source data fusion is provided for an embodiment of this application; Figure labeling: 11 Multi-source data acquisition and preprocessing module, 12 Distributed local fusion module, 13 Global fusion module, 14 Business analysis module, 15 Feedback decision module. Detailed Implementation

[0010] This application provides a business information processing method and system based on multi-source data fusion, which solves the technical problems in the prior art where marine acoustic equipment information processing relies on a single data source or centralized processing method, making it difficult to stably achieve high-precision information processing in complex marine noise, dynamic changes in environmental parameters, communication link fluctuations, and multi-node collaborative scenarios, resulting in low target recognition accuracy, insufficient real-time response capability, large impact range of node failures, and difficulty in system expansion.

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0012] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, apparatus, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0013] Example 1, as Figure 1 As shown, this application provides a business information processing method based on multi-source data fusion. This method can be deployed in a distributed marine monitoring network composed of underwater sensor nodes, buoy nodes, edge computing nodes, and a central fusion node. The method includes: S100: Continuously receive multi-source raw data frames sent by distributed acquisition nodes, and perform time synchronization, data calibration, noise suppression and standardization processing on the multi-source raw data frames to generate a standardized multi-source data frame sequence; In this embodiment, the multi-source raw data includes at least: acoustic data collected by underwater microphones or sonar arrays, environmental data collected by temperature, salinity, and depth sensors, and background noise data collected by noise monitoring sensors. To adapt to multi-point, multi-scale, and multi-task scenarios in the ocean, multiple acquisition nodes are distributed along the monitoring area. Each acquisition node has a unique node identifier and establishes communication connections with the upper-level distributed computing nodes through underwater acoustic communication links, fiber optic links, satellite links, or wireless links.

[0014] Acoustic data is used to reflect the characteristics of the target sound source and the sound propagation state in the marine environment; environmental data is used to reflect changes in sound speed profile, propagation conditions, and sea state; and background noise data is used to describe tidal noise, shipping noise, biological noise, and equipment self-noise. Simultaneous collection of these multiple types of data provides an information foundation for subsequent multi-source collaborative fusion.

[0015] Specifically, each acquisition node generates multi-source raw data frames according to a preset sampling period. The sampling rate for acoustic data can be set to 32kHz or 48kHz, the sampling rate for environmental data can be set to 0.1Hz to 1Hz, and the sampling rate for background noise data can be the same as or proportional to that of the acoustic data. In this embodiment, the sampling rate for acoustic data is set to 32kHz, the sampling rate for environmental data is set to 0.5Hz, and the sampling rate for background noise data is set to 32kHz. Using 32kHz as the sampling rate for acoustic and background noise data can fully cover the main characteristic frequency bands of common marine acoustic targets while controlling data transmission pressure and edge node computational load. Using 0.5Hz as the sampling rate for environmental data can meet the tracking requirements of slowly changing environmental parameters such as sea temperature, salinity, current velocity, and depth, while avoiding redundant environmental data acquisition that would cause invalid storage and computational overhead. To facilitate unified fusion, timestamp, node identifier, sensor identifier, and quality flag fields are added when generating raw data frames.

[0016] S101: Perform time synchronization and rearrangement processing on the multi-source raw data frames to obtain time-aligned multi-source raw data frames; Because clock errors, communication delays, and buffer accumulation may exist between different acquisition nodes, multi-source raw data frames may be out of order or have time offsets when being sent. Therefore, it is necessary to perform time synchronization and rearrangement processing on the received multi-source raw data frames. Specifically, a unified time reference can be provided for different acquisition nodes using Global Navigation Satellite System (GNSS), Network Time Protocol (NAT), or Central Reference Pulse (CRP) time synchronization methods, and the multi-source raw data frames can be rearranged according to the original timestamps. In this embodiment, GNSS time synchronization is used to provide a unified time reference for different acquisition nodes. The advantages of using GNSS time synchronization are high timing accuracy, good cross-regional consistency, and the ability to effectively reduce the accumulated time error in multi-node collaborative scenarios, thereby improving the timing alignment accuracy of subsequent local fusion and global fusion.

[0017] Let the first Each node at time... The collected raw timestamps are The system time reference is Then the clock deviation can be expressed as: in: Indicates the first Each node at time... The clock deviation relative to the system's timing reference, Indicates the first Each node at time... The collected raw timestamps, Indicates the system at time 10:00 A unified time synchronization reference time stamp, Indicates the data collection node number. Indicates the sampling time or data frame number.

[0018] In this embodiment, the synchronization threshold is set to 2 seconds. If the time exceeds 2 seconds, a resynchronization command is output to the time synchronization module, and a "clock pending correction" flag is added to the current period data of that node. The advantage of setting the synchronization threshold to 2 seconds is that it can effectively suppress the impact of time drift of multiple nodes on the spatiotemporal registration accuracy, while avoiding frequent resynchronization due to an excessively small threshold, thus balancing system stability and time synchronization accuracy.

[0019] S102: Perform data calibration, missing data repair and dimensional unification processing on the time-aligned multi-source raw data frame to obtain the calibrated multi-source data frame; Because different sensors have different output dimensions, ranges, and zero biases, data calibration processing is required for multi-source raw data frames. For acoustic data, the sampled values ​​can be converted into sound pressure amplitudes based on the sensor sensitivity coefficients; for environmental data, the sampled values ​​can be converted into standard physical quantities such as temperature, salinity, flow rate, and depth based on the sensor calibration table; and for background noise data, normalization processing can be performed based on the reference noise spectrum.

[0020] Suppose the original sample value of a certain data source is Zero bias value The proportionality coefficient is The calibrated data can then be expressed as: in: Indicates the first Data sources at time The calibrated, standardized output value Indicates the first The scaling factor or sensitivity correction factor corresponding to each data source Indicates the first Data sources at time The original sampled values, Indicates the first Zero bias value of each data source, Indicates the data source number. Indicates the sampling time or data frame number.

[0021] For missing samples, when the missing proportion within a single frame is not higher than a preset threshold, linear interpolation, spline interpolation, or nearest neighbor interpolation can be used for repair. When the missing proportion within a single frame is higher than the preset threshold, the frame is marked as a low-completeness data frame for subsequent quality scoring. In this embodiment, the missing proportion threshold within a single frame is set to 5%, and linear interpolation is used to repair missing samples not exceeding this threshold. Using a 5% missing proportion threshold can ensure data availability while avoiding over-repair of severely missing data. The advantages of using linear interpolation are low computational cost, simple implementation, and good preservation of the continuity of slowly changing signals, making it suitable for real-time processing scenarios at edge nodes.

[0022] S103: Perform noise suppression and feature enhancement processing on the acoustic data, environmental data and background noise data in the calibrated multi-source data frame to obtain the enhanced multi-source data frame; In this embodiment, bandpass filtering, spectral subtraction denoising, and short-time Fourier transform enhancement are sequentially performed on the acoustic data to highlight the effective frequency bands relevant to the target. Moving average, Kalman smoothing, or median filtering can be performed on the environmental data to suppress abrupt noise. Long-term average noise spectrum estimation is performed on the background noise data to form a current sea area noise reference model. Specifically, a moving average with a window length of 5 sampling points is performed on the environmental data, and a long-term average noise spectrum estimation with a duration of 60 seconds is performed on the background noise data. The advantage of using bandpass filtering combined with spectral subtraction and short-time Fourier transform enhancement is that it can effectively suppress background noise while maintaining the integrity of the target's time-frequency characteristics. The advantage of using a 5-point moving average is that it can smooth environmental parameter fluctuations without significantly weakening the environmental change trend. The advantage of using a 60-second long-term average noise spectrum estimation is that it can obtain a more stable background noise reference model, thereby improving the robustness of subsequent denoising compensation.

[0023] Let the time-domain signal of the acoustic data be The background noise reference signal is The acoustic enhancement spectrum after spectral subtraction can be expressed as: in: Indicates frequency Acoustic enhancement spectrum after spectral subtraction processing Indicates the acoustic signal at frequency The frequency domain spectrum at that location, Indicates the background noise reference signal at frequency The frequency domain spectrum at that location, Indicates the noise suppression coefficient. Indicates frequency index.

[0024] In this embodiment, the noise suppression coefficient The setting is 0.8, which aims to achieve a better balance between noise suppression intensity and target feature fidelity, and to avoid introducing obvious target spectrum distortion due to excessive spectral reduction.

[0025] S104: Perform standardization processing on the enhanced multi-source data frames to construct a standardized multi-source data frame sequence; In this embodiment, the processed multi-source data is divided into frames according to a uniform frame length and frame shift to construct a standardized multi-source data frame. The standardized multi-source data frame includes at least the following fields: timestamp, node identifier, sensor identifier, acoustic feature vector, environmental feature vector, background noise feature vector, integrity identifier, and synchronization status identifier.

[0026] To meet the demands of streaming processing, standardized multi-source data frame sequences are stored using a first-in, first-out (FIFO) queue structure. Newly generated standardized multi-source data frames are pushed to the tail of the queue, and read operations are performed from the head of the queue. When the queue length reaches a preset limit, the oldest data frame is evicted in chronological order to ensure a balance between memory usage and real-time performance.

[0027] S200: Continuously reads standardized multi-source data frame sequences, performs local data fusion based on multi-source data quality scores and dynamic weights within each distributed computing node, and generates a local fusion result sequence; In this embodiment, multiple acquisition nodes within the same monitoring area can correspond to one edge computing node. The edge computing node performs local data fusion on standardized multi-source data frames within its area to reduce the communication pressure caused by large-scale backhaul of the original data and improve real-time processing capabilities.

[0028] Specifically, local data fusion does not apply fixed weights to each data source. Instead, it dynamically calculates the fusion weights based on the quality status of each data source in the current processing cycle. This results in higher-reliability data sources contributing more to the fusion results and lower-reliability data sources interfering less with the fusion results.

[0029] S201: Perform quality assessment processing on the multi-source data in the standardized multi-source data frame sequence within the same distributed computing node to calculate the data quality score; For each data source, the signal-to-noise ratio, completeness, timeliness, and environmental stability metrics are calculated separately. The specific calculation method is as follows: 1. Signal-to-noise ratio (SNR): This can be calculated based on the ratio of effective signal energy to background noise energy. In this embodiment, the average effective signal energy within a 5-second sliding time window is used. With average background noise energy Calculate the original signal-to-noise ratio and map it to the interval according to preset upper and lower limits. Its calculation method can be expressed as: in: Indicates the first The j-th data source among the nodes is at time j. The original signal-to-noise ratio, Indicates the first The j-th data source among the nodes is at time j. Average effective signal energy, Indicates the first The j-th data source among the nodes is at time j. Average background noise energy Indicates the distributed computing node number. j represents the data source number within the node. Indicates the processing cycle or data frame sequence number.

[0030] in: Indicates the first The j-th data source among the nodes is at time j. Normalized signal-to-noise ratio (SNR) Indicates the first The j-th data source among the nodes is at time j. The original signal-to-noise ratio, This represents the lower threshold for signal-to-noise ratio normalization. This represents the upper threshold for signal-to-noise ratio normalization. Indicates the distributed computing node number. j represents the data source number within the node. Indicates the processing cycle or data frame sequence number.

[0031] In this embodiment, the following settings are provided: , This is to highlight the distinction between high-quality and low-quality signals and facilitate subsequent unified weighted calculations. At the same time, the advantage of using a 5-second sliding time window is that it can balance short-term fluctuation suppression and dynamic response speed. 2. Completeness Index: This can be calculated based on the ratio of valid sampling points to the theoretical number of sampling points within the current processing cycle. The calculation method can be expressed as follows: in: Indicates the first The j-th data source among the nodes is at time j. Completeness index Indicates the first The j-th data source among the nodes is at time j. The number of effective sampling points, Indicates the first The j-th data source among the nodes is at time j. The theoretical number of sampling points, Indicates the distributed computing node number. j represents the data source number within the node. Indicates the processing cycle or data frame sequence number.

[0032] In this embodiment, when When the score is below 0.95, the corresponding data frame is marked as a data frame with decreased integrity. The advantage of setting the integrity threshold to 0.95 is that it can promptly identify sampling defects or transmission packet loss, while avoiding excessive reduction in data quality score due to a very small number of occasional lost samples.

[0033] 3. Timeliness Indicators: These can be calculated based on data arrival delays. Let the first... The j-th data source among the nodes is at time j. The current arrival delay is The maximum allowed latency is Then the timeliness index can be expressed as: in: Indicates the first The j-th data source among the nodes is at time j. Timeliness indicators Indicates the first The j-th data source among the nodes is at time j. The current arrival delay, Indicates the maximum allowable delay. Indicates the distributed computing node number. j represents the data source number within the node. Indicates the processing cycle or data frame sequence number.

[0034] In this embodiment, the following settings are provided: Setting the maximum allowable latency to 2 seconds has the advantage of balancing the tolerance for link fluctuations and the requirements for real-time processing in marine monitoring scenarios, thereby avoiding significant delays in the fusion results caused by outdated data.

[0035] 4. Environmental stability index: This can be calculated based on the degree of fluctuation of environmental parameters within a sliding window. Let the first... Each node at time... The normalized standard deviations of temperature, salinity, flow rate, and depth parameters over the most recent 60-second window are as follows: , , and The overall environmental fluctuation can then be expressed as: in: Indicates the first Each node at time... The overall environmental fluctuations Indicates the first Each node at time... The normalized standard deviation of the temperature parameter within the statistical window. Indicates the first Each node at time... The normalized standard deviation of the salinity parameter within the statistical window. Indicates the first Each node at time... The normalized standard deviation of the flow velocity parameter within the statistical window, Indicates the first Each node at time... The normalized standard deviation of the depth parameter within the statistical window. Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number. The four scaling factors of 0.30, 0.25, 0.25 and 0.20 are derived from engineering experience. In practical applications, these scaling factors can be dynamically adjusted through the configuration file according to specific circumstances.

[0036] in: Indicates the first The j-th data source among the nodes is at time j. Environmental stability indicators, Indicates the first Each node at time... The overall environmental fluctuations This represents the upper threshold for normalizing environmental fluctuations. Indicates the distributed computing node number. j represents the data source number within the node. Indicates the processing cycle or data frame sequence number.

[0037] In this embodiment, the following settings are provided: The advantage of using the most recent 60-second window to statistically analyze environmental fluctuations is that it can reflect the trend of sea state changes more stably; the advantage of using the above weighting method is that it pays more attention to temperature and salinity changes, thus better reflecting the impact of changes in the marine propagation environment on the quality of acoustic processing.

[0038] Let the j-th data source be at node. The signal-to-noise ratio corresponding to the current processing cycle is The completeness index is The timeliness indicator is The environmental stability index is Then the data quality score corresponding to this data source can be expressed as: And satisfy in: Indicates the first The j-th data source among the nodes is at time j. Data quality score, Indicates the first The j-th data source among the nodes is at time j. The signal-to-noise ratio (SNR) metric. Indicates the first The j-th data source among the nodes is at time j. Completeness index Indicates the first The j-th data source among the nodes is at time j. Timeliness indicators Indicates the first The j-th data source among the nodes is at time j. Environmental stability indicators, This represents the weighting coefficient corresponding to the signal-to-noise ratio (SNR) metric. This represents the weighting coefficient corresponding to the completeness index. This represents the weighting coefficient corresponding to the timeliness indicator. This represents the weighting coefficient corresponding to the environmental stability index. Indicates the distributed computing node number. j represents the data source number within the node. Indicates the processing cycle or data frame sequence number.

[0039] In one alternative implementation, it can be set , , , This prioritizes signal-to-noise ratio (SNR) and completeness. The advantage of using these values ​​is that they assign high weights to SNR and completeness metrics, ensuring the quality of data entering the fusion computation while preserving the ability to correct for dynamic scenarios based on timeliness and environmental stability, thus improving the stability and reliability of local fusion results.

[0040] S202: Perform weight calculation processing on the data quality score to obtain the local fusion weight; Normalization is performed on the data quality scores of each data source within the same node to obtain the local fusion weights: in: Indicates the first The j-th data source among the nodes is at time j. The contribution weight to the local fusion result, Indicates the first The j-th data source among the nodes is at time j. Data quality score, Indicates the first Each node at time... The sum of the quality scores of all participating data sources. This indicates the number of data sources participating in the local fusion of the current node. Indicates the distributed computing node number. j represents the data source number within the node. Indicates the processing cycle or data frame sequence number.

[0041] In this embodiment, a preset confidence threshold is used. The value is set to 0.6. When the quality score of a data source is lower than 0.6, its corresponding local fusion weight is reduced to 20% of its original weight, and a low-confidence warning is output to the self-diagnosis module. Setting it to 0.6 has the advantage of being able to identify obviously low-quality data sources earlier, while avoiding misjudging slight fluctuations as invalid data; setting the weight decay ratio to 20% has the advantage of significantly reducing the interference of abnormal data sources on the fusion results, while retaining a small amount of information to participate in the fusion to enhance the system's fault tolerance.

[0042] S203: Perform local fusion processing on multi-source data based on local fusion weights to obtain local fusion results; Local data fusion can be achieved using weighted Kalman filtering, extended Kalman filtering, or particle filtering, with weighted Kalman filtering being the preferred method. Its advantages include low computational complexity, suitability for real-time deployment at edge nodes, and the ability to easily map data quality scores to the observation noise covariance matrix, thereby enabling dynamic adjustment of the reliability of different data sources. Let the... The state prediction vector of each node is: The predicted covariance matrix is The joint observation vector composed of multiple data sources is The observation matrix is Then the local observation noise covariance matrix The weighted construction is performed based on the local fusion weights. Let: in: Indicates the first Each node at time... The local observation noise covariance matrix, The operator represents the diagonal matrix formed by the terms within the parentheses. Indicates the first The fundamental noise variance of the j-th data source in the n nodes Indicates the first The j-th data source among the nodes is at time j. Local fusion weights, This indicates the smallest positive number whose denominator is zero. This indicates the number of data sources participating in the local fusion of the current node. Indicates the distributed computing node number. j represents the data source number within the node. Indicates the processing cycle or data frame sequence number.

[0043] The weighted Kalman gain can be expressed as: in: Indicates the first Each node at time... The weighted Kalman gain matrix, Indicates the first Each node at time... The predicted covariance matrix, Indicates the first Each node at time... The observation matrix Represents the observation matrix The transpose of the matrix, Indicates the first Each node at time... The local observation noise covariance matrix, This represents the matrix inversion operation. Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number.

[0044] The local fusion state vector is updated as follows: in: Indicates the first Each node at time... Updated local fusion state vector, Indicates the first Each node at time... State prediction vector Indicates the first Each node at time... The weighted Kalman gain matrix, Indicates the first Each node at time... The joint observation vector, Indicates the first Each node at time... The observation matrix Indicates the observation residual or innovative term. Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number.

[0045] The local fusion covariance matrix is ​​updated as follows: in: Indicates the first Each node at time... The updated local fusion covariance matrix, This represents the identity matrix, which has the same dimension as the state. Indicates the first Each node at time... The weighted Kalman gain matrix, Indicates the first Each node at time... The observation matrix Indicates the first Each node at time... The predicted covariance matrix, Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number.

[0046] in: It can characterize the target acoustic intensity, environmental correction parameters, noise compensation parameters, and local event confidence levels within the current monitoring area.

[0047] S204: Perform time-series organization and output processing on the local fusion results to obtain a sequence of local fusion results; Each local fusion result includes at least: a timestamp, a node identifier, a local fusion state vector, a local fusion covariance matrix, a local event label, the current node load value, and a communication latency value. The local fusion results are written into a local fusion result sequence in chronological order and sent to the global fusion layer.

[0048] S300: Continuously receives the local fusion results output by each distributed computing node, performs spatiotemporal registration, credibility assessment and global data fusion on the local fusion results, and generates a global fusion result; The global fusion layer is used to map the local fusion results of different regions and nodes to a unified spatiotemporal reference framework, and form a global fusion result for the entire business area. It can be deployed on shore-based central servers or cloud fusion platforms with sufficient computing resources, large storage capacity and easy linkage with upper-level monitoring platforms and historical databases, for performing high-dimensional fusion computing, long-term data archiving and remote unified scheduling.

[0049] S301: Perform spatiotemporal registration processing on the local fusion result to obtain the spatiotemporally registered local fusion result; Since the different local fusion results come from different geographical locations and different points in time, spatiotemporal registration is required first. Let the... The geographical locations of the nodes are The timestamp of the local fusion result is Then, through the unified coordinate transformation function Map it to a unified geographic coordinate system: in: Indicates the first Each node at time... Spatial location mapped to a unified geographic coordinate system Represents a unified coordinate transformation function. Indicates the first Each node at time... The original geographic location vector, Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number.

[0050] Meanwhile, in this embodiment, the local fusion results of each node are aligned to the same fusion time by using a unified time window. The system employs linear interpolation to time-align local fusion results that are earlier or later than the current fusion window but with a deviation of no more than one fusion cycle. Results with a deviation exceeding one fusion cycle are directly discarded. The advantage of this value selection and processing method is that it ensures temporal continuity while avoiding the introduction of outdated data, thereby improving the real-time performance and accuracy of the global fusion results.

[0051] S302: Perform a credibility assessment on the local fusion result after spatiotemporal registration, and calculate the node credibility and global fusion weight; To improve the adaptability of the global fusion results to link fluctuations and node anomalies, it is necessary to assess the trustworthiness of each distributed computing node. Let the first node be... Each node at time... The historical error index is The link latency metric is The node load metric is The current data quality indicators are Then the node credibility can be expressed as: And satisfy in: Indicates the first Each node at time... The credibility of the nodes. Indicates the first Each node at time... Historical error index, Indicates the first Each node at time... Current data quality metrics Indicates the first Each node at time... Link latency metrics Indicates the first Each node at time... The node load metric, This represents the reliability coefficient corresponding to the historical error index. This represents the credibility coefficient corresponding to the current data quality indicator. This represents the credibility coefficient corresponding to the link latency metric. This represents the reliability coefficient corresponding to the node load metric. Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number.

[0052] In this embodiment, the following settings are provided: , , , The advantage of using the above values ​​is that it highlights the dominant role of current data quality in node reliability, while taking into account the impact of link status, node load, and historical errors on long-term stability, thus making the global fusion weights more consistent with actual business scenarios.

[0053] Normalize the node credibility to obtain the global fusion weight: in: Indicates the first Each node at time... Global fusion weights, Indicates the first Each node at time... The credibility of the nodes. This represents the sum of the trust levels of all distributed computing nodes participating in the current global fusion. This indicates the number of distributed computing nodes participating in the current global fusion. Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number.

[0054] S303: Based on the node credibility and global fusion weight, perform global data fusion processing on the local fusion result after spatiotemporal registration to obtain the global fusion result; Global Kalman filtering, information filtering, or Bayesian networks can be used for global data fusion. In this embodiment, information filtering is used for global data fusion, specifically a weighted information fusion method. The advantage of using information filtering is that it facilitates the fusion of local information matrices and information vectors uploaded from multiple nodes, is suitable for the superposition of multi-node results in distributed scenarios, and maintains good numerical stability and computational efficiency even as the number of nodes increases. Let the... The local fusion state vector output by each node is The local fusion covariance matrix is The global information matrix can then be represented as: in: Indicates time The global information matrix Indicates the first Each node at time... Global fusion weights, Indicates the first Local fusion covariance matrix of each node The inverse matrix, This indicates the number of distributed computing nodes participating in the current global fusion. Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number.

[0055] The global information vector can be represented as: in: Indicates time The global information vector, Indicates the first Each node at time... Global fusion weights, Indicates the first Local fusion covariance matrix of each node The inverse matrix, Indicates the first Each node at time... The local fusion state vector, This indicates the number of distributed computing nodes participating in the current global fusion. Indicates the distributed computing node number. Indicates the processing cycle or data frame sequence number.

[0056] Therefore, the global fusion state vector can be represented as: in: Indicates time The global fusion state vector, Indicates time Global information matrix The inverse matrix, Indicates time The global information vector, Indicates the processing cycle or data frame sequence number.

[0057] The global fusion covariance matrix can be expressed as: in: Indicates time The global fusion covariance matrix, Indicates time Global information matrix The inverse matrix, Indicates the processing cycle or data frame sequence number.

[0058] Based on global fusion state vector and global fusion covariance matrix It can further generate global uncertainty parameters and global event labels. The global event labels are used to identify whether there are events such as target appearance, abnormal noise enhancement, environmental changes or node failures in the current business area.

[0059] S304: Perform node anomaly suppression and fusion result stabilization processing on the global fusion result to obtain a stabilized global fusion result; When the credibility of a node is lower than a preset threshold for multiple consecutive periods When an anomaly occurs, the node is marked as a node to be isolated, and its participation weight is reduced in the global fusion process. Once the isolated node recovers, its fusion weight is gradually restored to avoid drastic fluctuations in the global fusion result caused by a single anomaly. In this embodiment, a preset threshold is used. Setting it to 0.5 has the advantage of effectively distinguishing between available nodes and low-reliability nodes, while the judgment condition for three consecutive cycles can avoid false isolation caused by single-cycle fluctuations, thereby improving system stability.

[0060] Furthermore, to avoid severe fluctuations in the global fusion results under complex sea conditions, this embodiment performs exponential smoothing on the global fusion state vector at consecutive time points: in: Indicates time The exponentially smoothed global fusion state vector This represents the smoothing coefficient of the global fused state vector at the current moment. Indicates time The original global fusion state vector, This represents the smoothed global fusion state vector from the previous time step. Indicates the processing cycle or data frame sequence number.

[0061] In this embodiment, the smoothing coefficient Set it to 0.7. Setting it to 0.7 has the advantage of prioritizing the retention of the current fusion result changes while using the previous result to suppress short-term fluctuations, thus balancing response speed and smooth stability.

[0062] S400: Continuously reads global fusion results, performs real-time signal analysis, target recognition, target tracking and status determination, and generates business processing results; Local fusion and global fusion mainly address the problem of collaborative representation of multi-source data, while the business analysis stage is used to transform the fusion results into business information that can directly serve marine monitoring, target detection, and mission handling.

[0063] S401: Perform real-time signal enhancement and time-frequency feature construction processing on the global fusion result to obtain the enhanced time-frequency features; Using a preprocessed 32kHz acoustic signal as input, real-time signal enhancement and time-frequency feature construction are performed based on the global fusion state vector. Specifically, S401-1: Perform bandpass filtering and spectral subtraction denoising on the acoustic signal in the global fusion result to obtain the denoised acoustic signal; First, the key analysis frequency band is determined. In this embodiment, the key analysis frequency band is set through a configuration file. When identifying an uncertain target category, the key analysis frequency band is preferably set to 200Hz to 8kHz; when identifying a ship or submarine target, the key analysis frequency band is preferably set to 200Hz to 3kHz; when identifying a marine mammal target, the key analysis frequency band is preferably set to 1kHz to 8kHz.

[0064] After determining the key analysis frequency band, bandpass filtering is performed on the input acoustic signal. In this embodiment, a fourth-order Butterworth bandpass filter is used, with its lower and upper cutoff frequencies set at the lower and upper boundaries of the key analysis frequency band, respectively.

[0065] Subsequently, spectral subtraction denoising is performed on the bandpass filtered signal. Specifically, the average power spectrum of the background noise is first estimated based on the background noise monitoring sensor within the most recent 60-second window. Then, the average power spectrum of the background noise is used to perform spectral subtraction on the short-time power spectrum of the current acoustic signal to suppress stable background noise components and reduce the interference of irrelevant frequency band noise on subsequent identification.

[0066] S401-2: Perform delay summation beamforming processing on the denoised acoustic signal to obtain a beam-enhanced single-channel acoustic signal; When multi-channel array observation is available, delayed summation beamforming is performed on the acoustic signal after bandpass filtering and spectral subtraction denoising. Specifically, the propagation delay of each channel is calculated based on the spatial position of the sensors in the array. Then, the signals of each channel are time-aligned according to their corresponding propagation delays. Finally, the time-aligned multi-channel signals are weighted and summed to obtain the beam-enhanced single-channel output signal. When only a single-channel input is available, no processing is performed, and the single-channel signal after bandpass filtering and spectral subtraction denoising is directly used as the beam-enhanced single-channel output signal.

[0067] S401-3: Perform time-frequency transformation processing on the beam-enhanced single-channel acoustic signal to generate a Log-Mel spectrum. First, the enhanced acoustic signal is processed by frame-by-frame windowing. The analysis window length is set to 1024 sampling points, the window shift is set to 320 sampling points, and the window function is Hamming window. Then, a short-time Fourier transform is performed on each frame of the signal to obtain the amplitude spectrum. The amplitude spectrum is mapped onto a 64-dimensional Mel filter bank, and then logarithmic compression is performed to generate a Log-Mel spectrum.

[0068] S402: Perform target recognition and target localization processing on the enhanced time-frequency features to obtain target category, target recognition confidence and target location; S402-1: Input the Log-Mel spectrogram into the pre-trained target recognition model to perform inference processing and obtain the embedded features; The Log-Mel spectrogram obtained through real-time signal enhancement and time-frequency feature construction is input into a pre-trained target recognition model. This model can employ a pre-trained audio representation model to identify target categories. In this embodiment, CNN14 from the PANNs pre-trained model is used as the base network, and its pre-trained weights trained on the AudioSet dataset are directly loaded to extract embedded features from the enhanced time-frequency features. The extracted embedded features are then matched with template features in a pre-defined target template library to generate matching scores for each target category. The PANNs pre-trained model has strong representation capabilities for general environmental sounds and non-speech audio events, and exhibits good distinguishing ability for ship noise, submersible radiated noise, biological sounds, and background interference sounds.

[0069] In another optional implementation, to further improve the recognition accuracy for specific underwater acoustic targets, the pre-trained PANNs model can be fine-tuned based on direct transfer. Specifically, the PANNs backbone network parameters can be retained, and the last classifier layer can be replaced with a 5-dimensional output layer. The 5-dimensional output corresponds to five target categories: underwater vehicles, ships, schools of fish, marine mammals, and invalid interference sources. Then, the replaced model is fine-tuned with all parameters using labeled underwater acoustic data. The advantage of this fine-tuning method is that it can better adapt the model to the fine-grained category differences in current marine acoustic operational scenarios while retaining the general representation capabilities of the pre-trained model.

[0070] S402-2: Perform similarity matching processing on the embedded features and template features in the preset target template library to obtain the matching score corresponding to each target category; Suppose that the pre-trained PANN model at time 1 The output embedded feature vector is The first preset target template in the library The template feature vector corresponding to the class target is ,in, , This represents the total number of target categories. (This refers to the embedded feature vector.) With each template feature vector Perform cosine similarity calculations separately to obtain the first... Original matching score corresponding to the target class Its calculation method can be expressed as: in: Indicates time For the The original matching score of the target class. Indicates time The embedded feature vector, Indicates the first The template feature vector corresponding to the target class. Indicates the target category index. Indicates the total number of target categories. Indicates the processing cycle or data frame sequence number.

[0071] To facilitate comparison of matching results across different categories, the original matching scores for each category are normalized to obtain the category matching score. Its calculation method can be expressed as: in: Indicates time For the Normalized matching score of class targets, Indicates exponentiation. Indicates time For the The original matching score of the target class. Indicates the target category index. Indicates the total number of target categories. Indicates the processing cycle or data frame sequence number.

[0072] S402-3: Perform category determination processing on the matching score to obtain the target category and target recognition confidence; The target category can be represented as: in: Indicates time Target category determination results This indicates the operation corresponding to the category index that maximizes the matching score. Indicates time For the Normalized matching score of class targets, Indicates the target category index. Indicates the total number of target categories. Indicates the processing cycle or data frame sequence number.

[0073] The target confidence level can be expressed as: in: Indicates time Target recognition confidence, This represents the operation of taking the maximum value among the normalized matching scores of each category. Indicates time For the Normalized matching score of class targets, Indicates the target category index. Indicates the total number of target categories. Indicates the processing cycle or data frame sequence number.

[0074] To improve the stability of template matching, in this embodiment, when Output category index when it is not lower than 0.80. The corresponding target category; when If the value is below 0.80, the current target will be marked as a target to be reviewed.

[0075] S402-4: Perform target localization algorithm processing on the multi-node synchronous target acoustic data corresponding to the target category to obtain the target location; For target localization, Time Difference of Arrival (TDOA) features, Frequency Difference of Arrival (FDOA) features, array beam output features, or matched field response features can be extracted from the multi-node synchronized target acoustic data corresponding to the target category determination results. Based on the extracted localization features, a target localization algorithm is executed to determine the target location. In this embodiment, the TDOA localization algorithm is used to determine the target location. Specifically, time synchronization and target segment alignment processing are performed on the same target acoustic signals received by multiple distributed sensor nodes, the time difference of arrival between each node is calculated, and the target location is solved based on the time difference of arrival.

[0076] S403: Perform target tracking and business status determination processing on the target category, target recognition confidence level and target location to generate business processing results; The system continuously reads target detection results from adjacent time points and performs trajectory association on continuous observations of the same target based on nearest neighbor association, joint probability data association, or multi-hypothesis tracking methods to obtain the target trajectory result. In this embodiment, the trajectory association is performed on continuous observations of the same target based on the nearest neighbor association method, and the consistency of the trajectory association result is verified by combining the target category and target confidence output by the PANNs pre-trained model, so as to achieve simplicity, low computational complexity, and suitability for real-time business processing scenarios. Let the current target trajectory state be... Then it includes at least position, speed, heading and trajectory continuity confidence.

[0077] The business status determination is based not only on the target trajectory results, but also on global event tags and a preset rule base. For example, when a high-confidence target is detected continuously entering a preset warning area, it can be determined as a key tracking status; when a drastic change in the environment causes a significant decrease in the target's confidence, it can be determined as an enhanced environmental interference status; when multiple nodes fail consecutively, it can be determined as a degraded system operation status.

[0078] The business processing results include at least: target category, target location, target trajectory, event label, fusion confidence, node health status, and current task suggestion identified based on the PANNs pre-trained model. These business processing results are output in real time through a graphical interface, console, or remote terminal, and displayed in the form of text, charts, trajectory lines, and status color indicators.

[0079] In one optional implementation, when the system determines that a target has entered a key area of ​​interest, a red target marker is displayed in the graphical interface and a decision prompt of "It is recommended to start continuous tracking" is output; when the system determines that environmental interference has increased, a decision prompt of "It is recommended to increase the sampling rate of relevant nodes or adjust the array beam pointing" is output; when the system determines that some nodes are overloaded, a scheduling prompt of "It is recommended to migrate some fusion tasks to backup nodes" is output.

[0080] S500: Outputs real-time feedback information, decision suggestions, and alarm information based on business processing results; In this embodiment, the feedback decision module continuously monitors the business processing results, node load status, and link quality status, and outputs real-time feedback information, decision suggestions, and alarm information according to the business processing results, so that operators and the upper system can synchronously obtain the current target status, environmental status, abnormal status, and resource status.

[0081] S501: Performs real-time feedback and decision suggestion generation processing on business processing results, and outputs real-time feedback information and decision suggestions; In this embodiment, the real-time feedback information includes at least the target category, target location, target trajectory, fusion confidence level, environmental parameters, node health status, link status, and current task status, and is displayed in real time through a graphical interface. Specifically, the target category is displayed using text labels, the target location using nautical chart coordinates and trajectory lines, the fusion confidence level using percentages, and environmental parameters using dashboards and trend curves. The advantage of this real-time feedback method is its clear information hierarchy and high degree of visualization, allowing operators to quickly grasp the overall business situation.

[0082] In this embodiment, the decision recommendations include at least continuous tracking recommendations, sampling rate adjustment recommendations, array beam adjustment recommendations, task migration recommendations, and anomaly handling recommendations. When the system determines that the target has entered a key area of ​​interest, it outputs a decision recommendation to "initiate continuous tracking"; when the system determines that environmental interference has increased, it outputs a decision recommendation to "increase the sampling rate of relevant nodes". Through the above method, complex analysis results can be directly converted into actionable prompts, thereby shortening the time for manual judgment and improving the efficiency of handling.

[0083] S502: Perform alarm judgment and alarm generation processing on the business processing results, and output alarm information; Preset alarm conditions may include: a target entering the warning zone, a continuous increase in target confidence level that meets the characteristics of a key target, abnormally increased background noise, and sudden changes in environmental parameters exceeding limits. Alarm methods may include interface flashing prompts, audible and visual alarms, SMS notifications, network alarm message pushes, or interface notifications connected to the higher-level command system.

[0084] To avoid false alarms caused by short-term, occasional fluctuations, a continuous judgment mechanism can be adopted in one optional implementation, with a continuous alarm cycle threshold set. If the same alarm event occurs consecutively If the alarm conditions are met continuously within a processing cycle, a formal alarm is triggered; otherwise, it is displayed as an early warning. This effectively filters false alarms caused by short-term disturbances while maintaining timely response to continuous abnormal events.

[0085] Example 2 is based on the same inventive concept as the business information processing method based on multi-source data fusion in the foregoing examples, such as... Figure 2 As shown, this application also provides a business information processing system based on multi-source data fusion, the system comprising: The multi-source data acquisition and preprocessing module 11 is used to continuously receive multi-source raw data frames sent by distributed acquisition nodes, and perform time synchronization, data calibration, noise suppression and standardization processing on the multi-source raw data frames to generate a standardized multi-source data frame sequence. Distributed local fusion module 12 is used to continuously read standardized multi-source data frame sequences, perform local data fusion based on multi-source data quality scores and dynamic weights in each distributed computing node, and generate a local fusion result sequence. The global fusion module 13 is used to continuously receive the local fusion results output by each distributed computing node, perform spatiotemporal registration, credibility assessment and global data fusion on the local fusion results, and generate a global fusion result. Business analysis module 14 is used to continuously read global fusion results, perform real-time signal analysis, target recognition, target tracking and status determination, and generate business processing results; Feedback decision module 15 is used to output real-time feedback information, decision suggestions, and alarm information based on the business processing results.

[0086] In a specific application scenario, multiple underwater sensor nodes and buoy nodes can be deployed in a certain sea area, with each area handled by an edge computing node. Each node locally fuses the collected acoustic data, environmental data, and background noise data to obtain local target features and environmental compensation parameters. The central fusion node further integrates the local fusion results from multiple areas to obtain the target's global position, trajectory, and environmental impact assessment results, which are then displayed in real time on a monitoring platform to assist operators in monitoring, tracking, early warning, and task scheduling.

[0087] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0088] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0089] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A business information processing method based on multi-source data fusion, characterized in that, The method includes: The system continuously receives multi-source raw data frames sent by distributed acquisition nodes, performs time synchronization, data calibration, noise suppression, and standardization processing on the multi-source raw data frames, and generates a standardized multi-source data frame sequence. Continuously read standardized multi-source data frame sequences, and perform local data fusion based on multi-source data quality scores and dynamic weights within each distributed computing node to generate a local fusion result sequence; The system continuously receives local fusion results from each distributed computing node, performs spatiotemporal registration, reliability assessment, and global data fusion on the local fusion results, and generates a global fusion result. Continuously read the global fusion results, perform real-time signal analysis, target recognition, target tracking and status determination, and generate business processing results; Based on the business processing results, output real-time feedback information, decision suggestions, and alarm information; The step of performing local data fusion based on multi-source data quality scores and dynamic weights within each distributed computing node to generate a sequence of local fusion results includes: For the multi-source data in the standardized multi-source data frame sequence, a quality assessment process is performed within the same distributed computing node, and the data quality score is composed of four weighted factors: signal-to-noise ratio, completeness, timeliness, and environmental stability. The data quality score is processed by weight calculation to obtain the local fusion weight; Local fusion processing is performed on multi-source data based on local fusion weights to obtain local fusion results; The local fusion results are subjected to temporal organization and output processing to obtain a sequence of local fusion results; The process of performing spatiotemporal registration, credibility assessment, and global data fusion on the local fusion results to generate global fusion results includes: Spatiotemporal registration is performed on the local fusion results to obtain spatiotemporally registered local fusion results; The credibility assessment process is performed on the local fusion results after spatiotemporal registration to calculate the node credibility and global fusion weight. Based on the node credibility and global fusion weight, global data fusion processing is performed on the local fusion result after spatiotemporal registration to obtain the global fusion result; The global fusion result is subjected to node anomaly suppression and fusion result stabilization processing to obtain a stabilized global fusion result.

2. The business information processing method based on multi-source data fusion as described in claim 1, characterized in that, The process of continuously receiving multi-source raw data frames sent by distributed acquisition nodes, and performing time synchronization, data calibration, noise suppression, and standardization on the multi-source raw data frames to generate a standardized multi-source data frame sequence includes: Perform time synchronization and rearrangement processing on the multi-source raw data frames to obtain time-aligned multi-source raw data frames; Data calibration, missing data repair, and dimensional unification are performed on the time-aligned multi-source raw data frames to obtain calibrated multi-source data frames. Noise suppression and feature enhancement processes are performed on the acoustic data, environmental data, and background noise data in the calibrated multi-source data frame, respectively, to obtain the enhanced multi-source data frame. The enhanced multi-source data frames are subjected to standardization processing to construct a standardized multi-source data frame sequence.

3. The business information processing method based on multi-source data fusion as described in claim 1, characterized in that, The process of performing real-time signal analysis, target recognition, target tracking, and status determination to generate business processing results includes: Real-time signal enhancement and time-frequency feature construction are performed on the global fusion results to obtain the enhanced time-frequency features; The enhanced time-frequency features are processed for target recognition and target localization to obtain the target category, target recognition confidence, and target location. Target tracking and business status determination are performed on the target category, target recognition confidence level, and target location to generate business processing results.

4. The business information processing method based on multi-source data fusion as described in claim 1, characterized in that, The output of real-time feedback information, decision suggestions, and alarm information based on business processing results includes: Perform real-time feedback and decision suggestion generation on business processing results, and output real-time feedback information and decision suggestions; Perform alarm judgment and alarm generation processing on the business processing results, and output alarm information.

5. The business information processing method based on multi-source data fusion as described in claim 3, characterized in that, The execution of real-time signal enhancement and time-frequency feature construction includes: Bandpass filtering and spectral subtraction denoising are performed on the acoustic signal in the global fusion result to obtain the denoised acoustic signal; The denoised acoustic signal is subjected to delay summation beamforming processing to obtain a beam-enhanced single-channel acoustic signal; The beam-enhanced single-channel acoustic signal is subjected to time-frequency transformation processing to generate a Log-Mel spectrum.

6. The business information processing method based on multi-source data fusion as described in claim 5, characterized in that, The target identification and target localization include: The Log-Mel spectrogram is input into the pre-trained target recognition model to perform inference processing and obtain the embedded features; The embedded features are matched with template features in a preset target template library to obtain matching scores for each target category. Perform category determination processing on the matching score to obtain the target category and target recognition confidence level; The target location is obtained by performing a target localization algorithm on the multi-node synchronous target acoustic data corresponding to the target category.

7. The business information processing method based on multi-source data fusion as described in claim 4, characterized in that, The alarm determination and alarm generation process includes: A continuous alarm cycle threshold is preset for alarm events, and continuous judgment processing is performed on the continuous satisfaction of alarm events to obtain continuous judgment results; For alarm events that continuously reach the preset continuous alarm cycle threshold for a number of processing cycles and continuously meet the alarm conditions, perform formal alarm generation processing to obtain formal alarm information; For alarm events that do not continuously reach the preset continuous alarm cycle threshold for a certain number of processing cycles but meet the alarm conditions, an early warning prompt generation process is performed to obtain early warning prompt information.

8. A business information processing system based on multi-source data fusion, characterized in that, include: A multi-source data acquisition and preprocessing module is used to continuously receive multi-source raw data frames sent by distributed acquisition nodes, and perform time synchronization, data calibration, noise suppression and standardization processing on the multi-source raw data frames to generate a standardized multi-source data frame sequence. A distributed local fusion module is used to continuously read standardized multi-source data frame sequences, perform local data fusion based on multi-source data quality scores and dynamic weights in each distributed computing node, and generate a local fusion result sequence. A global fusion module is used to continuously receive the local fusion results output by each distributed computing node, perform spatiotemporal registration, credibility assessment and global data fusion on the local fusion results, and generate a global fusion result. The business analysis module is used to continuously read the global fusion results, perform real-time signal analysis, target recognition, target tracking and status determination, and generate business processing results. The feedback decision module is used to output real-time feedback information, decision suggestions, and alarm information based on the business processing results. The step of performing local data fusion based on multi-source data quality scores and dynamic weights within each distributed computing node to generate a sequence of local fusion results includes: For the multi-source data in the standardized multi-source data frame sequence, a quality assessment process is performed within the same distributed computing node, and the data quality score is composed of four weighted factors: signal-to-noise ratio, completeness, timeliness, and environmental stability. The data quality score is processed by weight calculation to obtain the local fusion weight; Local fusion processing is performed on multi-source data based on local fusion weights to obtain local fusion results; The local fusion results are subjected to temporal organization and output processing to obtain a sequence of local fusion results; The process of performing spatiotemporal registration, credibility assessment, and global data fusion on the local fusion results to generate global fusion results includes: Spatiotemporal registration is performed on the local fusion results to obtain spatiotemporally registered local fusion results; The credibility assessment process is performed on the local fusion results after spatiotemporal registration to calculate the node credibility and global fusion weight. Based on the node credibility and global fusion weight, global data fusion processing is performed on the local fusion result after spatiotemporal registration to obtain the global fusion result; The global fusion result is subjected to node anomaly suppression and fusion result stabilization processing to obtain a stabilized global fusion result.

Images