A method and device for classifying vibration events of a ship hitting a bridge

Abstract

The present invention provides a method and device for classifying vibration events caused by ship collisions with bridge bodies, and relates to the field of bridge safety technology. The method comprises: obtaining vibration data of a ship collision with a bridge structure through a distributed optical fiber vibration sensing system, and constructing a spatiotemporal matrix of vibration signals; using a preprocessing method combining dynamic segmentation with a sliding window to generate frequency band subband signals; splicing the spatiotemporal matrix with the frequency band subband signals; extracting features through a deformable convolution layer; processing through a space-time mixed attention layer and a multi-scale feature aggregation layer to obtain fused features; converting the fused features into a vector of fixed length, and outputting the classification probability through a fully connected layer. The present invention provides a complete technical chain from perception to abnormal event classification for smart bridge safety protection, which can significantly improve the monitoring efficiency and emergency response capabilities of bridges against ship collisions, while providing data support for structural health monitoring and long-term maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of bridge safety, and in particular to a method and device for classifying vibration events of a ship hitting a bridge. Background Art

[0002] With the rapid development of global transportation networks, long bridges (such as sea-crossing bridges, suspension bridges, and cable-stayed bridges) have become critical infrastructure connecting regional economies and transportation, owing to their ability to span complex geographical environments. However, these bridges face numerous potential operational threats, among which ship strikes ("ship strikes") are a significant factor in causing sudden damage to bridge structures. Ship strikes typically involve high impact loads, short duration, and random locations of damage. These impacts can cause irreversible damage to parts of, or even the entire bridge structure, posing a serious threat to bridge safety and public safety.

[0003] Ship collisions with bridges can have serious consequences, including direct threats to bridge structural safety, traffic disruptions, damage to the aquatic environment and ecosystem, and even casualties and significant economic losses. For example, damage to a bridge's structure, whether partial or complete, can require lengthy repairs, impacting regional economic and social activities. Collapses can also have irreversible and catastrophic consequences. Therefore, timely identification and classification of ship collisions are crucial for ensuring bridge safety and reducing accident risks. However, traditional monitoring methods have numerous limitations. Regarding vibration measurement, traditional ship collision vibration measurement methods primarily rely on point sensors (such as accelerometers, strain gauges, or pressure sensors). These sensors have limited coverage and can only monitor localized areas of the bridge, resulting in blind spots for many potential risk points. For example, if a ship collision occurs at a pier or tower where no sensors are installed, traditional systems may be completely unaware of the collision, resulting in missed early warning opportunities. Furthermore, the spatial resolution of point sensors is limited by their deployment density, making it difficult to precisely pinpoint the collision's location. Consequently, ship collision identification relies on manual analysis or simple thresholding, which is susceptible to local noise interference, leading to false alarms or missed detections. Especially on large bridges, when sensors are spaced too far apart, it's impossible to distinguish vibration sources from adjacent structures, further reducing positioning accuracy. More importantly, these sensors fail to record the propagation of vibration along the bridge structure, a process crucial for event classification. This incomplete information limits subsequent analysis. Secondly, traditional methods lack environmental adaptability. Visual monitoring methods (such as cameras or radar) are susceptible to interference from weather conditions (such as fog, heavy rain, and at night) and lighting conditions, leading to monitoring interruptions or data distortion. They also cannot directly perceive the vibration response of the bridge's internal structure. Electronic sensors, when operated in marine or corrosive environments for extended periods, may frequently fail due to electromagnetic interference, device aging, or power failures. This not only increases maintenance costs but also compromises data continuity and reliability. For underwater structures, such as pier foundations, most traditional sensors are unable to operate stably underwater for extended periods, resulting in a lack of ability to monitor events such as underwater collisions and scrapes. Furthermore, bridge structures are constantly exposed to the elements and traffic loads, making vibration data susceptible to temperature drift, structural aging, and external interference. Traditional algorithms lack adaptive capabilities and struggle to maintain stable performance under dynamically changing operating conditions, resulting in insufficient system reliability. In terms of real-time performance, traditional sensors rely on manual or complex systems for data collection and transmission, making real-time vibration monitoring and early warning difficult. The vibration signals generated by ship impacts exhibit complex spatiotemporal characteristics in the time-frequency domain (such as high-frequency impact, energy concentration, and propagation trajectory), which can be overlooked or misjudged due to delayed response. Furthermore, they have poor adaptability to dynamic scenarios. For example, the transient signals of a ship impact can be masked by low-frequency background noise (such as wind and vehicle vibration). The limited dynamic range of traditional sensors makes it difficult to effectively separate target signals, resulting in reduced classification accuracy.From a cost perspective, a large number of point sensors would need to be deployed to cover the entire bridge, resulting in high initial installation and ongoing maintenance costs. Furthermore, the complex wiring system increases construction difficulty and safety risks, especially on large, complex bridges, where wiring can present significant technical challenges. In this context, vibration event classification is crucial. Vibration signatures are needed to distinguish ship collisions from other common vibration sources (such as vehicle traffic, wind, or mechanical failures). This allows for accurate early warnings, enabling rapid responses (such as evacuation, traffic control, or initiation of emergency plans) and accumulating data for long-term structural health assessments.

[0004] Existing methods also face multiple challenges in vibration event classification. Most classification processes rely on manual analysis, which is inefficient and susceptible to subjective judgment. Differences in operator experience can lead to misjudgments or missed judgments, especially in complex vibration signals, where it is difficult to quickly identify key features. Furthermore, ship collisions can be accompanied by multi-parameter changes (such as local strain mutations, temperature fluctuations, or structural damage). Traditional algorithms for extracting features from vibration signals are often limited to a single parameter (such as frequency or amplitude), ignoring multi-dimensional information such as spatial distribution and temporal patterns. This results in insufficient ability to distinguish similar events (such as ship anchor chain scrapes and small collisions). The lack of collaborative processing capabilities for multi-source data leads to insufficient basis for classification decisions. Traditional signal processing methods (such as wavelet transforms or spectral analysis) struggle to effectively capture and distinguish these features from other background vibrations (such as vehicle traffic, wind vibrations, temperature changes, or mechanical construction). Classification accuracy decreases significantly, especially when complex environmental noise is superimposed. At the same time, bridge vibration signals are often mixed with environmental noise (such as wind, vehicles, and temperature fluctuations). Traditional classification models are sensitive to noise and prone to false positives or ambiguous classifications. This is especially true in scenarios with multiple sources of interference, such as strong winds and dense ships. Signal complexity increases, making it difficult for algorithms to accurately separate target events. A lack of training data is also a core issue. Ship collisions occur infrequently, making it difficult to obtain sufficient real-world case data for model training, resulting in insufficient classification models' ability to identify rare events. Furthermore, existing models are often designed based on specific bridge or ship types and lack the ability to generalize to the vibration characteristics of different structural forms (such as cable-stayed and suspension bridges) or ships of varying sizes, limiting their universality. Vibration patterns vary significantly across different bridge structures, ship types, or impact angles. Traditional classification models require manual feature design and parameter tuning for specific scenarios, making them difficult to achieve versatility and scalability, increasing the complexity of system deployment and maintenance. Finally, traditional methods have limited ability to identify complex vibration patterns. For example, they cannot effectively distinguish between minor collisions and major impacts. Alternatively, existing damage to the bridge (such as cracks) can alter the vibration response characteristics, leading to the misinterpretation of abnormal vibrations caused by structural damage as ship collisions, thus compromising the credibility of the early warning system. These shortcomings collectively lead to existing technologies for ship collision monitoring suffering from delayed response, high false alarm rates, and difficult maintenance. This makes it difficult to meet the requirements of modern bridge safety monitoring for full coverage, high precision, low latency, and intelligent capabilities. Therefore, there is an urgent need to combine distributed fiber optic sensing technology with intelligent algorithms to overcome the limitations of traditional methods and achieve accurate identification and real-time early warning of ship collisions.

[0005] The core principle of a distributed fiber-optic vibration sensing system is to utilize the optical fiber itself as a distributed sensor. By emitting laser pulses and analyzing the phase changes in the return signal, vibration sensing is achieved along the entire length of the fiber (up to tens of kilometers). Its technical advantages include: full-area continuous monitoring, covering key locations such as bridge piers, decks, and towers, reducing blind spots; high sensitivity and interference immunity, enabling detection of minute vibrations (such as minor ship collisions or anchor chain scrapes), and immunity to electromagnetic interference, corrosion, and extreme weather conditions; significant cost-effectiveness, with low fiber deployment costs and no need for additional sensor nodes, making it suitable for large-scale infrastructure monitoring; and real-time and data continuity, providing a real-time vibration data stream for rapid analysis and classification, while also accumulating long-term data for structural health assessment. These characteristics enable the distributed fiber-optic vibration sensing system to capture the unique vibration characteristics of ship collisions, such as high-frequency, high-amplitude transient signals, as well as differences in spatial distribution and temporal patterns from other vibration sources (such as low-frequency vehicle vibration). For example, the spatial localization function of the distributed fiber-optic vibration sensing system can accurately identify the vibration source location (e.g., near a bridge pier) and propagation path, while its temporal signal may exhibit the pulsed characteristics of a shock wave. Furthermore, by combining machine learning algorithms (such as deep learning convolutional neural networks and long-short-term memory artificial neural networks, or traditional methods like support vector machines and random forests), distributed fiber-optic vibration sensing systems can further improve classification accuracy. By building a vibration signature library from historical data or simulations, the model can be trained to distinguish between ship collisions and other events. Distributed fiber-optic vibration sensing systems have the potential to identify abnormal vibrations within the entire structure of a long bridge and, in conjunction with other monitoring methods (such as cameras, radar, or water level sensors), enhance classification reliability.

[0006] Distributed fiber-optic vibration sensing systems, with their global coverage, high sensitivity, and real-time performance, address the shortcomings of traditional monitoring methods and provide a reliable technical path for the accurate classification of ship collision incidents. This technology not only significantly reduces accident risks, ensures bridge safety and traffic continuity, but also provides data support for structural maintenance decisions, representing a key innovation in the fields of smart transportation and infrastructure protection. With the continued advancement of fiber-optic sensing technology and the deepening of artificial intelligence algorithms, the application of distributed fiber-optic vibration sensing systems in bridge monitoring will become more mature, promoting the development of more intelligent and automated safety warning systems.

[0007] Existing methods for measuring ship-collision bridge vibrations and classifying vibration events have many defects, which seriously restrict the reliability and classification efficiency of the monitoring system in practical applications. At the same time, they also highlight the necessity of introducing more advanced sensing technologies (such as distributed fiber optic vibration sensing systems). Summary of the Invention

[0008] In order to solve the multiple challenges faced by traditional bridge vibration monitoring technology in the classification of ship collision events, including technical problems such as insufficient positioning accuracy, difficulty in feature extraction, sensitivity to environmental interference, poor real-time performance, low long-term stability, and weak multi-source data fusion capabilities, an embodiment of the present invention provides a method and device for classifying ship collision bridge body vibration events. The present invention obtains high-resolution, full-domain coverage vibration space-time matrix data through a distributed fiber optic sensing system, and then uses the automated feature learning and space-time correlation modeling capabilities of a deep learning algorithm to break through the limitations of traditional methods in positioning accuracy, feature extraction, environmental interference resistance, real-time response, long-term stability, and multi-parameter collaborative analysis, thereby constructing an intelligent, highly reliable, and adaptive ship collision event classification and early warning system, providing accurate, real-time, and all-weather technical support for bridge safety protection, while reducing manual dependence and operation and maintenance costs, and promoting the development of infrastructure monitoring towards automation and intelligence. The technical solution is as follows:

[0009] In one aspect, a method for classifying vibration events caused by a ship striking a bridge is provided. The method is implemented by a device for classifying vibration events caused by a ship striking a bridge, and the method comprises:

[0010] S1. Vibration data from ship-impact bridge structures is acquired through a distributed fiber-optic vibration sensing system installed on the bridge. A vibration signal space-time matrix is ​​constructed based on the vibration data. The vibration signal space-time matrix records the evolution of vibration waves over time through the time dimension, reflects the propagation path of vibration waves in the bridge structure through the spatial dimension, and characterizes the changes in vibration intensity at different locations on the bridge structure through the amplitude.

[0011] S2. Divide the vibration signal space-time matrix into sliding windows of preset length along the time axis, perform frequency band decomposition on the data in each window using dynamic wavelet transform to generate frequency band sub-band signals; concatenate the vibration signal space-time matrix with the frequency band sub-band signals to obtain an enhanced input tensor; wherein the frequency band sub-band signals include high-frequency sub-band signals and low-frequency sub-band signals.

[0012] S3. Feature extraction is performed on the enhanced input tensor through the deformable convolution layer to obtain local spatiotemporal features.

[0013] S4. Process the local spatiotemporal features through the space-time mixed attention layer to obtain weighted fused spatiotemporal features.

[0014] S5. Process the local spatiotemporal features and the weighted fused spatiotemporal features through a multi-scale feature aggregation layer to obtain fused features.

[0015] S6. Convert the fused features into a vector of fixed length, output the classification probability through the fully connected layer, and obtain the vibration event classification result of the ship hitting the bridge structure.

[0016] Optionally, feature extraction is performed on the enhanced input tensor in S3 to obtain local spatiotemporal features, including:

[0017] The spatial offset and temporal offset are calculated based on the enhanced input tensor. The sampling position of the convolution kernel of the deformable convolution layer is dynamically adjusted according to the spatial offset and temporal offset to extract local spatiotemporal features, as shown in the following formula (1):

[0018] (1)

[0019] Where, represents the local spatiotemporal characteristics, Indicates that the elements in the tensor are real numbers, Indicates the number of spatial point sampling, Indicates the number of samples at a time point, Indicates the number of channels, represents a two-dimensional deformable convolution, represents the augmented input tensor, Indicates the spatial offset, Indicates the time offset.

[0020] Optionally, the local spatiotemporal features are processed by the space-time hybrid attention layer in S4 to obtain weighted fused spatiotemporal features, including:

[0021] The local spatiotemporal features are input into the spatial attention branch of the spatial-temporal hybrid attention layer. The spatial channel features are obtained by performing global average pooling and maximum pooling on the spatial point sampling vectors of the local spatiotemporal features in the time dimension, and concatenating the global average pooling and maximum pooling results. Based on the spatial channel features, a spatial attention weight matrix is ​​generated through a fully connected layer and Softmax. The spatial feature weights of each channel are obtained according to the spatial attention weight matrix. All spatial features are weighted and concatenated to obtain spatial weighted features.

[0022] The local spatiotemporal features are input into the temporal attention branch of the spatial-temporal hybrid attention layer. The temporal channel features are obtained by performing global average pooling and maximum pooling on the time series of the local spatiotemporal features in the spatial dimension and concatenating the results of global average pooling and maximum pooling. According to the temporal channel features, the temporal attention weight matrix is ​​generated through the fully connected layer and Softmax. The temporal feature weight of each channel is obtained according to the temporal attention weight matrix. All temporal features are weighted and concatenated to obtain the temporal weighted features.

[0023] The spatial weighted features and the temporal weighted features are concatenated to obtain the spatiotemporal weighted features;

[0024] The gating weight is generated by the Sigmoid activation function, and the spatiotemporal weighted features are weightedly fused according to the gating weight to obtain the weighted fused spatiotemporal features.

[0025] Optionally, weighted spatiotemporal features are fused as shown in the following formula (2):

[0026] (2)

[0027] Where, represents the weighted fusion of spatiotemporal features, represents the activation function, represents the gating weight matrix, represents the spatiotemporal weighted features, Represents element-wise multiplication.

[0028] Optionally, the local spatiotemporal features and the weighted fused spatiotemporal features are processed in S5 to obtain fused features, including:

[0029] Perform maximum pooling on the local spatiotemporal features, with the pooling direction being the time dimension, to obtain the time pooling result.

[0030] The weighted fused spatiotemporal features are average pooled, with the pooling direction being the spatial dimension, to obtain the spatial pooling result.

[0031] The temporal pooling results and the spatial pooling results are concatenated to obtain multi-scale features.

[0032] Dynamic gating weights are generated through a learnable gating parameter matrix, and local spatiotemporal features and weighted fused spatiotemporal features are fused according to the dynamic gating weights to obtain fused features.

[0033] Optionally, the fused features are as shown in the following formula (3):

[0034] (3)

[0035] Where, represents the fused features, represents the dynamic gating weight, represents the local spatiotemporal characteristics, Represents weighted fusion of spatiotemporal features.

[0036] Optionally, the fixed-length vector in S6 is as shown in the following equation (4):

[0037] (4)

[0038] Where, represents a fixed-length vector, represents two-dimensional global average pooling, Represents the fused features.

[0039] The classification probability is shown in the following formula (5):

[0040] (5)

[0041] Where, represents the classification probability, represents the normalized exponential function, represents the classification weight matrix, Represents the bias vector.

[0042] On the other hand, a device for classifying vibration events of a ship striking a bridge is provided. The device is applied to a method for classifying vibration events of a ship striking a bridge. The device comprises:

[0043] The data acquisition module is used to obtain vibration data of the ship-impact bridge structure through a distributed fiber-optic vibration sensing system laid on the bridge, and to construct a vibration signal space-time matrix based on the vibration data. The vibration signal space-time matrix records the evolution of the vibration wave over time through the time dimension, reflects the propagation path of the vibration wave in the bridge structure through the spatial dimension, and characterizes the changes in vibration intensity at different locations of the bridge structure through the amplitude.

[0044] The data preprocessing module is used to divide the vibration signal space-time matrix into sliding windows of preset length along the time axis, perform frequency band decomposition on the data in each window using dynamic wavelet transform to generate frequency band sub-band signals; splice the vibration signal space-time matrix with the frequency band sub-band signals to obtain an enhanced input tensor; wherein the frequency band sub-band signals include high-frequency sub-band signals and low-frequency sub-band signals.

[0045] The deformable convolution module is used to extract features from the enhanced input tensor through the deformable convolution layer to obtain local spatiotemporal features.

[0046] The spatial-temporal hybrid attention module is used to process local spatiotemporal features through the spatial-temporal hybrid attention layer to obtain weighted fused spatiotemporal features.

[0047] The multi-scale feature aggregation module is used to process the local spatiotemporal features and the weighted fused spatiotemporal features through the multi-scale feature aggregation layer to obtain the fused features.

[0048] The classification and output module is used to convert the fused features into a vector of fixed length, output the classification probability through the fully connected layer, and obtain the vibration event classification result of the ship hitting the bridge structure.

[0049] On the other hand, a ship-collision-bridge vibration event classification device is provided, which includes: a processor; a memory, wherein computer-readable instructions are stored on the memory, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned ship-collision-bridge vibration event classification methods is implemented.

[0050] On the other hand, a computer-readable storage medium is provided, wherein the storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned ship collision bridge body vibration event classification methods.

[0051] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0052] This innovative technical solution, combining a distributed fiber-optic vibration sensing system with a deep learning algorithm, achieves intelligent classification of ship-bridge collision vibration events by building a high-precision spatiotemporal sensing framework. The core idea behind this approach is to utilize the distributed fiber-optic vibration sensing system to capture the spatiotemporal distribution characteristics of vibration signals from ship collisions with bridge structures in wide waters. This approach, combined with multi-dimensional feature extraction and joint modeling using a deep learning model, addresses the limitations of traditional methods in classification accuracy and real-time performance under complex conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 This is a flow chart of a method for classifying vibration events of a ship striking a bridge provided by an embodiment of the present invention;

[0055] Figure 2 This is a block diagram of a device for classifying vibration events of a ship striking a bridge provided by an embodiment of the present invention;

[0056] Figure 3 It is a structural schematic diagram of a ship collision bridge vibration event classification device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0058] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0059] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0060] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0061] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0062] The embodiment of the present invention provides a method for classifying vibration events of a ship hitting a bridge. The method can be implemented by a device for classifying vibration events of a ship hitting a bridge. The device can be a terminal or a server. Figure 1 The flow chart of the ship-on-bridge vibration event classification method is shown in FIG. The processing flow of the method may include the following steps:

[0063] S1. Vibration data of the ship-impact bridge structure is obtained through a distributed fiber-optic vibration sensing system laid on the bridge, and a vibration signal spatiotemporal matrix is ​​constructed based on the vibration data.

[0064] Among them, the vibration signal space-time matrix records the evolution of the vibration wave over time through the time dimension, reflects the propagation path of the vibration wave in the bridge structure through the spatial dimension, and represents the changes in vibration intensity at different positions of the bridge structure through the amplitude.

[0065] In one feasible implementation, a distributed fiber-optic vibration sensing system is installed along key bridge locations (such as piers, towers, and decks) using phase optical time-domain reflectometry (PTDR) technology based on coherent heterodyne detection to form a high-density vibration monitoring network. The distributed fiber-optic vibration sensing system achieves a spatial resolution of meters and a sampling frequency exceeding 10 kHz, enabling real-time capture of vibration wave propagation along the fiber. By acquiring amplitude data corresponding to spatial locations and time series, a spatiotemporal matrix of vibration signals is constructed. The time dimension of this spatiotemporal matrix records the temporal evolution of the vibration wave, the spatial dimension reflects the propagation path of the vibration wave within the bridge structure, and the amplitude characterizes the variation in vibration intensity at different locations. When a ship strikes a pier, high-frequency vibration waves (such as shock waves) propagate upward along the fiber to the tower, while low-frequency structural responses diffuse to the bridge deck, resulting in a unique spatiotemporal amplitude distribution. By constructing this spatiotemporal matrix along the fiber, the system not only preserves the spatial distribution of vibration events but also captures the dynamic characteristics of the event through the continuity of the time series, providing a multi-dimensional physical basis for subsequent classification.

[0066] S2. Divide the vibration signal space-time matrix into sliding windows of preset length along the time axis, perform frequency band decomposition on the data in each window using dynamic wavelet transform to generate frequency band sub-band signals; concatenate the vibration signal space-time matrix with the frequency band sub-band signals to obtain an enhanced input tensor.

[0067] The frequency band sub-band signals include high frequency band sub-band signals and low frequency band sub-band signals.

[0068] In one feasible implementation, a preprocessing method combining dynamic segmentation and sliding window is used to address the high noise and complex interference issues of the original vibration signal. First, the space-time matrix is ​​divided into sliding windows of fixed length along the time axis to ensure the continuity of event features. For each signal in the window (space-time matrix, three-dimensional tensor ,in is the number of spatial point samples, is the number of time point samples, 1 is a single channel, preferably, is the number of spatial point samples of 100, The spatial point sampling vector (the number of time samples is 400) uses the dynamic wavelet transform (Continuous Wavelet Transform, CWT) to perform frequency band decomposition to generate frequency band subband signals , Indicates the number of frequency band decompositions. The high-frequency sub-band corresponds to the transient shock wave generated by the ship collision, while the low-frequency sub-band corresponds to the long-term variation of the structural response. The center frequency of the wavelet transform is dynamically adjusted through the frequency domain-based self-attention module (FSA). and bandwidth The center frequency of the wavelet transform It can be expressed as:

[0069] (1)

[0070] in, is a learnable attention weight matrix, It is The original spatial point sampling vibration signal of time intervals. Bandwidth It can be expressed as:

[0071] (2)

[0072] in is the learnable parameter weight matrix, Represents the activation function. According to the attention weight matrix , before dynamic selection Key frequency bands. Preferably, the wavelet transform is decomposed into 8 frequency bands. In order to enhance the data, the original signal and frequency band sub-band signals Splicing to form the enhanced input tensor , perform Z-score normalization on each channel. Preferably, High-frequency subband signals can capture the instantaneous burst of energy at a specific spatial location during a ship collision, while low-frequency subband signals can reflect the vibration mode changes in the bridge's overall structure caused by the impact. This multi-scale fusion not only preserves information about the spatial propagation path of vibration waves, but also enhances the ability to distinguish between different interference sources (such as vehicle loads, wind vibration, and temperature changes) through the complementary characteristics of the frequency and time domains.

[0073] S3. Feature extraction is performed on the enhanced input tensor through the deformable convolution layer to obtain local spatiotemporal features.

[0074] In a feasible implementation, the core advantage of Deformable Convolutional (DC) over traditional convolution is that it breaks the fixed sampling position and receptive field in traditional convolution operations by introducing learnable offsets, thereby significantly improving the model's adaptability to complex geometric shapes and multi-scale targets. Deformable convolution also improves efficiency by reducing computational redundancy. The fixed grid of traditional convolution may lead to unnecessary sampling in the background or irrelevant areas, while deformable convolution can dynamically adjust the sampling points according to the position of the target, concentrating computing resources on key areas. In terms of performance, deformable convolution significantly improves the effect of the model in complex scenes by giving the network the ability to model geometric deformations. Generate spatial offsets through a deformable offset network and time offset . Spatial offset It can be expressed as:

[0075] (3)

[0076] Time offset It can be expressed as:

[0077] (4)

[0078] in and is a learnable parameter matrix. The sampling position of the convolution kernel is dynamically adjusted according to the offset to extract local spatiotemporal features:

[0079] (5)

[0080] Where, represents the local spatiotemporal characteristics, Indicates that the elements in the tensor are real numbers, Indicates the number of spatial point sampling, Indicates the number of samples at a time point, Indicates the number of channels, represents a two-dimensional deformable convolution, represents the augmented input tensor, Indicates the spatial offset, Indicates the time offset.

[0081] Preferably, the convolution kernel size is 3×3, the stride is 1, and the padding is the same. Feature tensor The key frequency band is retained. Preferably, the number of output channels is 64. .

[0082] S4. Process the local spatiotemporal features through the space-time mixed attention layer to obtain weighted fused spatiotemporal features.

[0083] In one feasible implementation, the Space-Time Hybrid Attention (STHA) layer is a deep learning module that combines the spatial and temporal dimension attention mechanisms. It is mainly used to process data with spatiotemporal characteristics (such as video, action recognition, spatiotemporal sequences, etc.). Its core idea is to improve the model's ability to capture complex spatiotemporal information by simultaneously modeling the spatial features within the same frame (spatial attention) and the temporal relationship between different frames (temporal attention). The feature tensor processed by the deformable convolutional network is converted into Processing is divided into two parallel channels (spatial channel and temporal channel).

[0084] For the spatial attention branch, global average pooling and maximum pooling are performed on the spatial point sampling vector in the time dimension, and channel features are generated after splicing. :

[0085] (6)

[0086] Preferably, Generate spatial attention weight matrix through full connection layer and Softmax , which can be expressed as:

[0087] (7)

[0088] Preferably, ; Represents a learnable weight matrix, which is used to Projection to Query space; Represents a learnable weight matrix, which is used to Projection to Key space; Represents the feature dimension of Query and Key; Represents a learnable weight matrix, which is used to Projected into the Value space.

[0089] The spatial feature weight of each channel can be expressed as:

[0090] (8)

[0091] Preferred Then the spatial features are fused and the spatial features of all time series are spliced ​​into , preferred .

[0092] Furthermore, for the temporal attention branch, global average pooling and maximum pooling are performed on the time series in the spatial dimension, and channel features are generated after splicing. :

[0093] (9)

[0094] Preferably, Generate the time attention weight matrix through the fully connected layer and Softmax , which can be expressed as:

[0095] (10)

[0096] Preferred . Represents a learnable weight matrix, which is used to Projection to Query space; Represents a learnable weight matrix, which is used to Projection to Key space; Represents the feature dimension of Query and Key; Represents a learnable weight matrix, which is used to Projected into the Value space.

[0097] The temporal feature weight of each channel can be expressed as:

[0098] (11)

[0099] Then the time features are fused and the time features of all spatial point sampling intervals are spliced ​​into .

[0100] Preferred Then the spatial features are fused and the time series of all spatial point sampling vectors are spliced ​​into , preferred .

[0101] After obtaining the spatial weighted features and time weighted features, the spatiotemporal weighted features are concatenated. The spatiotemporal weighted features can be expressed as:

[0102] (12)

[0103] Preferred . Generate gate weights through Sigmoid activation function , weighted fusion of spatiotemporal features:

[0104] (13)

[0105] Where, represents the weighted fusion of spatiotemporal features, represents the activation function, represents the gating weight matrix, represents the spatiotemporal weighted features, Represents element-wise multiplication.

[0106] S5. Process the local spatiotemporal features and the weighted fused spatiotemporal features through a multi-scale feature aggregation layer to obtain fused features.

[0107] In one feasible implementation, the Multi-Scale Feature Aggregation (MSFA) layer effectively fuses features from different frequency bands or scales to enhance the model's ability to represent multi-scale information. This layer enables the model to simultaneously focus on global structure (low-frequency) and local details (high-frequency), thereby improving its ability to capture complex textures or patterns. By aggregating multi-scale features, the MSFA layer enhances the model's robustness to different timescales and frequency components in the input signal.

[0108] Deformable convolutional layer output and the spatial-temporal mixed attention layer output All In order to extract multi-scale features of the time dimension through time-direction pooling, Perform max pooling:

[0109] (14)

[0110] The pooling direction is the time dimension. Preferably, the pooling kernel is , the window size in the time direction is 3, there is no change in the spatial direction, and the filling is the same. .

[0111] In order to extract multi-scale features of spatial dimensions through pooling in spatial directions, Perform average pooling:

[0112] (15)

[0113] The pooling direction is the spatial dimension. Preferably, the pooling kernel is , the window size in the spatial direction is 3, the time direction does not change, and the filling is the same. .

[0114] The spatial and temporal pooling results are concatenated into multi-scale features, which can be expressed as :

[0115] (16)

[0116] Preferably, .

[0117] Through the learnable gating parameter matrix Generating dynamic gating weights , dynamic weight It can be expressed as:

[0118] (17)

[0119] Then the deformable convolution layer output and the spatial-temporal mixed attention layer output Fusion, the final fusion variable It can be expressed as:

[0120] (18)

[0121] Preferably, .

[0122] Where, represents the fused features, represents the dynamic gating weight, represents the local spatiotemporal characteristics, Represents weighted fusion of spatiotemporal features.

[0123] S6. Convert the fused features into a vector of fixed length, output the classification probability through the fully connected layer, and obtain the vibration event classification result of the ship hitting the bridge structure.

[0124] In a feasible implementation, the fused features Convert to a fixed-length vector , that is, compressing the spatial dimension and time dimension to 1. A vector of fixed length It can be expressed as:

[0125] (19)

[0126] Preferably, .

[0127] Where, represents a fixed-length vector, represents two-dimensional global average pooling, Represents the fused features.

[0128] Output classification probability through the fully connected layer:

[0129] (20)

[0130] in , . Classification probability vector ,in is the number of categories.

[0131] Where, represents the classification probability, represents the normalized exponential function, represents the classification weight matrix, Represents the bias vector.

[0132] The key technical point of the present invention lies in its high-precision perception and classification capabilities for vibration signals of ship collision events. Its core innovation is achieved through the following irreplaceable technical features: First, a distributed fiber optic vibration sensing system is used as the basis for data acquisition to achieve spatially continuous, high-resolution, real-time monitoring of vibration signals of the bridge body. Without the use of a distributed fiber optic vibration sensing system, traditional point sensors (such as accelerometers) cannot capture the spatial distribution characteristics of vibration waves propagating along the bridge (such as path and intensity changes), resulting in a lack of key information in the classification model, making it difficult to distinguish between ship collisions and other vibration sources (such as vehicle traffic or wind vibration). Secondly, the invention proposes a joint spatiotemporal feature extraction method that couples the time series data of the vibration signal with the spatial position information to construct a multi-dimensional feature vector. If only the features of a single dimension of time or space are extracted, the model cannot learn the spatiotemporal coupling laws unique to ship collision events, and the classification accuracy will drop significantly.

[0133] In terms of deep learning architecture design, this paper employs a deep learning algorithm based on deformable convolution and a hybrid spatial-temporal attention mechanism to address the complexity of different bridge structures and ship collision scenarios. Using a single neural network would be incapable of simultaneously processing the spatial distribution and temporal dynamics of vibration signals, resulting in insufficient adaptability to diverse vibration patterns.

[0134] A synergistic architecture combining a distributed fiber-optic vibration sensing system and deep learning: This system combines the fiber-path vibration data from the distributed fiber-optic vibration sensing system with the automatic feature learning capabilities of deep learning to form an end-to-end data flow design, encompassing the entire process from continuous spatial and temporal acquisition to feature extraction, model training, and real-time classification. This design addresses the problem of traditional methods relying on manual feature extraction and insufficient classification accuracy. Without the distributed fiber-optic vibration sensing system's continuous spatial and temporal acquisition capabilities or the adaptive feature learning capabilities of deep learning, the system would be unable to effectively identify ship collision events.

[0135] A joint spatiotemporal feature modeling approach: By mapping the time series and spatial coordinates of vibration signals into a three-dimensional tensor and designing a deep learning network structure adapted to this feature, we can accurately model the spatiotemporal coupling of ship collision events. Without joint spatiotemporal feature modeling, the model will fail to capture the physical properties of vibration wave propagation, resulting in classification failure.

[0136] In summary, the present invention obtains high-spatial-resolution vibration data along the optical fiber through a distributed optical fiber vibration sensing system. Combined with spatiotemporal joint feature modeling and a hybrid deep learning architecture, it solves the problems of missing spatial information and insufficient model generalization ability in vibration event classification in traditional methods, and realizes real-time, high-precision classification in complex environments.

[0137] The application of a distributed fiber-optic vibration sensing system in the classification of ship-to-bridge vibration events has significantly improved the overall effectiveness of monitoring and early warning. Through continuous monitoring with high spatial resolution, the system can locate vibration sources with meter-level or even sub-meter-level accuracy, precisely distinguishing ship impacts from other common vibration sources such as vehicle traffic, wind vibration, and temperature fluctuations. This avoids misjudgments caused by sparse distribution or ambiguous positioning of traditional sensors. Its global coverage allows fiber optic cables to be laid only in key bridge areas for comprehensive monitoring. Compared to the multi-point deployment limitations of traditional point sensors such as accelerometers, this reduces hardware costs while ensuring comprehensive data, providing more complete vibration signature information for event classification. Furthermore, the system boasts real-time data acquisition capabilities, capturing transient vibration signals from ship collisions with millisecond-level response. Combining traditional machine learning or deep learning algorithms, it dynamically analyzes the time-frequency characteristics of vibration (such as frequency, amplitude, and duration), rapidly identifying dangerous events and triggering early warnings, saving valuable time for emergency response and reducing false alarms and missed alerts. Because fiber optic sensors are insensitive to electromagnetic interference and corrosion-resistant, the system can operate stably and continuously in complex electromagnetic environments or harsh conditions such as humidity and salt spray. Frequent calibration or maintenance is unnecessary, ensuring data continuity and reliability, enabling long-term health monitoring and identifying potential structural damage or abnormal event patterns. Its surface-mount deployment also avoids damage to the bridge structure while remaining highly concealed, minimizing the risk of human interference. By integrating vibration data with other monitoring methods (such as cameras, radar, and water level sensors), the system not only assists in optimizing waterway management to reduce the probability of accidental ship collisions, but also provides critical information such as vibration energy and impact location for accident investigations, aiding in the development of targeted reinforcement plans. Importantly, the distributed fiber optic vibration sensing system reduces reliance on manual inspections through its AI-based classification and alarm mechanism, making it particularly suitable for the intelligent operation and maintenance of large or remote bridges. Its electromagnetic interference resistance and optical signal transmission ensure data security and integrity. With its advantages of full-area coverage, high resolution, real-time performance, environmental adaptability, and multi-parameter collaborative analysis, the distributed fiber-optic vibration sensing system not only achieves efficient classification and early warning of ship collision incidents, but also provides comprehensive data support for bridge safety protection, structural health assessment, and traffic environment management, becoming an important innovative means of intelligent infrastructure monitoring.

[0138] Deep learning algorithms, processing the spatiotemporal matrices acquired by distributed fiber-optic vibration sensing systems, have improved the classification of ship-to-bridge vibration events. Their core advantage lies in their ability to efficiently analyze the spatiotemporal correlations of complex vibration data, significantly enhancing classification accuracy and intelligence. Using a continuously laid fiber-optic network, the distributed fiber-optic vibration sensing system captures vibration signals from the surface or interior of bridge structures with high spatial and temporal resolution, generating a three-dimensional spatiotemporal matrix containing spatial location, time series, and vibration intensity. While this multidimensional data structure is rich in information, traditional machine learning methods struggle to effectively extract its complex features. Deep learning algorithms, however, can use automated feature learning to identify unique spatiotemporal patterns in the data from ship strikes and other vibration sources (such as vehicle traffic, wind vibration, temperature fluctuations, or mechanical construction). For example, ship strikes typically manifest as localized, high-frequency, high-amplitude transient vibrations with concentrated energy and transient duration, whereas vehicle traffic exhibits periodic, low-frequency vibrations that propagate along the bridge deck. Deep learning, through its multi-layered network architecture, can simultaneously capture the spatial distribution characteristics of vibration signals (such as localized strain changes at the impact site) and temporal dynamic characteristics (such as the propagation trajectory and attenuation of shock waves), automatically distinguishing these patterns and significantly reducing false positives and false negatives. Furthermore, the introduction of deep learning algorithms enables the system to adapt to complex vibration scenarios in diverse environments. Because bridge vibration data can be affected by environmental noise (such as wind, water flow, and temperature fluctuations) and background interference (such as passing vehicles and subway operations), deep learning, through end-to-end training, can extract key vibration features from this noise. It can even continuously optimize the model through transfer learning or incremental learning to account for data variations across seasons, weather conditions, or traffic volume. For example, the vibration energy of a ship collision can vary depending on the size of the ship, the angle of impact, or the material of the bridge structure. Deep learning models can learn these subtle differences using large amounts of annotated data, achieving more robust classification capabilities. Furthermore, deep learning can also combine multi-parameter data (such as the coordinated changes in vibration signals with temperature and strain) for joint analysis, further improving classification reliability. For example, a ship collision may be accompanied by a localized temperature increase (due to friction or structural deformation) or a sudden change in strain. This multi-dimensional information can be integrated and processed through a deep learning network to enhance the judgment of the event type. The combination of deep learning algorithms and distributed fiber optic systems also demonstrates significant advantages in terms of real-time performance. Because distributed fiber optic vibration sensing systems can generate spatiotemporal matrix data in real time with millisecond-level accuracy, deep learning models can achieve real-time online classification and early warning through lightweight design or edge computing deployment. This immediate response capability provides a more efficient protection mechanism for bridge safety, quickly triggering alarms in the event of a ship collision, buying time for emergency response.At the same time, the scalability of deep learning enables it to adapt to bridge structures of different sizes, whether it is a long-span cable-stayed bridge or a short-span beam bridge. The model performance can be optimized by simply adjusting the training dataset without the need for complex parameter tuning.

[0139] Deep learning algorithms can mine changes in the health status of bridge structures from long-term monitoring data. Ship collisions can cause hidden damage to bridges, and subsequent vibration patterns will gradually change due to structural damage. Deep learning models can identify these subtle changes through time series analysis, not only classifying current events but also predicting structural degradation trends to assist in preventive maintenance. By combining historical vibration data with real-time event classification results, deep learning models can determine whether a region has fatigue damage due to multiple impacts or long-term vibration loads, thereby providing dynamic support for bridge safety assessments. In addition, the adaptability of deep learning enables it to continuously learn new vibration patterns, such as vibration characteristics under different ship types or impact angles, thereby continuously improving the generalization ability of the classification system and adapting to complex and changing actual working conditions.

[0140] Overall, the deep learning algorithm's processing of the spatiotemporal matrix of a distributed fiber-optic vibration sensing system not only overcomes the limitations of traditional methods in feature extraction and pattern recognition, but also significantly improves the effectiveness of ship collision classification through automation, high precision, and real-time performance. It transforms the spatiotemporal information of vibration data into interpretable event types, while supporting structural health monitoring and traffic environment analysis. This provides an intelligent and comprehensive solution for bridge safety protection, and promotes the development of infrastructure monitoring technology towards a higher level of automation and intelligence.

[0141] In this embodiment of the present invention, the innovative technical solution of a distributed fiber-optic vibration sensing system and a deep learning algorithm achieves intelligent classification of ship-bridge collision vibration events by building a high-precision spatiotemporal perception framework. The core concept of this solution is to use the distributed fiber-optic vibration sensing system to obtain the spatiotemporal distribution characteristics of vibration signals from ship collisions with bridge structures in wide waters. This solution then uses a deep learning model to extract multi-dimensional features and conduct joint modeling to address the limitations of traditional methods in classification accuracy and real-time performance under complex working conditions.

[0142] Figure 2 This is a block diagram of a device for classifying vibration events caused by a ship hitting a bridge according to an exemplary embodiment. The device is used in a method for classifying vibration events caused by a ship hitting a bridge. Figure 2 The device includes a data acquisition module 310, a data preprocessing module 320, a deformable convolution module 330, a spatial-temporal hybrid attention module 340, a multi-scale feature aggregation module 350, and a classification and output module 360.

[0143] The data acquisition module 310 is used to obtain vibration data of the ship-impact bridge structure through a distributed fiber-optic vibration sensing system laid on the bridge, and to construct a vibration signal space-time matrix based on the vibration data; wherein the vibration signal space-time matrix records the evolution of the vibration wave over time through the time dimension, reflects the propagation path of the vibration wave in the bridge structure through the spatial dimension, and represents the change in vibration intensity at different positions of the bridge structure through the amplitude.

[0144] The data preprocessing module 320 is used to divide the vibration signal space-time matrix into sliding windows of preset length along the time axis, perform frequency band decomposition on the data in each window using dynamic wavelet transform to generate frequency band subband signals; and concatenate the vibration signal space-time matrix with the frequency band subband signals to obtain an enhanced input tensor; wherein the frequency band subband signals include high-frequency subband signals and low-frequency subband signals.

[0145] The deformable convolution module 330 is used to extract features from the enhanced input tensor through a deformable convolution layer to obtain local spatiotemporal features.

[0146] The space-time hybrid attention module 340 is used to process the local space-time features through the space-time hybrid attention layer to obtain weighted fused space-time features.

[0147] The multi-scale feature aggregation module 350 is used to process the local spatiotemporal features and the weighted fused spatiotemporal features through a multi-scale feature aggregation layer to obtain fused features.

[0148] The classification and output module 360 ​​is used to convert the fused features into a vector of fixed length, output the classification probability through the fully connected layer, and obtain the vibration event classification result of the ship hitting the bridge structure.

[0149] In this embodiment of the present invention, the innovative technical solution of a distributed fiber-optic vibration sensing system and a deep learning algorithm achieves intelligent classification of ship-bridge collision vibration events by building a high-precision spatiotemporal perception framework. The core concept of this solution is to use the distributed fiber-optic vibration sensing system to obtain the spatiotemporal distribution characteristics of vibration signals from ship collisions with bridge structures in wide waters. This solution then uses a deep learning model to extract multi-dimensional features and conduct joint modeling to address the limitations of traditional methods in classification accuracy and real-time performance under complex working conditions.

[0150] Figure 3 FIG. 1 is a structural diagram of a device for classifying vibration events of a ship-collision bridge provided by an embodiment of the present invention. Figure 3 As shown, the ship collision bridge vibration event classification device may include the above Figure 2 Optionally, the ship-collision-bridge vibration event classification device 410 may include a first processor 2001 .

[0151] Optionally, the ship-collision-bridge vibration event classification device 410 may further include a memory 2002 and a transceiver 2003 .

[0152] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0153] The following combination Figure 3 The components of the ship-on-bridge vibration event classification device 410 are described in detail:

[0154] The first processor 2001 is the control center of the ship-collision-bridge vibration event classification device 410 and can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs) or one or more field programmable gate arrays (FPGAs).

[0155] Optionally, the first processor 2001 may execute various functions of the ship-collision-with-bridge vibration event classification device 410 by running or executing a software program stored in the memory 2002 and calling data stored in the memory 2002 .

[0156] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 3 CPU0 and CPU1 are shown in FIG.

[0157] In a specific implementation, as an embodiment, the ship collision bridge vibration event classification device 410 may also include multiple processors, such as Figure 3 1 and 2. The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). A processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0158] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled by the first processor 2001 for execution. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0159] Alternatively, the memory 2002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and capable of being accessed by a computer, but not limited thereto. The memory 2002 may be integrated with the first processor 2001 or exist independently and accessed through the interface circuit ( Figure 3 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0160] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0161] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 3 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0162] Optionally, the transceiver 2003 may be integrated with the first processor 2001 or may exist independently and receive the vibration event of a ship hitting a bridge through the interface circuit ( Figure 3 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0163] It should be noted that Figure 3 The structure of the ship-collision-bridge vibration event classification device 410 shown in the figure does not constitute a limitation on the router. The actual knowledge structure recognition device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0164] In addition, the technical effects of the ship-collision-bridge vibration event classification device 410 can refer to the technical effects of the ship-collision-bridge vibration event classification method described in the above method embodiment, and will not be repeated here.

[0165] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, etc.

[0166] It should also be understood that the memory in the embodiments of the present invention may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0167] The above embodiments can be implemented in whole or in part via software, hardware (e.g., circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0168] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0169] In this disclosure, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0170] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0171] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0172] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0173] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of the device or unit, which can be electrical, mechanical or other forms.

[0174] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0175] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0176] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical disks.

[0177] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for classifying vibration events caused by a ship hitting a bridge, characterized in that: The method comprises: S1. Vibration data from a ship impacting the bridge structure is acquired using a distributed fiber-optic vibration sensing system installed on the bridge. A vibration signal space-time matrix is ​​constructed based on the vibration data. The vibration signal space-time matrix records the evolution of vibration waves over time using the time dimension, reflects the propagation path of vibration waves in the bridge structure using the spatial dimension, and characterizes the changes in vibration intensity at different locations on the bridge structure using amplitude. S2. Dividing the vibration signal spatiotemporal matrix into sliding windows of preset length along the time axis, performing frequency band decomposition on the data in each window using a dynamic wavelet transform to generate frequency band subband signals; concatenating the vibration signal spatiotemporal matrix with the frequency band subband signals to obtain an enhanced input tensor; wherein the frequency band subband signals include high-frequency subband signals and low-frequency subband signals; S3. performing feature extraction on the enhanced input tensor through a deformable convolutional layer to obtain local spatiotemporal features; S4. Processing the local spatiotemporal features through a space-time hybrid attention layer to obtain weighted fused spatiotemporal features; S5. Processing the local spatiotemporal features and the weighted fused spatiotemporal features through a multi-scale feature aggregation layer to obtain fused features; S6. Convert the fused features into a vector of fixed length, output the classification probability through a fully connected layer, and obtain a vibration event classification result of the ship hitting the bridge structure; The step of processing the local spatiotemporal features and the weighted fused spatiotemporal features in S5 to obtain fused features includes: Performing maximum pooling on the local spatiotemporal features, with the pooling direction being the time dimension, to obtain a time pooling result; Performing average pooling on the weighted fused spatiotemporal features, with the pooling direction being the spatial dimension, to obtain a spatial pooling result; splicing the temporal pooling result and the spatial pooling result to obtain multi-scale features; Dynamic gating weights are generated through a learnable gating parameter matrix, and the local spatiotemporal features and the weighted fused spatiotemporal features are fused according to the dynamic gating weights to obtain fused features.

2. The ship collision bridge vibration event classification method according to claim 1 is characterized in that: The step S3 extracts features from the enhanced input tensor to obtain local spatiotemporal features, including: The spatial offset and the temporal offset are calculated according to the enhanced input tensor; the sampling position of the convolution kernel of the deformable convolution layer is dynamically adjusted according to the spatial offset and the temporal offset to extract the local spatiotemporal features, as shown in the following formula (1): （1） Where, represents the local spatiotemporal characteristics, Indicates that the elements in the tensor are real numbers, Indicates the number of spatial point sampling, Indicates the number of samples at a time point, Indicates the number of channels, represents a two-dimensional deformable convolution, represents the augmented input tensor, Indicates the spatial offset, Indicates the time offset.

3. The ship collision bridge vibration event classification method according to claim 1 is characterized in that: The step S4 processes the local spatiotemporal features through a space-time hybrid attention layer to obtain weighted fused spatiotemporal features, including: The local spatiotemporal features are input into the spatial attention branch of the spatial-temporal hybrid attention layer, and the spatial channel features are obtained by performing global average pooling and maximum pooling on the spatial point sampling vectors of the local spatiotemporal features in the time dimension, and concatenating the global average pooling and maximum pooling results; based on the spatial channel features, a spatial attention weight matrix is ​​generated through a fully connected layer and Softmax; the spatial feature weights of each channel are obtained according to the spatial attention weight matrix; all spatial features are weighted and concatenated to obtain spatial weighted features; The local spatiotemporal features are input into the temporal attention branch of the space-time hybrid attention layer, and the temporal channel features are obtained by performing global average pooling and maximum pooling on the time series of the local spatiotemporal features in the spatial dimension and concatenating the global average pooling and maximum pooling results; based on the temporal channel features, a temporal attention weight matrix is ​​generated through a fully connected layer and Softmax; the temporal feature weight of each channel is obtained according to the temporal attention weight matrix; all temporal features are weighted and concatenated to obtain a temporal weighted feature; Concatenating the spatial weighted features and the temporal weighted features to obtain spatiotemporal weighted features; The gating weight is generated by the Sigmoid activation function, and the spatiotemporal weighted features are weightedly fused according to the gating weight to obtain the weighted fused spatiotemporal features.

4. The ship collision bridge vibration event classification method according to claim 3 is characterized in that: The weighted fusion spatiotemporal features are shown in the following formula (2): （2） Where, represents the weighted fusion of spatiotemporal features, represents the activation function, represents the gating weight matrix, represents the spatiotemporal weighted features, Represents element-wise multiplication.

5. The method for classifying vibration events caused by a ship collision with a bridge according to claim 1, characterized in that: The fused features are shown in the following formula (3): （3） Where, represents the fused features, represents the dynamic gating weight, represents the local spatiotemporal characteristics, Represents weighted fusion of spatiotemporal features.

6. The ship collision bridge vibration event classification method according to claim 1 is characterized in that: The fixed-length vector in S6 is as shown in the following formula (4): （4） Where, represents a fixed-length vector, represents two-dimensional global average pooling, Represents the fused features; The classification probability is shown in the following formula (5): （5） Where, represents the classification probability, represents the normalized exponential function, represents the classification weight matrix, Represents the bias vector.

7. A device for classifying vibration events caused by a ship hitting a bridge, the device for classifying vibration events caused by a ship hitting a bridge being used to implement the method for classifying vibration events caused by a ship hitting a bridge being as claimed in any one of claims 1 to 6, characterized in that: The device comprises: A data acquisition module is configured to acquire vibration data from a ship impacting the bridge structure using a distributed optical fiber vibration sensing system installed on the bridge, and to construct a vibration signal spatiotemporal matrix based on the vibration data. The spatiotemporal matrix records the evolution of vibration waves over time using the time dimension, reflects the propagation path of vibration waves in the bridge structure using the spatial dimension, and characterizes the changes in vibration intensity at different locations on the bridge structure using amplitude. a data preprocessing module configured to divide the vibration signal spatiotemporal matrix into sliding windows of preset length along the time axis, perform frequency band decomposition on the data within each window using a dynamic wavelet transform to generate frequency band subband signals; and concatenate the vibration signal spatiotemporal matrix with the frequency band subband signals to obtain an enhanced input tensor; wherein the frequency band subband signals include high-frequency subband signals and low-frequency subband signals; A deformable convolution module, configured to extract features from the enhanced input tensor through a deformable convolution layer to obtain local spatiotemporal features; A space-time hybrid attention module is used to process the local space-time features through a space-time hybrid attention layer to obtain weighted fused space-time features; A multi-scale feature aggregation module, configured to process the local spatiotemporal features and the weighted fused spatiotemporal features through a multi-scale feature aggregation layer to obtain fused features; The classification and output module is used to convert the fused features into a vector of fixed length, output the classification probability through a fully connected layer, and obtain the vibration event classification result of the ship hitting the bridge structure.

8. A device for classifying vibration events caused by a ship hitting a bridge, characterized in that: The ship-collision bridge vibration event classification device includes: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 6.

Images