Underground electric power optical cable multi-scene vibration region positioning method based on PLGS-YOLO model

Abstract

The underground electric power optical cable multi-scene vibration area positioning method based on the PLGS-YOLO model relates to the technical fields of deep learning models and electric power optical cable monitoring, solves the problem that the model used for positioning the vibration area in the existing underground electric power optical cable is interfered by background noise or non-target area signals, so that the model cannot meet the application requirements, and the method is used for judging the laying scene of a vibration event and positioning the corresponding signal area.Through optimizing YOLOv11n, using SPPFDMSCA and PLMSAM to enhance the feature extraction efficiency through a multi-scale attention mechanism, combining the advantages of traditional convolution and deep separable convolution through GSConv, reducing the calculation complexity while retaining multi-channel features, and optimizing feature fusion through VoVGSCSP, the detection precision and model efficiency are improved, and the calculation complexity is significantly reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning models and power cable monitoring, and particularly relates to an underground power cable multi-scene vibration area positioning method based on a YOLO model with progressive lightweight grouped-shuffle feature fusion and spatial attention (PLGS-YOLO). BACKGROUND

[0002] As an important communication device in the power system, underground power cables have become a key infrastructure to ensure the city's power supply and efficient operation of the power grid. Underground power cables have multiple scene laying methods such as buried, underground and cable trench, among which buried and underground laying often face the risk of natural and man-made damage. For example, in buried laying, construction activities may cause the cable to be pressed, displaced or broken; underground laying may encounter illegal excavation or foreign object impact, affecting the transmission quality of the optical fiber. How to effectively manage and maintain underground power cables to ensure smooth communication has become a problem to be solved.

[0003] Φ-OTDR optical fiber sensing technology has the advantages of high sensitivity, long distance detection, high reliability and low cost. This technology uses optical fiber as a medium and carrier to perceive various vibrations occurring along the optical fiber and collects vibration information in real time. Man-made damage events can be analyzed by collecting vibration signals. This technology is widely used in pipeline safety warning, perimeter security, railway transportation, submarine cable safety detection and other fields. A portion of unused optical fibers will be reserved in power cables to replace the main optical fibers in case of failure. Therefore, unused internal optical fibers can be used as a sensing medium for vibration signals when abnormal vibration events threaten underground power cables. It is very suitable for application in underground power cable monitoring tasks. Therefore, Φ-OTDR optical fiber sensing technology is used to collect vibration signals in the environment around the underground power cable, and related research methods are used to locate the area of possible damage caused by human events.

[0004] Currently, the single-stage method of You Only Look Once (YOLO) model is mainly used to locate the area corresponding to the vibration event in the signal space-time graph collected by the Φ-OTDR system. The input of the YOLO model is the time-space graph of the vibration signal, which converts time series data into an image. However, this method only measures the accuracy and real-time performance of the method, while the underground power cable laying scene is complex, the vibration event type is diverse, and the interference of background noise or non-target area signals is different. The deployment difficulty of the proposed method also needs to be fully considered.

[0005] Currently, the Φ-OTDR vibration region is mainly divided into two methods based on signal processing technology and target detection model. In the actual environment, the interference noise is large, which will submerge the useful signal, which is not conducive to the vibration region. Based on the above problems, a series of positioning methods using signal processing technology have been proposed in recent years. Wu et al. used a simulated annealing algorithm based on an adaptive annealing threshold for wavelet denoising, and then calculated the spatial gradient of the gray image using a two-dimensional edge detection method for axial positioning. Huang et al. proposed a regional axial positioning method based on overlapping phase cross-correlation, according to the linear relationship of the axial point vibration signal on the link with time, the autocorrelation coefficient of the adjacent time vibration signal matrix is calculated, and the maximum value in the link is taken as the position of the vibration.

[0006] The positioning method based on signal processing technology has good positioning effect on high-frequency and large-amplitude vibration region, but poor positioning effect on low-frequency, small-amplitude event and buried deep or distant vibration region. Therefore, a series of positioning methods based on target detection model are proposed. The target detection model has high precision and real-time. Xu et al. proposed a YOLOv3 multi-class vibration detection model for intrusion event vibration region positioning and real-time detection. Wang et al. proposed a Φ-OTDR perimeter security event vibration region positioning method based on CBAM-YOLOv8. However, both types of positioning methods are collected in a specific laying scene and positioned, while the laying scene of a complete underground power cable line is complex and needs to consider the positioning problem in different laying scenes of the same cable.

[0007] To solve the problem of vibration region positioning of underground power cable in multiple scenes, the present application proposes a PLGS-YOLO model based on YOLOv11n, which can judge the laying scene of the vibration region and locate the vibration region at the same time. This method considers the buried and underground laying scenes where human damage occurs more frequently, as well as the size, accuracy and real-time performance of the model used by the method. It can be directly applied to real underground power cables to prevent potential damage and better protect the safety of underground optical cables, which has important practical significance and application prospect. SUMMARY

[0008] The present application is to solve the problem that the existing vibration region positioning of underground power cable is mainly concentrated in a single laying scene, and such methods are difficult to effectively balance between performance, real-time performance and model lightweight, limiting their application in multiple laying scenes. In addition, the model used is disturbed by background noise or non-target region signal, which leads to the problem that the model cannot meet the application requirements. The present application provides a PLGS-YOLO model-based underground power cable multi-scene vibration region positioning method.

[0009] The underground power optical cable multi-scene vibration area positioning method based on the PLGS-YOLO model is realized through the following steps:

[0010] Step one, pre-processing the obtained vibration signal, and converting the pre-processed signal into a space-time graph;

[0011] Step two, vibration area positioning;

[0012] By constructing the PLGS-YOLO model and training the PLGS-YOLO model, the trained PLGS-YOLO model is used to position the vibration area in the space-time graph obtained in step one, and the positioned space-time graph and the corresponding laying scene type are generated;

[0013] The PLGS-YOLO model improves the YOLOv11n network, replaces the SPPF module and the C2PSA module with the SPPFDMSCA module and the PLMSAM module in the backbone network respectively, replaces the convolution layer in the neck network with the GSConv module, and replaces the C3k2 module with the VoVGSCSP module;

[0014] The DMSCA module is introduced in the SPPFDMSCA module to enhance the global feature extraction capability of the backbone network;

[0015] The LMSCA module is used in the PLMSAM module to obtain a deep feature map;

[0016] The space-time graph is input into the YOLOv11n backbone network, the feature is enhanced through the dynamic multi-scale convolution attention mechanism in the SPPFDMSCA module, and the obtained global feature map is input into the PLMSAM module;

[0017] The global feature map is input into the neck network after being operated by the LMSCA module in the PLMSAM module;

[0018] The feature maps after multi-scale convolution operation of the GSConv module and the VoVGSCSP module in the neck network are fused, and finally the positioned space-time graph and the corresponding laying scene type are obtained through the detection head.

[0019] The beneficial effects of the present application are:

[0020] The Φ-OTDR system is deployed in a certain communication station, the underground power optical cable line with two scenes of buried and underground is selected as the scene where human threats are more frequent, the optical cable is used as a vibration sensing medium, and a vibration event experiment simulating a real threat is designed. The obtained original signal is pre-processed to obtain a self-built positioning model data set.

[0021] The method of the present application proposes a PLGS-YOLO model to realize the positioning of the vibration area in the two laying scenes of buried and underground. On the basis of the YOLOv11n model, the spatial pyramid pooling structure (SPPFDMSCA) with dynamic multi-scale convolution attention and the progressive lightweight module (PLMSAM) based on partial self-attention mechanism are introduced to enhance the vibration feature extraction efficiency, and the grouped shuffle convolution (GSConv) and the one-shot aggregation-based grouped shuffle cross stage partial block (VoVGSCSP) are combined to optimize the feature fusion process. Compared with the YOLO series of lightweight models (including the models applied to the vibration area positioning), it is proved that the PLGS-YOLO can significantly reduce the calculation complexity and memory occupation while improving the positioning accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The flow chart of the underground power optical cable multi-scene vibration area positioning method based on the PLGS-YOLO model of the present application;

[0023] Figure 2 The principle diagram of the PLGS-YOLO model in the underground power optical cable multi-scene vibration area positioning method based on the PLGS-YOLO model of the present application;

[0024] Figure 3 The principle diagram of the SPPFDMSCA module;

[0025] Figure 4 The principle diagram of the DMSCA module;

[0026] Figure 5 The principle diagram of the PLMSAM module;

[0027] Figure 6 The principle diagram of the LMSCA module;

[0028] Figure 7 The principle diagram of the GSConv module;

[0029] Figure 8 is a VoVGSCSP module schematic diagram;

[0030] Figure 9 is an experimental scene and a deployment effect diagram of the Φ-OTDR system;

[0031] Figure 10 (a), (b), (c) and (d) in FIG. 1 are φ-OTDR signal effect diagrams of foreign object falling, digging, knocking and vibration events of electric drill and hammer, respectively;

[0032] Figure 11 (a), (b), (c) and (d) in FIG. 2 are denoised φ-OTDR signal effect diagrams of foreign object falling, digging, knocking and vibration events of electric drill and hammer, respectively;

[0033] Figure 12 is a space-time diagram for positioning a buried vibration area, wherein (a), (b), (c), (d), (e), (f), (g) and (h) are space-time diagrams of background noise, motorcycle, small cart, digging, pickaxe, lead ball, walking and electric drill and hammer, respectively;

[0034] Figure 13 is a space-time diagram for positioning an underground vibration area, wherein (a), (b), (c), (d), (e), (f) and (g) are space-time diagrams of background noise, foreign object falling, electric drill, hooking, climbing, knocking and walking, respectively;

[0035] Figure 14 is a performance curve effect diagram of a PLGS-YOLO model on a verification set; wherein (a), (b), (c), (d), (e) and (f) are curve effect diagrams of Boxloss, Clsloss, Precision, Recall, mAP@50 and mAP@50-95, respectively.

[0036] Figure 15 is a visualization effect diagram of a detection result of an underground power optical cable multi-scene vibration area positioning method based on a PLGS-YOLO model; wherein (a), (b), (c), (d), (e), (f), (g) and (h) are visualization effect diagrams of a buried motorcycle, a buried knocking, a buried digging, a buried small cart, an underground drilling, an underground walking, an underground knocking and a foreign object falling, respectively. DETAILED DESCRIPTION

[0037] DETAILED DESCRIPTION ONE, COMBINATION Figures 1 to 8The underground power optical cable multi-scene vibration region positioning method based on the PLGS-YOLO model is illustrated. For the underground power optical cable multi-scene vibration event, the PLGS-YOLO model is constructed by improving YOLOv11n. The overall process of the method includes two stages, as shown in FIG. 1. The green line represents the vibration signal preprocessing stage, and the red line represents the vibration region positioning stage. The specific steps of the method are as follows: Figure 1

[0038] Step 1: Vibration signal preprocessing. The Φ-OTDR signal is denoised by high-pass and low-pass filtering to obtain the preprocessed spatiotemporal signal. The spatiotemporal signal is converted into a spatiotemporal graph.

[0039] In the processing of underground power optical cable vibration signals in multi-laying scenarios, the signals collected by the Φ-OTDR system often contain high-frequency machine internal noise and low-frequency environmental noise interference. These noises can mask the effective features of the vibration signals and affect the positioning performance of subsequent methods. Therefore, in this embodiment, a combination of low-pass filtering and high-pass filtering is used to preprocess the vibration signals, remove irrelevant noise from the frequency domain, and improve signal quality. At the same time, high-pass filtering and low-pass filtering have short processing time and can meet the real-time requirements of signal preprocessing of the system.

[0040] Step 1-1: High-pass filtering.

[0041] High-pass filtering (High-Pass Filter, HPF) allows high-frequency components in the signal to pass through while suppressing low-frequency components. In the Φ-OTDR signal, low-frequency noise usually comes from environmental noise during system operation, and these low-frequency disturbances may mask the detailed features in the vibration event. By high-pass filtering, these low-frequency noises can be removed, highlighting the key change information in the vibration signal. The transfer function of high-pass filtering is defined as:

[0042]

[0043] In the formula, H HPF (f) represents the transfer function of the high-pass filter, f is the frequency of the input signal, and f c1 is the cutoff frequency of the high-pass filter.

[0044] The output signal of the high-pass filter can be represented as:

[0045]

[0046] where x[k] is the original input signal, h HPF ​[n-k] is the impulse response of the high-pass filter at time step n, N represents the length of the filter, n represents the current time step, the filter calculates the corresponding output signal y HPF [n].

[0047] Step 1-2: Low-pass filtering;

[0048] Low-pass filtering allows low-frequency components in the signal to pass through while suppressing high-frequency noise. In the Φ-OTDR signal, high-frequency noise is usually caused by electromagnetic interference or random fluctuations of the micro-vibration signal inside the machine. By low-pass filtering, these high-frequency components can be removed, retaining the main trend information in the optical cable vibration signal, thus more effectively reflecting the spatial distribution of vibration. The transfer function of low-pass filtering is defined as:

[0049]

[0050] where H LPF (f) represents the transfer function of the low-pass filter, f c2 is the cut-off frequency of the low-pass filter. The output signal of the low-pass filter can be represented as:

[0051]

[0052]

[0053] where h LPF [l-m] is the impulse response of the high-pass filter in the Φ-OTDR system at time step l, y LPF [n] is the output signal of the high-pass filter at time step n.

[0054] Step 1-3: Combination of high-pass and low-pass filtering;

[0055] In the processing of underground power optical cable vibration signals, low-pass filtering is used to retain the main trend of the vibration signal, such as large amplitude vibration events caused by foreign object falling, excavation, etc.; high-pass filtering is used to extract the detailed changes of the vibration signal, such as event characteristics caused by high-frequency vibrations such as tapping or electric drill hammer. By combining low-pass and high-pass filtering, the signal-to-noise ratio can be significantly improved, and the performance of the proposed method in the vibration event positioning task can be optimized.

[0056] First, the method of high-pass filtering is used to remove low-frequency noise, and it is defined as:

[0057]

[0058] where x[m] is the original Φ-OTDR signal from the Φ-OTDR system before filtering, h HPF [l-m] is the impulse response of the high-pass filter in the Φ-OTDR system at time step l, y HPF ​[l] is the output signal of the original Φ-OTDR signal after high-pass filtering at time step l.

[0059] Secondly, a low-pass filtering process is adopted to smooth the high-frequency components. The final combined filtering equation containing both high-pass and low-pass filtering is:

[0060]

[0061] where h LPF [n-l] is the impulse response of the low-pass filter in the Φ-OTDR system at time step n, y filtered [n] is the final output signal after high-pass and low-pass filtering.

[0062] In this embodiment, the combined filtering process not only retains the main trend information of low frequency, but also extracts important detailed features of high frequency, providing high-quality input data for subsequent positioning. The filter parameters are determined according to the characteristics of the data set, optimizing the filtering process for a given vibration signal.

[0063] Step 2: Vibration area positioning;

[0064] By constructing the PLGS-YOLO model and training the PLGS-YOLO model, the trained PLGS-YOLO model is used to position the vibration area in the spatio-temporal graph obtained in step one, generating the positioned spatio-temporal graph and the corresponding laying scene type;

[0065] In this embodiment, the underground power optical cable vibration area positioning method in multiple laying scenes needs to position the vibration area and detect the corresponding laying scene of the vibration area, with high requirements for real-time performance and accuracy. The YOLO series model has shown good real-time performance and accuracy in target detection, effectively solving the problems of vibration area positioning and scene detection. Among them, YOLOv11n has significantly improved real-time performance and accuracy compared to previous YOLO versions, but the computational complexity and parameter amount are still high, which is not suitable for deployment in environments with limited computing power.

[0066] To this end, the embodiment proposes a vibration region positioning method based on a PLGS-YOLO model. In view of the differences in signal strength and distribution between underground power optical cable burying and underground scene, the YOLOv11n backbone network and neck network are optimized to meet the requirements of real-time performance, accuracy and lightweight, and to better position the regional vibration signal and detect the laying scene. By introducing the SPPFDMSCA module to replace the SPPF module, the global feature extraction capability of the backbone network is enhanced. The PLMSAM module is introduced to replace the C2PSA module to enhance the deep feature extraction capability of the backbone network. The GSConv module and VoVGSCSP module are used to balance the model calculation complexity and detection accuracy, and the feature fusion quality of the neck network is optimized. The overall structure of the model is as shown in Figure 2

[0067] Step 2-1, a PLGS-YOLO model is constructed. The PLGS-YOLO model improves the YOLOv11n model. In the backbone network, the SPPF module and the C2PSA module are replaced by the SPPFDMSCA module and the PLMSAM module respectively. In the neck network, the convolution layer is replaced by the GSConv module, and the C3k2 module is replaced by the VoVGSCSP module.

[0068] A dynamic multi-scale convolution attention mechanism module (Dynamic Multi-Scale Convolution Attention, abbreviated as DMSCA) is introduced in the SPPFDMSCA module to enhance the global feature extraction capability of the backbone network.

[0069] The LMSCA module is used in the PLMSAM module to obtain deep feature maps.

[0070] In this embodiment, three different scale features are extracted by the backbone network, which is composed of convolution layer 1, convolution layer 2, C3k2 module 1, convolution layer 3, C3k2 module 2, convolution layer 4, C3k2 module 3, convolution layer 5, C3k2 module 4, SPPFDMSCA module and PLMSAM module.

[0071] The space-time graph in step 1 is input into the backbone network, and the processed feature maps are output to the neck network and the convolution layer 4 through the convolution layer 1, the convolution layer 2, the C3k2 module 1, the convolution layer 3 and the C3k2 module 2 in turn.

[0072] The feature maps output by the convolution layer 4 are output to the neck network and the convolution layer 5 through the C3k2 module 3.

[0073] ​The feature map output by the convolution layer 5 is output to the up-sampling module 1 and the Concat module 4 of the neck network through the C3k2 module 4, the SPPFD MSCA module and the PLMSA module;

[0074] In YOLOv11n, the original SPPF module uses a multi-scale max pooling operation to fuse spatial information of different receptive fields, but this structure has limitations when processing underground optical cable vibration signals. There is a significant scale difference in vibration signals in space, and the signal coverage and feature complexity of different vibration events are different, and the propagation mode is also different in underground and downhole laying scenes. The traditional pooling operation may lose important spatial details while compressing information, making it difficult to fully express complex vibration region features. To improve the modeling capability of the model for multi-scale vibration signals, in the embodiment, the SPPF module in the backbone network is replaced with an SPPFDMSCA module, as shown in Figure 3 The module introduces a DMSCA module based on the original multi-scale structure, enhances feature expression capability through parallel convolution and attention fusion, further optimizes spatial context modeling effect, and enhances the global feature extraction capability of the backbone network, so as to more accurately extract key vibration region features and improve vibration region positioning performance.

[0075] As shown in Figure 3 In the embodiment, the SPPFDMSCA module includes a convolution layer 1, a max pooling layer 1, a max pooling layer 2, a max pooling layer 3, a Concat module, a convolution layer 2 and a dynamic multi-scale convolution attention mechanism module (DMSCA module).

[0076] The SPPFDMSCA module receives the high semantic feature map after the C3k2 module 4 as input, first passes through the multi-layer residual connection of the convolution layer 1, the max pooling layer 1, the max pooling layer 2, the max pooling layer 3 and the Concat module, then passes through the convolution layer 2 and the DMSCA module for feature enhancement, and finally inputs the obtained global feature map to the PLMSAM module.

[0077] As shown in Figure 4 In the DMSCA module, the input is a global feature map with a size of DMSCA Input ∈R N×C×H×W N represents batch size, C represents input channel number, H and W represent height and width of the feature map respectively. Sigmoid represents an activation function, Softmax represents a normalization function, Concat represents a channel splicing operation, · represents matrix multiplication, Conv represents convolution operation, and reshape represents adjustment of the channel dimension of the feature map.

[0078] Firstly, the module introduces multi-scale convolution operation, using 3x3, 5x5 and 7x7 convolution kernels on input features Figure X Parallel convolution extraction is performed, and the features extracted under three scales are spliced and fused in the channel dimension to obtain multi-scale features Figure X ms This operation can capture vibration features under different receptive fields and enhance the model's perception of multi-type vibration regions. The specific calculation process is as follows:

[0079] X ms =Concat(Conv 3×3 (DMSCA Input ),Conv 5×5 (DMSCA Input )),Conv 7×7 (DMSCA Input )))

[0080] Then, the multi-scale features Figure X ms are generated through 1x1 convolution to generate attention map A, and are activated through Sigmoid function, so that the attention map focuses on the key response area. At the same time, X ms is generated through 1x1 convolution to generate key-value pair K and V, which are used to build context modeling module. K is normalized through Softmax function and multiplied with V to calculate context enhanced features Y. The specific calculation process is as follows:

[0081] A=Sigmoid(Conv 1×1 (X ms ))

[0082] K=Softmax(reshape(Conv 1×1 (X ms )))

[0083] V=reshape(Conv 1×1 (X ms ))

[0084] Y=Conv 1×1 (V·K)

[0085] Finally, the attention map A and the context enhanced features Y are added to the input feature map DMSCA Input , and the output global feature map DMSCA output of the module is obtained, so as to realize the joint modeling and enhancement of multi-scale and context information. The specific calculation process is as follows:

[0086] DMSCA output =DMSCA Input +A·Y

[0087] The DMSCA module significantly improves the modeling and perception ability of the model for complex vibration patterns by fusing multi-scale spatial receptive fields and context attention mechanisms, which helps to finely locate the vibration region.

[0088] The PLMSAM module receives the global feature map DMSCA output , and further compresses redundancy and enhances key region perception through an improved lightweight multi-scale attention mechanism module (LMSAM module) to obtain a deep feature map.

[0089] The C2PSA module in the original YOLOv11n backbone network extracts deep features of the vibration spatiotemporal graph through multiple PSA and cross-connections, and is used for feature fusion of the neck network. However, in the processing of vibration signals of underground power optical cables, similar spatiotemporal graph features are generated under the action of foreign objects falling and lead balls in different laying scenarios. This similarity increases the difficulty of the model in laying scenario detection, especially when the vibration events have similar frequencies and time domain changes. To solve the above problems, in the present embodiment, a PLMSAM module is designed to replace the C2PSA module for deep feature extraction, thereby enhancing the deep feature extraction capability of the backbone network. The PLMSAM module follows the framework structure of PSA, but replaces the Attention module with an LMSAM module, which has a structure as shown in Figure 5 , which can more effectively capture key information in the vibration spatiotemporal graph while reducing model parameters and computational complexity, significantly improving the adaptability of the model to complex vibration signals of underground power optical cables and providing high-quality input for subsequent feature fusion of the neck network.

[0090] As shown in Figure 5 , in the present embodiment, the PLMSAM module includes a batch normalization layer 1, an LMSAM module, a batch normalization layer 2, a feedforward neural network module (FFN module), the LMSAM module includes a spatial attention module (SAM module) and a lightweight multi-scale channel attention module (LMSCA module);

[0091] The PLMSAM module receives the global feature map after the SPPF DMSCA module as input, first obtains a multi-branch fusion feature map through a residual connection structure composed of a batch normalization layer 1 and an LMSAM module, then obtains a deep feature map through a residual connection structure composed of a batch normalization layer 2 and an FFN module, and finally inputs the deep feature map obtained by the PLMSAM module into an up-sampling module 1 and a Concat module 4 of the neck network.

[0092] The channel attention module calculates the input feature map by a pooling operation to convert it into pixels, resulting in a large amount of spatial information loss. The spatial channel attention mechanism module (CBAM module) alleviates this problem by combining the channel attention module (CAM module) and the spatial attention module (SAM module), but CBAM mainly optimizes the features by weighting the channel and space of the feature map, which is difficult to effectively capture the relationship between different scale targets. Therefore, the LMSAM module is proposed to solve this problem in the embodiment, in which the channel attention module of the CBAM module is replaced by the LMSCA module. This mechanism better captures key features in the signal through multi-scale convolution, enhances the model's perception of complex vibration signals, and significantly improves the accurate positioning of the same vibration event in different laying scenarios. The LMSCA structure is shown in Figure 6

[0093] The LMSCA structure includes three parts: a 5x5 convolution layer, a multi-branch convolution layer, a 1x1 convolution layer, and an average pooling layer, wherein the multi-branch convolution includes a 1x7 convolution layer, a 1x11 convolution layer, and a 1x11 convolution layer. The 5x5 convolution layer obtains local features, the multi-branch convolution layer captures multi-scale target relationships in different channels, and the 1x1 convolution layer weights the original feature map to enhance the expression ability of the feature map. The calculation process is as follows:

[0094] M = Conv 5×5 (LMSCA Input )

[0095] M1 = Conv 1×7 (M)

[0096] M2 = Conv 1×11 (M)

[0097] M3 = Conv 1×21 (M)

[0098]

[0099] wherein, LMCA Input represents the input feature map, Conv 1×i represents a two-dimensional convolution with different convolution kernel sizes, M1, M2, and M3 represent three branches, and each branch uses a convolution kernel of different size to perform convolution processing on the input. LMCA Output represents the deep feature map obtained after the LMCA Input is processed by the average pooling layer and the 1x1 convolution weight.

[0100] ​Step 2-2, feature fusion is performed in the neck network, which is that all convolution layers in the original YOLOv11n neck network are replaced by GSConv modules, and all C3k2 modules are replaced by VoVGSCSP modules. The output of each module is sequentially input to the subsequent module, which constitutes a feature flow path from the backbone to the detection head, supporting the model to realize multi-scale vibration area detection.

[0101] The convolution layers and C3k2 modules in the YOLOv11n neck network process all channel information through global aggregation, which can extract rich feature information, but due to the high computational complexity and parameter quantity, it increases the computational burden of the model in complex environments. In the two laying scenes of underground and downhole, the characteristics of the underground power optical cable vibration signal have obvious differences. The downhole signal usually has strong local variability, while the buried signal is greatly affected by environmental factors (such as soil medium, optical cable sleeve, etc.), and the signal is relatively flat. Therefore, the model not only needs to extract rich multi-channel features, but also needs to effectively process the temporal and spatial differences of these signals. In order to solve the above problems, all convolution layers in the YOLOv11n neck network are replaced by GSConv module 1 and GSConv module 2 in the embodiment.

[0102] Compared with traditional convolution, depth separable convolution performs outstandingly in reducing computational complexity. However, it has the disadvantage of partially ignoring the information relationship between channels, leading to the phenomenon of feature information loss. In order to combine the advantages of depth separable convolution and traditional convolution, GSConv fuses the global feature information obtained by traditional convolution with the channel feature information obtained by depth separable convolution, effectively preserving multi-channel information while reducing computational complexity.

[0103] In the embodiment, the neck network includes GSConv modules (GSConv module 1 and GSConv module 2), VoVGSCSP modules (VoVGSCSP module 1, VoVGSCSP module 2, VoVGSCSP module 3, and VoVGSCSP module 4), up-sampling modules (up-sampling module 1 and up-sampling module 2), and Concat modules (Concat module 1, Concat module 2, Concat module 3, and Concat module 4);

[0104] The feature map output by the C3k2 module 2 in the backbone network is input to the Concat module 2 in the neck network, and the feature map after the Concat module 2 is input to the VoVGSCSP module 2. The VoVGSCSP module 2 outputs compressed shallow high-resolution feature maps and inputs them to the Concat module 3 through the GSConv module 1, and simultaneously inputs them to the detection head 1 of the head network, for shallow vibration area positioning and laying scene prediction.

[0105] The feature map output by the C3k2 module 3 in the backbone network is input to the Concat module 1 in the neck network, and after being input to the VoVGSCSP module 1 through the Concat module 1, the feature map output by the VoVGSCSP module 1 is input to the upsampling module 2 and the Concat module 3, and after being input to the Concat module 2 through the upsampling module 2, the feature map is input to the VoVGSCSP module 3 through the Concat module 3. The VoVGSCSP module 3 outputs a compressed middle-layer high-resolution feature map; the high-resolution feature map is input to the Concat module 4 through the GSConv module 2, and at the same time, is input to the detection head 2 of the head network, so as to perform middle-layer vibration area positioning and laying scene prediction.

[0106] The feature map output by the PLMSAM module in the backbone network is input to the upsampling module 1 and the Concat module 4 in the neck network, and after being input to the Concat module 1 through the upsampling module 1, the feature map is input to the VoVGSCSP module 4 through the Concat module 4. The VoVGSCSP module 4 outputs a deep-layer fusion feature map, which is input to the detection head 3 in the head network, so as to perform deep-layer vibration area positioning and laying scene prediction.

[0107] In the embodiment, the input of the GSConv module 1 is the feature map of the VoVGSCSP module 2, and the output is a compressed shallow-layer high-resolution feature map for use by the Concat module 3; the input of the GSConv module 2 is the feature map of the VoVGSCSP module 3, and the output is a compressed middle-layer high-resolution feature map for use by the Concat module 4.

[0108] The input of the VoVGSCSP module 1 is the spliced middle-deep-layer feature map obtained by the Concat module 1, and the output is a middle-layer fusion feature map for use by the upsampling module 2 and the Concat module 3; the input of the VoVGSCSP module 2 is the spliced shallow-middle-layer feature map obtained by the Concat module 2, and the output is a shallow-layer fusion feature map for use by the detection head 1 and the GSConv module 1; the input of the VoVGSCSP module 3 is the spliced shallow-middle-layer feature map obtained by the Concat module 3, and the output is a middle-layer fusion feature map for use by the detection head 2 and the GSConv module 2; the input of the VoVGSCSP module 4 is the spliced middle-deep-layer feature map obtained by the Concat module 4, and the output is a deep-layer fusion feature map for use by the detection head 3.

[0109] As Figure 7As shown, in the embodiment, the GSConv module includes a convolution layer Conv, a depth separable convolution layer DWConv, a Concat module, and a channel rearrangement module Shuffle; wherein the Conv is composed of a traditional convolution, a batch normalization operation, and an activation function, and the DWConv is composed of a depth separable convolution, a batch normalization operation, and an activation function. Assuming that a feature map with C1 channels is input, the feature map is first processed by the traditional convolution to generate a feature map with C2 / 2 channels, and then processed by the depth separable convolution to generate another feature map with C2 / 2 channels. In the case of keeping the number of channels unchanged, the results of the depth convolution and the depth separable convolution are connected, and finally the channel fusion mechanism of the Shuffle module is used to integrate the feature information from the traditional convolution and the depth separable convolution to generate an output feature map with C2 channels.

[0110] In order to fully utilize the function of GSConv in the neck network of YOLOv11n, in the embodiment, all C3k2 modules are replaced with VoVGSCSP module 1, VoVGSCSP module 2, VoVGSCSP module 3, and VoVGSCSP module 4 without changing the original input and output size of C3k2. The module adopts a single-level aggregation strategy and designs an efficient cross-level network module. By using a residual structure to integrate traditional convolution and GSConv while maintaining good accuracy, the inference time complexity and model parameter count are reduced. The structure of VoVGSCSP is as shown in Figure 8

[0111] In the embodiment, each VoVGSCSP module is composed of convolution layer 1, convolution layer 2, convolution layer 3, GSConv module branch, Concat module, and convolution layer 5, wherein the GSConv module branch includes GSConv module 1 and GSConv module 2.

[0112] The first part of the C1 channel feature map is obtained by performing shallow feature extraction on the C1 / 2 channel feature map through the convolution layer 1, and then performing deep feature extraction on the C1 / 2 channel feature map through the convolution layer 2 and the GSConv module branch, respectively, and then obtaining the C2 / 2 channel deep feature map through residual addition.

[0113] The second part of the C1 channel feature map is directly extracted through the convolution layer 3 to obtain the C2 / 2 channel backbone feature map.

[0114] Finally, the C2 / 2 channel deep feature map and the C2 / 2 channel backbone feature map are spliced and feature-extracted through the Concat module and the convolution layer 5 to obtain the C2 channel fusion feature map. ​

[0115] Specific Implementation Method Two: Combination Figures 9 to 15 This embodiment describes an experimental example of the multi-scenario vibration zone localization method for underground power optical cables based on the PLGS-YOLO model described in Specific Embodiment 1:

[0116] 1. Experimental design and deployment;

[0117] In this embodiment, the experimental object is an underground power optical cable line with multiple scenarios at a communication station in Jilin Province, with a total length of 1.98 km. The total length of the optical cable inside the communication station is 0.42 km, the cumulative length of the buried optical cable outside the communication station is approximately 1.5 km, and the length of each underground optical cable is approximately 0.03 km. During the laying of the underground power optical cable, the optical cable in the buried area is protected by conduit, while the optical cable in the underground and station areas is not protected by conduit. The red line segment in the figure represents the external conduit protection structure of the optical cable. To accurately collect vibration signals from the specified buried area and underground optical cable, a 1 km long G652D single-mode optical fiber was first connected to the Ф-OTDR system. The other end of the single-mode optical fiber was connected to one of the fiber cores of the underground power optical cable. The pulse width of the acquisition card was set to 32 ns, the frequency to 1000 Hz, and the spatial resolution to 2 m. The scenario and instrument deployment are as follows. Figure 9 As shown.

[0118] Fifteen different types of vibration events were tested in designated areas of underground and manhole-mounted fiber optic cable lines, such as... Figure 12 As shown in the table. The underground scene includes background noise, motorcycles, wheelbarrows, digging, pickaxes, shot puts, walking, and electric drills / hammers; the underground scene includes background noise, falling objects, electric drills, hooking, climbing, knocking, and walking. Specific event descriptions are shown in Tables 1 and 2. Table 1 describes the specific vibration events underground, and Table 2 describes the specific vibration events underground.

[0119] Table 1

[0120] Event type Detail description Background noise No any disturbance in natural state Motor One person riding a motorbike to and fro in the collecting area of the buried objects Trolley One trolley carrying heavy objects to and fro in the collecting area of the buried objects Digging One shovel to dig in the collecting area of the buried objects every 3 seconds Pick One pick to hit the ground in the collecting area of the buried objects Lead ball One person to hit the ground with a 5kg lead ball every 3 seconds Walking Two persons to walk to and fro in the collecting area of the buried objects in opposite directions at a constant speed Electric drill and hammer One multifunctional electric drill to drill continuously on the stone in the collecting area of the buried objects

[0121] Table 2

[0122] Event type Detail description Background noise No any disturbance on the ground and no person in the well Foreign object falling One person to throw a hammer or wrench down from the well by a rope Electric drill One person to drill continuously on the stone in the vertical direction of the optical cable Hooking and pulling One person to pull up and drop down the optical cable by the sharp end of a hammer every 2 seconds Climbing One person to climb up and down a ladder Knocking One person to knock on different areas of the optical cable in the well every 1 second Walking One person to walk to and fro along the line at a constant speed

[0123] 2. Data preprocessing and dataset creation;

[0124] The Φ-OTDR system is used to acquire vibration signals within a designated test area, while simultaneously manually recording the event type, spatial location, and placement scenario of different vibrations. The Φ-OTDR system outputs multiple 1024×781 MAT files.

[0125] The Φ-OTDR signal is processed using a high-pass filter (instrument preset high-pass filter coefficient b2, cutoff frequency 2Hz, pass frequency 5Hz, stopband attenuation -80dB). Then it is processed using a low-pass filter (instrument preset low-pass filter coefficient b3, cutoff frequency 150Hz, pass frequency 100Hz, stopband attenuation -80dB) to finally obtain the preprocessed spatiotemporal signal.

[0126] By combining low-pass and high-pass filters, their effects on different vibration events were compared. For example, falling objects and digging require low-pass filtering to maintain the trend of the main signal, while knocking or electric drills and hammers require high-pass filtering to extract high-frequency details. Figure 10 , Figure 11 These visualizations more clearly demonstrate the impact of filtering on signal quality and feature preservation.

[0127] The spatiotemporal signal was converted into a 1024×781 spatiotemporal image. Based on manually recorded spatial locations and laying scenarios, labels were used to annotate the laying scenario type and vibration area, forming a location dataset. In the location dataset, background noise refers to signals without vibration events; these samples are not labeled with laying scenario type or vibration area. Here, 1024 represents the length of each data sample in the time dimension, and 781 represents the number of spatial sampling points.

[0128] The location dataset was randomly divided into training, validation, and test sets in a 7:2:1 ratio. Detailed information about the dataset is shown in Table 3. Table 3 describes vibration events in multi-layout scenarios.

[0129] Table 3

[0130]

[0131]

[0132] exist Figure 12 , 13 In the diagram, the green box represents the underground laying scenario, and the red box represents the underground laying scenario. Regardless of the laying scenario, the vibration signal will spread to the surrounding area. However, the vibration diffusion characteristics are more significant in the underground scenario, the signal intensity is more concentrated, and the vibration area is easier to distinguish; while in the underground scenario, the signal propagation is affected by the casing and soil medium, the diffusion range is smaller, the vibration intensity is weaker, and the signal characteristics are more difficult to distinguish.

[0133] 3. Parameter settings;

[0134] In this embodiment, pytorch1.11.0 is selected as the deep learning framework. All experiments are conducted on a workstation equipped with a 64 GB memory NVIDIA GeForce RTX 2080 GPU, with CUDA version 11.3. In this experiment, the batch size of all positioning models is set to 8, and the epoch is set to 100.

[0135] 4. Evaluation index;

[0136] When evaluating the models in the positioning method, the size of the model is measured by Parameters and GFLOPs in this embodiment. Parameters represent the spatial complexity of the model, while GFLOPs (floating-point operations per second) are used to describe the computational complexity. The test time reflects the average positioning time of the model.

[0137] To evaluate the performance of the vibration region positioning, Precision, Recall, mAP@50 and mAP@50-95 are used as key indicators, each of which reflects different aspects of the positioning process:

[0138] Precision focuses on the accuracy of vibration region prediction. Higher Precision can ensure that the detected vibration region truly corresponds to the actual vibration source, reducing false detection of background noise or environmental interference as vibration events.

[0139] Recall emphasizes the model's ability to capture all existing vibration regions. Higher Recall means that the system can effectively detect weak or small-scale vibrations, which is crucial for detecting subtle disturbances in underground power cables.

[0140] The mean average precision (mAP@50) at an IoU threshold of 0.5 evaluates the correctness of vibration region positioning. This metric is particularly suitable for practical applications where rough localization of vibration events is sufficient for early warning and monitoring, as it allows for moderate overlap with the true value (intersection over union IoU = 50%).

[0141] The mean average precision (mAP@50-95) at an IoU threshold of 0.5 to 0.95 provides a more rigorous evaluation by assessing the positioning accuracy at multiple IoU thresholds. This reflects the model's performance in accurately dividing the boundaries of vibration regions, which is crucial for distinguishing overlapping or closely spaced vibration events.

[0142] The bounding box regression loss (Boxloss) is used to measure the deviation between the bounding box predicted by the model and the true labeled box. In the vibration area positioning, the bounding box corresponds to the vibration occurrence area predicted by the model. The smaller the Boxloss, the more accurately the model locates the actual vibration area range.

[0143] The class classification loss (Clsloss) measures whether the model's judgment of the vibration area class (i.e., different laying scenarios or vibration event types) is accurate. In the vibration area positioning, Clsloss reflects the model's ability to identify whether the area belongs to a buried or underground laying scenario.

[0144] To verify the effectiveness of the PLGS-YOLO model described in the embodiment, ablation experiments and comparative experiments are performed, and the results of the test set in the positioning data set are analyzed. In the ablation experiment, by gradually introducing the PLMSAM module, the GSConv module and the VoVGSCSP module (referred to as GSVO), and the SPPFDMSCA module, the improvement effect of each module is verified. In the comparative experiment, the PLGS-YOLO model is compared with the YOLO series lightweight models (such as YOLOv3-tiny, YOLOv5-tiny, YOLOv8n, etc.), and the results show that the PLGS-YOLO model achieves the best performance in terms of detection accuracy, computational complexity and real-time performance.

[0145] First, the PLGS-YOLO model is used for ablation experiment;

[0146] Based on YOLOv11n, combined with PLMSAM, GSConv and VoVGSCSP, SPPFDMSCA, ablation experiments are performed, where GSVO represents GSConv and VoVGSCSP, and the PLGS-YOLO model in different ablation experiment results is shown in Table 4.

[0147] Table 4

[0148]

[0149] First, the C2PSA of the backbone network is improved using the PLMSAM module. The experimental results show that the model parameters are reduced, the computational complexity is basically unchanged, the Precison is improved from 93.99% to 94.30%, the mAP@50 is improved from 97.27% to 97.34%, the mAP@50-95 is improved from 59.94% to 60.53%, the Recall is improved from 94.05% to 94.72%, and the test time is reduced from 6.896ms to 6.790ms. PLMSAM improves the performance of YOLOv11n by improving the PSA module and discarding the original model's cross-connection structure. Then, we replace the Conv in the neck network with GSConv and replace C3k2 with VoVGSCSP. The experimental results show that GSConv and VoVGSCSP fully extract the deep features of PLMSAM. Compared with only improving PLMSAM in YOLOv11n, the model's parameters, computational complexity, and test time are reduced, and the Precison, mAP@50, mAP@50-95, and Recall are improved. However, compared with YOLOv11n, the Recall is still decreased by 0.12%. The SPPF of YOLOv11n loses part of the information of the deep features, which is not conducive to feature fusion in the neck network. Based on the improved PLMSAM and GSVO in YOLOv11n, we replace SPPF with SPPFDMSCA. The experimental results show that although some indicators are not as good as the remaining ablation models, all performance indicators of PLGS-YOLO are better than those of the original YOLOv11n model. The Precison is improved from 93.99% to 94.46%, the Recall is improved from 94.05% to 94.55%, the mAP@50 is improved from 97.27% to 97.51%, and the mAP@50-95 is improved from 59.94% to 61.27%. The model's computational complexity, computational complexity, and test time are reduced.

[0150] Then, the performance analysis of the PLGS-YOLO model is performed.

[0151] The mainstream YOLO series lightweight models are selected to verify the effectiveness of PLGS-YOLO. These models include YOLOv3-tiny, YOLOv5-tiny, YOLOv7-tiny, YOLOv8n, YOLOv8n+CBAM, YOLOv9n, YOLOv10n, and YOLOv11n. The experiments are conducted under the same conditions. Among them, YOLOv3-tiny and YOLOv8n+CBAM have been applied to the vibration positioning task, and the remaining models are directly transplanted to the vibration image processing to conduct comparative analysis. All experiments are conducted under the same conditions, and the results are shown in Table 5. Table 5 is a comparison of different target detection models.

[0152] Table 5

[0153]

[0154]

[0155] YOLOv5 simplifies the backbone network to CSPDarknet-tiny, which greatly reduces the number of layers and residual blocks. YOLOv7-tiny only retains the basic spatial pyramid pooling module of SPPCSPC, reducing the level of feature fusion. Although the YOLOv5-tiny and YOLOv7-tiny models meet the real-time requirements of underground power cable scene detection, the detection accuracy is significantly lower than our improved model. Compared with YOLOv5-tiny, the Precision of PLGS-YOLO increases by 1.77%, the Recall increases by 1.67%, the mAP@50 increases by 1.13%, and the mAP@50-95 increases by 2.01%. Compared with YOLOv7-tiny, the Precision of PLGS-YOLO increases by 3.97%, the Recall increases by 5.74%, the mAP@50 increases by 4.19%, and the mAP@50-95 increases by 7.70%.

[0156] The network structure of YOLOv3-tiny model is relatively larger and more complex, and the fixed parameters of anchor frame cannot fully adapt to multi-scale vibration signals, which requires more computing resources for detection and has poor real-time performance. Compared with PLGS-YOLO, the GFLOPS of YOLOv3-tiny model increases by 12.60, the parameters increase by 9.55M, and the test time increases by 1.18ms.

[0157] The SPPF of YOLOv8n and YOLOv8n-CBAM obtains deep features through multi-scale pooling, but features of different scales may contain redundant information, which will affect the detection accuracy if directly used for feature fusion in the neck network. Compared with PLGS-YOLO, the Precision of YOLOv8n decreases by 1.41%, the Recall decreases by 0.51%, the mAP@50 decreases by 0.27%, and the mAP@50-95 decreases by 1.09%. The Precision of YOLOv8n-CBAM decreases by 2.50%, the Recall decreases by 0.69%, the mAP@50 decreases by 0.75%, and the mAP@50-95 decreases by 1.53%.

[0158] YOLOv9n replaces the traditional convolutional layer and residual block with GELAN as part of the backbone network, improving the ability of feature extraction, but GELAN adopts a multi-branch structure, which will significantly increase the computational load and test time of the model. Compared with PLGS-YOLO, the GFLOPS of YOLOv9n model increases by 5.70, and the test time increases by 3.46 ms.

[0159] YOLOv10n uses PSA to further extract deep features of SPPF, focusing on key information of features and reducing interference of redundant information, improving the feature fusion efficiency of the neck network, but the multi-head attention mechanism inside PSA will cause partial local detail information to be lost, which is not conducive to precise positioning of the vibration signal and increases the number of data to be detected. Compared with PLGS-YOLO, the Precision of YOLOv10n decreases by 2.99%, the Recall decreases by 1.55%, the mAP@50 decreases by 1.53%, and the mAP@50-95 decreases by 4.02%.

[0160] By comparing the Precision, Recall, mAP@50, mAP@50-95 of YOLOv3-tiny, YOLOv8n, YOLOv8n+CBAM, YOLOv9n, YOLOv10n, the detection accuracy of the PLGS-YOLO model is the best. The GFLOPS, parameters, and test time of PLGS-YOLO are the lowest, meeting the real-time performance and having small deployment difficulty. In summary, PLGS-YOLO has the best overall performance in detecting and positioning the underground power cable vibration signal collected by the OTDR system. -OTDR system.

[0161] As shown in Figure 14 To further evaluate the performance of PGSDYOLO in vibration area positioning, the Box loss, Cls loss, Precision, Recall, mAP@50 (when IoU threshold is 0.5), and mAP@50-95 (when IoU threshold is 0.5 to 0.95) of PGSDYOLO on the validation set are compared with existing lightweight YOLO models. With the increase of iteration number, PLGS-YOLO can always obtain higher performance.

[0162] In this embodiment, the underground power cable vibration signal collected by the OTDR system is used to evaluate the performance of the PGSDYOLO model in the vibration area positioning. The samples collected in real time by the OTDR are used for laying scene detection and positioning, and the PLGS-YOLO correctly labels the laying scene type in each image, wherein the green box represents the buried laying scene, and the red box represents the underground laying scene, and the different types of vibration signal regions can be accurately positioned, and part of the visualization results are as follows Figure 15 .

[0163] The buried motorcycle signal is represented as a continuous high-frequency energy stripe in the time-space graph, and the signal intensity is high and the range is wide. The detection box completely covers the high-frequency signal region and is not disturbed by other noise. The buried knocking signal is irregularly distributed, and there are multiple short-time high-energy shock waves, and the background noise is less. Multiple detection boxes label each high-energy signal region to ensure complete signal coverage. The buried excavation signal is represented as an intermittent strong signal in the time-space graph, and the frequency is similar to the background noise, but the energy is higher. The model successfully filters the background signal interference and accurately frames the vibration signal region. The buried trolley signal is represented as a low-frequency, continuous energy stripe, and the distribution is relatively uniform, and the signal intensity is low. The detection box completely labels the low-frequency signal region and does not miss important information.

[0164] The downhole drilling signal is concentrated in the middle of the image and is represented as a continuous high-frequency energy stripe with a moderate distribution range. The detection box accurately labels the signal region and avoids background interference. The downhole walking signal is intermittent, with sparse stripes but obvious energy, and the background noise is weak. The detection box completely labels all intermittent signals without missing the main region. The downhole knocking signal is scattered and represented as multiple short-time high-energy signal peaks, and the interval region is background noise. The model covers all high-energy signal regions through multiple detection boxes without being disturbed by noise. The downhole falling object signal is represented as a continuous high-energy stripe in the time-space graph, and the signal gradually decays with dynamic characteristics. The detection box covers the entire signal path from high energy to low energy, and the signal region is complete.

[0165] In summary, the PLGS-YOLO has superior vibration region positioning and scene detection performance in both buried and underground laying scenes.

[0166] The underground power optical cable vibration region positioning method for multiple laying scenes described in the embodiment uses the PLGS-YOLO to detect the laying scene and extract the regional vibration signal. The model's rationality is verified from the model's real-time performance, ease of deployment, detection accuracy, and positioning accuracy. The PLGS-YOLO replaces the corresponding position modules with PLMSAM, GSConv, and VoVGSCSP, SPPFDMSCA based on YOLOv11n. The improved PLGS-YOLO is optimal in Precison, mAP@50, mAP@50-95, and Recall, and the model's parameter amount, computational complexity, and test time are relatively low.

[0167] Any combination of the technical features in the above-described embodiments can be made, and for the sake of brevity, not all possible combinations are described, however, as long as there is no conflict, any combination of the technical features should be considered within the scope of the present disclosure.

[0168] The above-described embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for ordinary skilled persons in the art, some modifications and improvements can be made without departing from the concept of the present application, and these are within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for locating vibration zones in underground power optical cables across multiple scenarios based on the PLGS-YOLO model, characterized by: This method is implemented by the following steps: Step 1: Preprocess the acquired vibration signal and convert the preprocessed signal into a spatiotemporal diagram; Step 2: Locating the vibration area; By constructing a PLGS-YOLO model and training the PLGS-YOLO model, the vibration area in the spatiotemporal map obtained in step one is located using the trained PLGS-YOLO model, and the located spatiotemporal map and the corresponding laying scene type are generated. The PLGS-YOLO model improves the YOLOv11n network by replacing the SPPF module and C2PSA module with the SPPFDMSCA module and PLMSAM module, respectively, in the backbone network. Replace the convolutional layers in the neck network with the GSConv module and the C3k2 module with the VoVGSCSP module; The DMSCA module is introduced into the SPFFDMSCA module to enhance the global feature extraction capability of the backbone network. The LMSCA module is used in the PLMSAM module to obtain deep feature maps; The spatiotemporal map is input into the YOLOv11n backbone network, and feature enhancement is performed through the dynamic multi-scale convolutional attention mechanism in the SPFFDMSCA module. The obtained global feature map is then input into the PLMSAM module. The global feature map is then input into the neck network after undergoing multi-scale convolution operations through the LMSCA module in the PLMSAM module. The feature maps after multi-scale convolution operations of the GSConv and VoVGSCSP modules in the neck network are fused to obtain the spatiotemporal map after localization and the corresponding laying scene type through the detection head.

2. The method for locating vibration zones in underground power optical cables in multiple scenarios based on the PLGS-YOLO model according to claim 1, characterized in that: In step one, the acquired Φ-OTDR vibration signal is denoised by high-pass and low-pass filtering to obtain a preprocessed spatiotemporal signal; the spatiotemporal signal is then converted into a spatiotemporal map.

3. The method for locating vibration zones in underground power optical cables in multiple scenarios based on the PLGS-YOLO model according to claim 1, characterized in that: The SPFFDMSCA module also includes a first convolutional layer, multiple max pooling layers, a Concat module, and a second convolutional layer; The feature map output from the convolutional layers and C3k2 module in the backbone network is processed by the first convolutional layer, multiple max pooling layers, and multiple residual connection operations of the SPFFDMSCA module. After passing through the second convolutional layer, it is enhanced by the DMSCA module to output a global feature map.

4. The method for locating vibration zones in underground power optical cables in multiple scenarios based on the PLGS-YOLO model according to claim 3, characterized in that: In the DMSCA module, multi-scale convolution operations are introduced. Parallel convolution extraction is performed on the input feature map X using 3×3, 5×5, and 7×7 convolution kernels, respectively. The features extracted at the three scales are then concatenated and fused along the channel dimension to obtain the multi-scale feature map X. ms ; The multi-scale feature map X ms An attention map A is generated through a 1×1 convolution and activated by a sigmoid function; simultaneously, the multi-scale feature map X... ms After 1×1 convolution, key-value pairs K and V are generated respectively. K is normalized by the Softmax function and multiplied with V to calculate the context enhancement feature Y. The attention map A and the context enhancement feature Y are then added to the input feature map X to obtain the output global feature map.

5. The method for locating vibration zones in underground power optical cables in multiple scenarios based on the PLGS-YOLO model according to claim 1, characterized in that: The PLMSAM module includes batch normalization layer 1, LMSAM module, batch normalization layer 2 and FFN module. The LMSAM module includes spatial attention module and LMSCA module. The PLMSAM module receives the global feature map from the SPFFDMSCA module as input. First, it obtains a multi-branch fused feature map through the residual connection structure composed of batch normalization layer 1 and the LMSAM module. Then, it obtains a deep feature map through the residual connection structure composed of batch normalization layer 2 and the FFN module. Finally, the deep feature map obtained by the PLMSAM module is input to the upsampling module 1 and the Concat module 4 of the neck network.

6. The method for locating vibration zones in underground power optical cables in multiple scenarios based on the PLGS-YOLO model according to claim 5, characterized in that: The LMSCA module includes a 5×5 convolutional layer, a multi-branch convolutional layer, a 1×1 convolutional layer, and an average pooling layer. Local features are obtained through the 5×5 convolutional layer, the multi-branch convolutional layer captures multi-scale target relationships in different channels, and the original feature map is weighted by the average pooling layer and the 1×1 convolutional layer to obtain a weighted deep feature map.

7. The method for locating vibration zones in underground power optical cables in multiple scenarios based on the PLGS-YOLO model according to claim 3, characterized in that: The neck network includes multiple upsampling modules, multiple Concat modules, multiple GSConv modules, and multiple VoVGSCSP modules; The feature map output from the C3k2 module 2 in the backbone network is fed into the second Concat module in the neck network. The feature map after passing through the second Concat module is input into the second VoVGSCSP module, which outputs a compressed shallow high-resolution feature map. This map is then fed into the third Concat module via the first GSConv module and simultaneously into the detection head 1 of the head network for shallow vibration area localization and laying scene prediction. The feature map output from module 3 of C3k2 in the backbone network is transmitted to the first Concat module in the neck network. After passing through the first Concat module, it is input to the first VoVGSCSP module. The feature map output from the first VoVGSCSP module is input to the second upsampling module and the third Concat module. After being upsampled by the second upsampling module, it is input to the second Concat module. After passing through the third Concat module, it is input to the third VoVGSCSP module. The output of the third VoVGSCSP module is a compressed mid-layer high-resolution feature map. The high-resolution feature map is input to the fourth Concat module through the second GSConv module, and simultaneously input to the detection head 2 of the head network for mid-layer vibration region localization and laying scene prediction. The feature map output from the PLMSAM module in the backbone network is sent to the first upsampling module and the fourth Concat module in the neck network. The feature map is upsampled by the first upsampling module and then input to the first Concat module. After passing through the fourth Concat module, it is input to the fourth VoVGSCSP module. The output of the fourth VoVGSCSP module is a deep fusion feature map, which is input to the detection head 3 in the head network for deep vibration area localization and laying scene prediction.

8. The method for locating vibration zones in underground power optical cables in multiple scenarios based on the PLGS-YOLO model according to claim 7, characterized in that: Each GSConv module consists of a convolutional layer Conv, a depthwise separable convolutional layer DWConv, a Concat module, and a Shuffle module. The feature map with C1 channel is used to generate a feature map with C2 / 2 channels through a convolutional layer. At the same time, the feature map with C2 / 2 channels is processed through a depthwise separable convolutional layer to generate another feature map with C2 / 2 channels. The convolutional layer generates a feature map with C2 / 2 channels, and the depthwise separable convolutional layer generates another feature map with C2 / 2 channels. These feature maps are then connected by the Concat module and integrated through the channel fusion mechanism of the Shuffle module to finally generate an output feature map with C2 channels.

9. The method for locating vibration zones in underground power optical cables in multiple scenarios based on the PLGS-YOLO model according to claim 7, characterized in that: Each VoVGSCSP module consists of four convolutional layers, two GSConv modules, and a Concat module. A portion of the C1 channel feature map is used for shallow feature extraction through the first convolutional layer to obtain the C1 / 2 channel feature map. The C1 / 2 channel feature maps are then used for deep feature extraction through the second convolutional layer and two GSConv modules, respectively. Finally, the residuals are summed to obtain the C2 / 2 channel deep feature map. The other part of the C1 channel feature map is extracted through the third convolutional layer to obtain the C2 / 2 channel backbone feature map; The C2 / 2 channel deep feature map and the C2 / 2 channel backbone feature map are concatenated and feature extracted using the Concat module and the fourth convolutional layer to obtain the C2 channel fused feature map.

Images