A method and system for detecting external damage to optical cables

By dividing the optical cable into spatial sensing channels and generating multi-band spatiotemporal color images, combined with an improved YOLO network and a spatiotemporal convolutional attention module, the problems of missed detection and false detection in optical cable external damage event detection are solved, and high-precision optical cable external damage event detection is achieved.

CN123122647APending Publication Date: 2026-11-17STATE GRID ZHEJIANG ELECTRIC POWER CO LTD QUZHOU POWER SUPPLY CO
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611300446.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-26
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing optical cable external damage event detection methods based on one-dimensional signal analysis have high false alarm and high missed detection rates under complex environmental noise conditions, cannot achieve accurate concurrent multi-dimensional positioning and classification, and lose spatiotemporal two-dimensional information of the event.

Method used

The optical cable is divided into several spatial sensing channels. The vibration signal energy matrix is ​​obtained through multi-scale wavelet packet decomposition, generating a multi-band spatiotemporal color image. An improved YOLO network is used for event detection, and a spatiotemporal convolutional attention module is combined to enhance feature extraction.

Benefits of technology

It effectively distinguishes between real external damage events and environmental noise, improving detection accuracy and robustness, and achieving high-precision and reliable detection of optical cable external damage events.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122647A_ABST
    Figure CN123122647A_ABST
Patent Text Reader

Abstract

This invention relates to the field of power communication optical cable safety monitoring technology, and in particular to a method and system for detecting external damage events in optical cables. The invention divides the optical cable to be tested into several spatial sensing channels, acquiring the original one-dimensional vibration signal corresponding to each spatial sensing channel; it then performs multi-scale wavelet packet decomposition on the original one-dimensional vibration signal of each spatial sensing channel to obtain sub-signals for each spatial sensing channel in different frequency bands; for each frequency band, it calculates the energy sequence corresponding to the sub-signals of each spatial sensing channel within a set detection time window; based on the energy sequences of all spatial sensing channels in the same frequency band, it constructs an energy matrix for the corresponding frequency band; and based on the energy matrices of the sub-signals of each frequency band, it generates a multi-frequency band spatiotemporal color image to obtain the detection result of external damage events in the optical cable. This invention effectively improves the accuracy of detecting external damage events in optical cables.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power communication optical cable safety monitoring technology, and in particular to a method and system for detecting external damage events to optical cables. Background Technology

[0002] Optical cables for power communication are crucial carriers supporting core power services such as power grid dispatch automation, relay protection, and stability control. Their operational stability directly impacts the overall safe and stable operation of the power grid. During their long-term service, external forces such as construction excavation, mechanical crushing, and impacts are the primary causes of cable line faults and communication interruptions, easily triggering safety risks such as abnormal power data transmission and relay protection failures. Therefore, achieving all-weather, high-precision, and low-missibility intelligent detection of external damage events along power optical cables is of significant engineering value and practical importance for mitigating cable faults and ensuring the safe and reliable operation of the power communication network.

[0003] Currently, monitoring technologies for external force damage to power optical cables are mainly divided into three categories: manual inspection and video surveillance, fiber optic quasi-distributed sensing, and phase-sensitive optical time-domain reflectometry (OTDR) sensing. Manual inspection relies on on-site manual checks, making continuous 24 / 7 monitoring impossible, resulting in high labor costs and low efficiency. Video surveillance is susceptible to interference from weather conditions such as sunlight, rain, snow, and fog, limiting its coverage and making it unsuitable for monitoring the entire length of long-distance power transmission optical cables. Fiber optic quasi-distributed sensing technology can only collect signals from preset discrete monitoring points, leaving numerous blind spots along the cable route. Achieving full coverage monitoring over long distances requires densely deploying a large number of sensing points, significantly increasing equipment and construction costs, resulting in poor engineering practicality and economy.

[0004] Compared to the two technologies mentioned above, phase-sensitive optical time-domain reflectometry (φ-OTDR) sensing technology, with its advantage of continuous distributed monitoring along the entire line, has become the mainstream technical solution for monitoring external damage to power optical cables. This technology can rely on the existing communication optical cable itself to achieve blind-spot-free, continuous vibration signal acquisition over tens of kilometers of line, without the need for additional sensing equipment. It can acquire vibration data at various spatial locations along the optical cable around the clock, effectively compensating for the shortcomings of manual inspection, video surveillance, and grating point sensing, and is suitable for online monitoring scenarios of long-distance power optical cables. Current mainstream monitoring solutions mostly rely on one-dimensional vibration time-series signals acquired by φ-OTDR for external damage event identification and analysis.

[0005] However, existing analysis and identification methods based on one-dimensional vibration signals still have certain shortcomings. These methods only perform independent, point-by-point analysis and processing of single-point one-dimensional time-series vibration signals from each spatial sensing channel of the optical cable, completely losing the spatiotemporal two-dimensional correlation information of the external force damage event. They sever the spatial distribution correlation and temporal evolution characteristics of the external damage event, failing to fully preserve the inherent spatiotemporal overall texture of the external damage event, resulting in weak perception of the complete event. In complex noise scenarios such as wind and rain interference and environmental noise, the single one-dimensional signal feature dimension is limited, with low identification and weak anti-interference ability, making it difficult to effectively distinguish between real external damage events and random environmental noise. It is prone to missed detections and false detections, and cannot meet the high-precision and high-reliability routine monitoring requirements of power optical cable external damage events under complex working conditions. Summary of the Invention

[0006] Therefore, the technical problem to be solved by the present invention is to overcome the shortcomings of existing optical cable external damage event detection methods based on one-dimensional signal analysis, which suffer from high false alarm rates and high missed detection rates under complex environmental noise due to the loss of two-dimensional spatiotemporal information of the event, and are unable to achieve accurate concurrent multi-dimensional positioning and classification.

[0007] To address the aforementioned technical problems, this invention provides a method for detecting external damage to optical cables, comprising: The optical cable under test is divided into several spatial sensing channels, and the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable under test is obtained. Multi-scale wavelet packet decomposition was performed on the original one-dimensional vibration signals of each spatial sensing channel of the optical cable to be tested to obtain sub-signals of each spatial sensing channel in different frequency bands. For each frequency band, calculate the energy sequence of the sub-signal corresponding to each spatial sensing channel within the set detection time window in that frequency band; Based on the energy sequences of all spatial sensing channels in the same frequency band, an energy matrix for the corresponding frequency band is constructed. Each row in the energy matrix represents a spatial sensing channel, and each column represents a time sampling point within a set detection time window. The matrix elements are the vibration energy of the sub-signal of the corresponding frequency band under the corresponding time sampling point and the corresponding spatial sensing channel. Based on the energy matrix of sub-signals in each frequency band, a multi-band spatiotemporal color image is generated; Based on multi-band spatiotemporal color images, the detection results of optical cable external damage events are obtained.

[0008] Preferably, the method for obtaining the original one-dimensional vibration signal corresponding to each spatial sensing channel to be detected includes: Based on a set acquisition frequency, a detection light pulse is emitted to the optical cable under test; The phase or amplitude change of the Rayleigh backscattered light interference signal returned by the optical cable under test is received and demodulated to obtain the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable under test.

[0009] Preferably, the method of transmitting probe light pulses to the optical cable under test based on a set acquisition frequency includes: Using a phase-sensitive optical time-domain reflectometer, probe light pulses are emitted toward the optical cable under test based on a set acquisition frequency.

[0010] Preferably, the method for generating a multi-band spatiotemporal color image based on the energy matrix of sub-signals in each frequency band includes: The sum of all elements in the energy matrix of the sub-signals of each frequency band is used as the total energy of each frequency band. Based on the total energy, the energy matrices of the sub-signals in each frequency band are sorted in descending order. The top three energy matrices after sorting are then mapped to grayscale images and used as the R, G, and B channels of the multi-frequency band spatiotemporal color image, respectively, to synthesize the multi-frequency band spatiotemporal color image.

[0011] Preferably, the detection results of optical cable external damage events include: the location of the target event on the optical cable to be detected, the start and duration, the spatial impact range, the event category, and the confidence level.

[0012] Preferably, the method for obtaining the detection results of optical cable external damage events based on multi-band spatiotemporal color images includes: By passing the multi-band spatiotemporal color image through the optical cable breach event detection network, the bounding box coordinates and class probabilities of all target events in the multi-band spatiotemporal color image are obtained. The bounding box coordinates of all target events in the multi-band spatiotemporal color image are mapped to obtain the detection results of optical cable external damage events.

[0013] Preferably, the optical cable external damage event detection network is an improved YOLO network, which is obtained by setting a spatiotemporal convolutional attention module after the backbone network of the YOLO network.

[0014] Preferably, the method for processing the features extracted from the backbone network through a spatiotemporal convolutional attention module to obtain spatiotemporal convolutional attention features includes: The features extracted by the backbone network are subjected to global average pooling and global max pooling along the spatial dimension, respectively. The pooling results are then concatenated and spatial orientation attention weights are generated by a one-dimensional convolutional kernel. The features extracted by the backbone network are subjected to global average pooling and global max pooling along the time dimension, respectively. The pooling results are then concatenated and used to generate temporal attention weights through a one-dimensional convolutional kernel. After fusing the spatial and temporal attention weights through an outer product, the features extracted by the backbone network are weighted to obtain spatiotemporal convolutional attention features.

[0015] Preferably, the improved YOLO network further includes: replacing convolutions in the neck network with deformable convolutions to obtain the target improved YOLO network, wherein the selection method includes: By sequentially replacing convolutions at different locations within the neck network with deformable convolutions, multiple candidate improved YOLO networks are constructed, and each candidate improved YOLO network is iteratively trained using the same training dataset. By comparing the detection accuracy, recall, and robustness metrics of each candidate improved YOLO network, the convolutions in the neck network to be replaced with deformable convolutions are determined.

[0016] The present invention also provides a system for detecting external damage to optical cables, comprising: The data acquisition module is used to divide the optical cable under test into several spatial sensing channels and acquire the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable under test. The decomposition module is used to perform multi-scale wavelet packet decomposition on the original one-dimensional vibration signals of each spatial sensing channel of the optical cable to be tested, so as to obtain the sub-signals of each spatial sensing channel under different frequency bands. The energy sequence acquisition module is used to calculate the energy sequence of each spatial sensing channel within a set detection time window for each frequency band. The energy matrix construction module is used to construct an energy matrix for a corresponding frequency band based on the energy sequences of all spatial sensing channels in the same frequency band. Each row in the energy matrix represents a spatial sensing channel, and each column represents a time sampling point within a set detection time window. The matrix elements are the vibration energy of the sub-signal of the corresponding frequency band under the corresponding time sampling point and the corresponding spatial sensing channel. The image generation module is used to generate multi-band spatiotemporal color images based on the energy matrix of sub-signals in each frequency band; The detection module is used to obtain the detection results of optical cable external damage events based on multi-band spatiotemporal color images.

[0017] The optical cable external damage detection method and system provided in this application have the following beneficial effects: This invention divides the optical cable to be tested into several spatial sensing channels. The optical cable under test is divided into several spatial sensing channels, and the original one-dimensional vibration signal of each channel is collected. The sub-signals of different frequency bands of each channel are obtained by multi-scale wavelet packet decomposition. The energy sequence of the sub-signals of each spatial sensing channel in each frequency band within a set detection time window is further calculated. Based on the energy sequence of all spatial sensing channels in the same frequency band, a frequency band energy matrix is ​​constructed with time sampling points as rows and spatial sensing channels as columns. Then, a multi-frequency band spatiotemporal color image is generated from each frequency band energy matrix. In this way, the originally independent one-dimensional temporal vibration signal is converted into a two-dimensional image representation that integrates time evolution, spatial distribution and multi-frequency band vibration energy information. This fully restores the inherent spatiotemporal overall texture of the optical cable external damage event, no longer separating the spatial dimension distribution correlation and the temporal dimension continuous evolution characteristics of the event, enriching the feature representation dimensions, improving the feature recognition of external damage events, effectively distinguishing real external damage events from random environmental noise caused by wind, rain and environmental noise, improving the problem of missed detection and false detection under complex working conditions, and realizing the effective detection of external damage events of power optical cables.

[0018] Furthermore, the improved YOLO network proposed in this invention sets up a spatiotemporal convolutional attention module after the YOLO network backbone to adaptively mine feature weights from both spatial and temporal dimensions. First, the output features of the backbone network are subjected to global average pooling and global max pooling in the spatial dimension, respectively. After concatenation, spatial attention weights are generated through one-dimensional convolution to accurately capture the continuous distribution features of external damage events in the optical cable spatial channel. Simultaneously, the features are subjected to dual pooling in the temporal dimension, and temporal attention weights are generated through one-dimensional convolution to effectively extract the continuous evolution features of external damage events in the temporal dimension. On this basis, the spatial and temporal attention weights are deeply fused through outer product operations. Finally, the fused spatiotemporal attention weights are used to adaptively weight and correct the original features. This design fully leverages the prior features of spatially continuous and temporally sustained strip-like morphology in spatiotemporal images of external damage events in power fiber optic cables. This distinguishes it from the irregular and spatiotemporally uncorrelated characteristics of random noise such as wind, rain, and environmental vibrations. It enhances the effective feature response of real external damage events while adaptively suppressing interference from cluttered environmental noise and invalid background features, improving the feature recognition and anti-interference capabilities of the damaged target. This further ensures the accuracy and reliability of fiber optic cable external damage detection under complex operating conditions. Furthermore, convolutions at different locations within the neck network are replaced with deformable convolutions. These deformable convolutions adaptively adapt to the geometric deformation of the external damage texture under different operating conditions, accurately identifying external damage target events in images and effectively improving the detection accuracy and robustness of fiber optic cable external force damage events in complex noise environments. Attached Figure Description

[0019] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the steps of a method for detecting external damage to optical cables according to the present invention. Figure 2 To improve the structure diagram of the YOLO network; Figure 3 This is a structural diagram of the spatiotemporal convolutional attention module; Figure 4 This is a structural diagram of an optical cable external damage detection system according to the present invention. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0021] Please see Figure 1 This embodiment provides a method for detecting optical cable external damage events, including: Step S1: Divide the optical cable to be tested into N spatial sensing channels and acquire the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable to be tested; where N is the number of spatial sensing channels; the spatial sensing channel is an independent sensing segment obtained by spatially dividing along the axial direction of the optical cable to be tested, and each spatial sensing channel uniquely corresponds to the spatial position of a fixed length of optical fiber on the optical cable; each spatial sensing channel acquires a one-dimensional vibration time sequence signal that changes with time, which characterizes the external vibration disturbance information at the location of the optical cable segment.

[0022] In this embodiment, specifically, the method for obtaining the original one-dimensional vibration signal corresponding to each spatial sensing channel to be detected includes: Using a phase-sensitive optical time-domain reflectometer (φ-OTDR), probe light pulses are emitted toward the optical cable under test based on a set acquisition frequency; The phase or amplitude change of the Rayleigh backscattered light interference signal returned by the optical cable under test is received and demodulated to obtain the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable under test.

[0023] The original one-dimensional vibration signals corresponding to each spatial sensing channel of the optical cable under test are summarized to obtain a time series dataset of the original one-dimensional vibration signals consisting of N spatial sensing channels within the monitoring line length L and duration T.

[0024] Step S2: Perform multi-scale wavelet packet decomposition on the original one-dimensional vibration signals of each spatial sensing channel of the optical cable to be tested to obtain sub-signals of each spatial sensing channel in K frequency bands; where K is the total number of frequency bands. In this embodiment, optionally, the multi-scale wavelet packet decomposition adopts 4-layer wavelet packet decomposition to obtain sub-signals of each spatial sensing channel under 16 frequency bands.

[0025] Step S3: For each frequency band, calculate the energy sequence of the sub-signal corresponding to each spatial sensing channel within the set detection time window in that frequency band; Step S4: Based on the energy sequences of all spatial sensing channels in the same frequency band, construct the energy matrix of the corresponding frequency band; each row in the energy matrix represents a spatial sensing channel, each column represents a time sampling point within a set detection time window, and the matrix element values ​​are the vibration energy of the sub-signal of the corresponding frequency band under the corresponding time sampling point and the corresponding spatial sensing channel. If the energy matrix of the current frequency band Total OK In the energy matrix, each row represents a spatial sensing channel, and each column represents a time sampling point within a set detection time window. Therefore, the energy matrix for the current frequency band... for: , Among them, the energy matrix of the current frequency band elements in Indicates the first Each spatial sensing channel, within the set detection time window, the first... Vibrational energy at each time sampling point To set the total number of time sampling points within the detection time window, For spatial sensing channel indexing, This sets the index of time sampling points within the detection time window.

[0026] Step S5: Generate a multi-band spatiotemporal color image based on the energy matrix of the sub-signals of each frequency band; In this embodiment, the method for generating a multi-band spatiotemporal color image based on the energy matrix of sub-signals in each frequency band includes: The total energy of each frequency band is calculated by summing the values ​​of all elements in the energy matrix of the sub-signals of each frequency band. The formula is as follows: , in, This represents the total energy of the current frequency band.

[0027] Based on the total energy of each frequency band, the energy matrices of the sub-signals of each frequency band are sorted in descending order. The top three energy matrices after sorting are taken, mapped to grayscale images, and used as the R, G, and B channels of the multi-frequency band spatiotemporal color image, respectively, to synthesize the multi-frequency band spatiotemporal color image.

[0028] Step S6: Based on multi-band spatiotemporal color images, obtain the detection results of optical cable external damage events.

[0029] Existing research in other fields that converts one-dimensional signals into two-dimensional images and then processes them using convolutional neural networks (e.g., spectrograms in speech recognition, Gram angle fields or wavelet time-frequency maps in mechanical fault diagnosis) only retains time-frequency two-dimensional information in the generated images, losing the crucial spatial dimension information in fiber optic sensing signals. This makes it impossible to achieve concurrent detection and localization of multiple events along the optical cable line. In contrast, this invention acquires one-dimensional vibration time-series signals corresponding to multiple spatial sensing channels on the optical cable, constructing a multi-band spatiotemporal color image. Continuously acting external damage events appear as stripes or patches of specific colors and textures on the multi-band spatiotemporal color image, while random noise appears as irregular scattered points, effectively improving detection accuracy. Based on this, a target detection network is used to identify the target region corresponding to the external damage vibration in the image. The spatial location of the external damage event on the optical cable line can then be deduced from the pixel position of the target region in the row direction, thereby achieving the detection and localization of external damage events along the optical cable line and supporting the synchronous identification of multiple concurrent vibration events along the line.

[0030] In this embodiment, the detection results of the optical cable external damage event specifically include: the location of the target event on the optical cable to be detected, the start and duration, the spatial impact range, the event category and the confidence level.

[0031] In this embodiment, specifically, the method for obtaining the detection result of optical cable external damage event based on multi-band spatiotemporal color images includes: By passing the multi-band spatiotemporal color image through the optical cable breach event detection network, the bounding box coordinates and class probabilities of all target events in the multi-band spatiotemporal color image are obtained. The bounding box coordinates of all target events in the multi-band spatiotemporal color image are mapped to obtain the detection results of optical cable external damage events.

[0032] In this embodiment, the generated multi-band spatiotemporal color image is input into a trained improved YOLO network. The improved YOLO network performs deep mining and adaptive enhancement of the spatiotemporal texture features of the multi-band spatiotemporal color image, predicts and identifies the external damage target regions in the image, and outputs pixel-level detection results for all target events, including the bounding box pixel coordinates, pixel width, pixel height, and corresponding event category and category confidence of each target. The horizontal axis of the multi-band spatiotemporal color image corresponds to the detection time dimension, and the vertical axis corresponds to the spatial sensing channel dimension of the optical cable mileage. The pixel position and span of the bounding box output by the network in the horizontal direction of the image represent the occurrence time and duration of the external damage event, while the pixel position and span of the bounding box in the vertical direction of the image represent the occurrence location and spatial coverage of the external damage event on the optical cable. The target events specifically refer to external force damage events to the optical cable, distinguished from irregular environmental interference noise such as wind and rain disturbances, environmental vibrations, and equipment operating noise. Specifically, they are external force disturbance events that can cause mechanical damage to power optical cables, such as knocking, digging, crushing, and friction.

[0033] To further establish a one-to-one mapping relationship between the pixel coordinates of multi-band spatiotemporal color images and the actual physical spatiotemporal parameters of the optical cable, the target bounding box pixel information output by the network is reverse-calculated and its coordinates are transformed. Specifically: based on the calibration mapping relationship between the horizontal axis pixels of the image and the sampling points of the detection time, the start time of the external damage event is determined by the starting position of the horizontal axis pixels of the target bounding box, and the duration of the external damage event is calculated from the horizontal axis pixel span of the bounding box; based on the calibration mapping relationship between the vertical axis pixels of the image and the spatial sensing channel number and mileage of the optical cable, the precise location of the external damage event on the optical cable to be detected is determined by the center position of the vertical axis pixels of the target bounding box, and the spatial influence range of the external damage event along the optical cable is calculated from the vertical axis pixel span of the bounding box. Finally, by combining the event category and corresponding confidence level obtained by the network, a complete optical cable external damage event detection result is obtained. The optical cable external damage event detection result specifically includes: the spatial location of the target event on the optical cable to be detected, the event start time, the event duration, the spatial influence range of the optical cable, the external damage event category, and the event detection confidence level.

[0034] In this embodiment, the optical cable external damage event detection network is not limited to the improved YOLO network. Any neural network capable of extracting features from target events in multi-band spatiotemporal color images and outputting the bounding box coordinates and category probabilities of the target events can be used as the optical cable external damage event detection network in this invention.

[0035] In some embodiments, the optical cable external damage event detection network can adopt target detection networks such as YOLO series target detection networks, Faster R-CNN, RetinaNet, DETR, etc.; it can also adopt segmentation networks such as Mask R-CNN, U-Net, DeepLab, etc., and further obtain the bounding box coordinates of the target event based on the segmentation results.

[0036] In this embodiment, preferably, to improve the detection speed and deployment efficiency, the optical cable external damage event detection network is an improved YOLO (You Only Look Once) network. The improved YOLO network is obtained by setting a backbone network after the backbone network and then setting a spatiotemporal convolutional attention module.

[0037] Furthermore, in the neck network (FPN / PAN), convolutions are replaced with deformable convolutions to obtain the target improved YOLO network.

[0038] In this embodiment, an improved YOLO (You Only Look Once) network is constructed and trained to receive multi-band spatiotemporal color images and directly output the bounding box coordinates of all target events in the multi-band spatiotemporal color images. and category probability (C); where, The pixel coordinates of the predicted outer target bounding box center point on the horizontal axis (time direction) of the multi-band spatiotemporal color image. The pixel coordinates of the center point of the predicted outer target bounding box on the vertical axis (direction of the optical cable spatial channel) of the multi-band spatiotemporal color image. To predict the width of the bounding box, To predict the height of the bounding box; like Figure 2 As shown, Figure 2 The structure diagram for improving the YOLO network.

[0039] In this embodiment, the process of inputting a multi-band spatiotemporal color image into an improved YOLO network and outputting the bounding box coordinates and class probabilities of all target events in the multi-band spatiotemporal color image includes: Multi-band spatiotemporal color images are passed through a backbone network to obtain the features extracted by the backbone network; like Figure 3 As shown, Figure 3 This is a structural diagram of the spatiotemporal convolutional attention module.

[0040] Methods for obtaining spatiotemporal convolutional attention features by processing the features extracted from the backbone network through a spatiotemporal convolutional attention module include: The features extracted by the backbone network are subjected to global average pooling and global max pooling along the spatial dimension, respectively. The pooling results are then concatenated and spatial orientation attention weights are generated by a one-dimensional convolutional kernel. The features extracted by the backbone network are subjected to global average pooling and global max pooling along the time dimension, respectively. The pooling results are then concatenated and used to generate temporal attention weights through a one-dimensional convolutional kernel. After fusing the spatial and temporal attention weights through an outer product, the features extracted by the backbone network are weighted to obtain the spatiotemporal convolutional attention features. The spatiotemporal convolutional attention features are sequentially passed through the Neck network and the Head detector to obtain the bounding box coordinates and class probabilities of all target events in the multi-band spatiotemporal color image; the Neck network includes the FPN feature pyramid and the PAN path aggregation.

[0041] In the neck network (FPN / PAN), deformable convolutions are used to replace some conventional convolutions, enabling the improved YOLO network to adaptively learn the geometric deformation of image texture caused by different geological conditions and laying methods for the same type of external damage event, which significantly improves the robustness of the model.

[0042] In this embodiment, preferably, the selection of convolutions in the neck network is obtained by replacing them with deformable convolutions, and the selection method includes: By sequentially replacing convolutions at different locations within the neck network with deformable convolutions, multiple candidate improved YOLO networks are constructed, and each candidate improved YOLO network is iteratively trained using the same training dataset. By comparing the detection accuracy, recall, and robustness metrics of each candidate improved YOLO network, the convolutions in the neck network to be replaced with deformable convolutions are determined.

[0043] This invention performs a reverse physical mapping of the bounding box coordinates and dimensions output by the optical cable external damage event detection network to obtain the precise location of the event on the optical cable (kilometer marker or tower number), start and duration, spatial impact range, and event category and confidence level. Only when the confidence level exceeds a preset threshold (e.g., 0.8) will a precise alarm containing all multidimensional information be pushed to the monitoring center.

[0044] Compared with existing technologies, this invention constructs multi-band spatiotemporal color images, which simultaneously preserve the spatiotemporal two-dimensional morphological information and frequency domain energy distribution information of events. This allows different types of external damage events to have distinguishable color and texture features on the image, significantly improving detection accuracy and reducing false alarm rate.

[0045] By introducing a spatiotemporal convolutional attention module guided by the prior knowledge of external failure morphology into the YOLO network, the morphological prior of external failure events in spatiotemporal images is used to target and enhance features, suppressing random noise and further improving detection performance. Compared with existing attention modules such as CoordAttention, the spatiotemporal convolutional attention module of this invention performs average pooling and max pooling through two branches, which can simultaneously capture the energy concentration and peak distribution characteristics of vibration events. Combined with the outer product fusion method, it achieves accurate weighting of strip-shaped targets.

[0046] Based on Example 1, Example 2 takes the monitoring of external damage to the optical cable accompanying a buried gas pipeline as an example. On a 40-kilometer communication optical cable laid in the same trench as a natural gas pipeline next to a heavy industrial area, a phase-sensitive optical time-domain reflectometer was connected to the spare fiber core in the pipeline valve chamber. The equipment was set with a spatial resolution of 8m and a pulse transmission frequency of 2.5 kHz to conduct long-term uninterrupted vibration signal acquisition. The continuous acquisition period for this experiment lasted three months, during which 126 real external disturbance samples were fully recorded. The disturbance types covered typical risk events such as excavator breaker operation, pile driver construction, and illegal dumping by dump trucks. High-intensity environmental interference such as typhoons, rainstorms, and vibrations from passing vehicles also occurred during the acquisition period to test the algorithm's anti-interference performance. A 5-second detection time window was set during the experiment, and 5000 spatial sensing channels were configured. The collected one-dimensional vibration time-series signal was processed to generate a multi-band spatiotemporal color image with a size of 416×416 pixels. The staff labeled the corresponding detection bounding boxes and event categories for all 126 external force damage events. High-intensity environmental noise from wind, rain, and traffic was separately classified and labeled as strong noise samples. Then, all samples were divided into training set, validation set, and test set in a ratio of 8:1:1.

[0047] This embodiment uses an improved YOLOv5s network as the detection model, adds a spatiotemporal convolutional attention module to the output of the backbone network, and adds a neck FPN. In the PAN structure, some conventional convolutional layers were replaced with deformable convolutions. On the hardware side, four NVIDIA Tesla T4 GPUs were used for parallel training, with AdamW as the optimizer and an initial learning rate of 0.001, coupled with a cosine annealing learning rate adjustment scheme. The overall training iterations were set to 500 rounds. Testing showed that the mAP@0.5 index reached 96.8% under various external impact targets; the average accuracy for excavator breaker hammer operations was 98.1%, and the average accuracy for pile driver construction events was 97.5%. The system achieved a 99.1% accuracy rate in suppressing high-intensity environmental noise, with a noise-triggered false alarm rate of only 0.9%. The average absolute positioning error for detecting the location of optical cable damage was controlled within ±12.5 m. The trained model was subsequently deployed to an NVIDIA Jetson AGX Orin edge computing terminal, where the inference time for a single multi-band spatiotemporal color image was only 12.7 ms, meeting the requirements for real-time on-site monitoring.

[0048] This third embodiment's test scenario selected communication optical cables laid inside urban cable trenches for monitoring human-induced external force damage. The spatial resolution of the sensing equipment was configured to 3m. During the experiment, typical human-induced damage vibration signals were collected, including those from pickaxe digging, shovel prying, and handheld drill operation. Simultaneously, frequent interference samples from pedestrian walking and motorcycle engine idling vibrations were also collected. During model training, two data augmentation techniques—Mosaic stitching and random affine transformation—were introduced to expand the sample morphology and enhance the algorithm's performance in recognizing small-scale, weak disturbances. After testing, the algorithm achieved a recall rate of 93.5% and a detection precision of 91.2% for pickaxe digging events, and zero false alarms for frequent environmental disturbances such as motorcycle idling vibrations. This set of experiments verifies that this monitoring scheme can adapt to various vibration intensities and complex environmental backgrounds, and has good general detection performance for various optical cable disturbance events.

[0049] refer to Figure 4 This fourth embodiment provides a fiber optic cable external damage event detection system, including: The data acquisition module 10 is used to divide the optical cable under test into several spatial sensing channels and acquire the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable under test. The decomposition module 20 is used to perform multi-scale wavelet packet decomposition on the original one-dimensional vibration signals of each spatial sensing channel of the optical cable to be tested, so as to obtain the sub-signals of each spatial sensing channel under different frequency bands. The energy sequence acquisition module 30 is used to calculate the energy sequence of each spatial sensing channel within a set detection time window for each frequency band. The energy matrix construction module 40 is used to construct an energy matrix for a corresponding frequency band based on the energy sequences of all spatial sensing channels in the same frequency band. Each row in the energy matrix represents a time sampling point within a set detection time window, and each column represents a spatial sensing channel. The matrix elements are taken as the vibration energy of the sub-signal of the corresponding frequency band at the corresponding time sampling point and in the corresponding spatial sensing channel. Image generation module 50 is used to generate multi-band spatiotemporal color images based on the energy matrix of sub-signals in each frequency band; The detection module 60 is used to obtain the detection results of optical cable external damage events based on multi-band spatiotemporal color images.

[0050] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for detecting external damage to optical cables, characterized in that, include: The optical cable under test is divided into several spatial sensing channels, and the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable under test is obtained. Multi-scale wavelet packet decomposition was performed on the original one-dimensional vibration signals of each spatial sensing channel of the optical cable to be tested to obtain sub-signals of each spatial sensing channel in different frequency bands. For each frequency band, calculate the energy sequence of the sub-signal corresponding to each spatial sensing channel within the set detection time window in that frequency band; Based on the energy sequences of all spatial sensing channels in the same frequency band, construct the energy matrix of the corresponding frequency band; Each row in the energy matrix represents a spatial sensing channel, and each column represents a time sampling point within a set detection time window. The matrix elements are the vibration energy of the sub-signal of the corresponding frequency band at the corresponding time sampling point and in the corresponding spatial sensing channel. Based on the energy matrix of sub-signals in each frequency band, a multi-band spatiotemporal color image is generated; Based on multi-band spatiotemporal color images, the detection results of optical cable external damage events are obtained.

2. The method for detecting external damage to optical cables according to claim 1, characterized in that, Methods for obtaining the original one-dimensional vibration signals corresponding to each spatial sensing channel to be detected include: Based on a set acquisition frequency, a detection light pulse is emitted to the optical cable under test; The phase or amplitude change of the Rayleigh backscattered light interference signal returned by the optical cable under test is received and demodulated to obtain the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable under test.

3. The method for detecting external damage to optical cables according to claim 2, characterized in that, Methods for transmitting probe light pulses to the optical cable under test based on a set acquisition frequency include: Using a phase-sensitive optical time-domain reflectometer, probe light pulses are emitted toward the optical cable under test based on a set acquisition frequency.

4. The method for detecting external damage to optical cables according to claim 1, characterized in that, Methods for generating multi-band spatiotemporal color images based on the energy matrices of sub-signals in each frequency band include: The sum of all elements in the energy matrix of the sub-signals of each frequency band is used as the total energy of each frequency band. Based on the total energy, the energy matrices of the sub-signals in each frequency band are sorted in descending order. The top three energy matrices after sorting are then mapped to grayscale images and used as the R, G, and B channels of the multi-frequency band spatiotemporal color image, respectively, to synthesize the multi-frequency band spatiotemporal color image.

5. The method for detecting external damage to optical cables according to claim 1, characterized in that, The results of optical cable external damage event detection include: the location of the target event on the optical cable to be detected, the start and duration, the spatial impact range, the event category and confidence level.

6. The method for detecting external damage to optical cables according to claim 5, characterized in that, Methods for obtaining optical cable external damage event detection results based on multi-band spatiotemporal color images include: By passing the multi-band spatiotemporal color image through the optical cable breach event detection network, the bounding box coordinates and class probabilities of all target events in the multi-band spatiotemporal color image are obtained. The bounding box coordinates of all target events in the multi-band spatiotemporal color image are mapped to obtain the detection results of optical cable external damage events.

7. The method for detecting external damage to optical cables according to claim 6, characterized in that, The optical cable external damage event detection network is an improved YOLO network, which is obtained by adding a spatiotemporal convolutional attention module after the backbone network of the YOLO network.

8. The method for detecting external damage to optical cables according to claim 7, characterized in that, Methods for obtaining spatiotemporal convolutional attention features by processing the features extracted from the backbone network through a spatiotemporal convolutional attention module include: The features extracted by the backbone network are subjected to global average pooling and global max pooling along the spatial dimension, respectively. The pooling results are then concatenated and spatial orientation attention weights are generated by a one-dimensional convolutional kernel. The features extracted by the backbone network are subjected to global average pooling and global max pooling along the time dimension, respectively. The pooling results are then concatenated and used to generate temporal attention weights through a one-dimensional convolutional kernel. After fusing the spatial and temporal attention weights through an outer product, the features extracted by the backbone network are weighted to obtain spatiotemporal convolutional attention features.

9. The method for detecting external damage to optical cables according to claim 7, characterized in that, The improved YOLO network also includes: replacing convolutions in the neck network with deformable convolutions to obtain the target improved YOLO network. The methods selected include: By sequentially replacing convolutions at different locations within the neck network with deformable convolutions, multiple candidate improved YOLO networks are constructed, and each candidate improved YOLO network is iteratively trained using the same training dataset. By comparing the detection accuracy, recall, and robustness metrics of each candidate improved YOLO network, the convolutions in the neck network to be replaced with deformable convolutions are determined.

10. A system for detecting external damage to optical cables, characterized in that, include: The data acquisition module is used to divide the optical cable under test into several spatial sensing channels and acquire the original one-dimensional vibration signal corresponding to each spatial sensing channel of the optical cable under test. The decomposition module is used to perform multi-scale wavelet packet decomposition on the original one-dimensional vibration signals of each spatial sensing channel of the optical cable to be tested, so as to obtain the sub-signals of each spatial sensing channel under different frequency bands. The energy sequence acquisition module is used to calculate the energy sequence of each spatial sensing channel within a set detection time window for each frequency band. The energy matrix construction module is used to construct the energy matrix of the corresponding frequency band based on the energy sequences of all spatial sensing channels in the same frequency band. Each row in the energy matrix represents a spatial sensing channel, and each column represents a time sampling point within a set detection time window. The matrix elements are the vibration energy of the sub-signal of the corresponding frequency band at the corresponding time sampling point and in the corresponding spatial sensing channel. The image generation module is used to generate multi-band spatiotemporal color images based on the energy matrix of sub-signals in each frequency band; The detection module is used to obtain the detection results of optical cable external damage events based on multi-band spatiotemporal color images.