Event detection methods, systems, devices, and storage media for optical cable detection

Abstract

The application relates to the field of optical fiber sensing, and more particularly to an event detection method, system and device for optical cable detection and a storage medium. The method comprises the following steps: inputting optical fiber data of an optical cable into a first processing unit for processing to obtain space-time features; inputting the optical fiber data into a second processing unit for processing to obtain frequency domain features; inputting the space-time features and the frequency domain features into an attention processing unit after splicing for feature enhancement to obtain enhanced features; inputting the enhanced features into a bottleneck layer processing unit for processing to obtain bottleneck features; wherein the bottleneck layer processing unit comprises at least two bottleneck layer processing subunits, and the bottleneck layer processing subunits can perform feature transformation processing on the enhanced features; and inputting the bottleneck features into a backbone network for optical fiber event detection to obtain an optical fiber event detection result corresponding to the optical fiber data. The application can improve the accuracy of optical fiber event detection in an optical cable.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing, and more specifically, to an event detection method, system, device, and storage medium for optical cable detection. Background Technology

[0002] Fiber optic event detection refers to the use of distributed fiber optic sensing technology, such as distributed optical fiber vibration sensing systems (DVS), to sense signals along the fiber optic line and detect various external events such as construction, intrusion, and environmental interference. Current mainstream detection methods include detection models based on fixed thresholds or traditional machine learning, spatiotemporal feature extraction models, and deep learning convolutional neural networks. However, these approaches each have drawbacks. For example, most detection models can only extract features from a single dimension, failing to fully integrate the interaction relationships of multi-dimensional features. They also typically have poor real-time performance, relying on a large number of parameters and computations, which is unfriendly to the performance requirements of processing equipment and real-time processing. Furthermore, existing detection models have weak generalization capabilities, lacking adaptability and robustness in scenarios with strong noise interference, small sample sizes, or low signal-to-noise ratio events. These shortcomings of existing solutions often make it difficult to achieve high-precision detection and classification of fiber optic events. Summary of the Invention

[0003] This invention provides an event detection method, system, device, and storage medium for optical cable inspection, which improves the accuracy of optical fiber event detection in optical cables.

[0004] According to a first aspect of this application, a fiber optic event detection method for optical cable inspection is provided, the method comprising: The optical fiber data from the optical cable is input into the first processing unit for processing to obtain the spatiotemporal characteristics; The fiber optic data is input into the second processing unit for processing to obtain frequency domain characteristics; The spatiotemporal features and the frequency domain features are concatenated and then input into the attention processing unit for feature enhancement to obtain enhanced features. The enhanced features are input into the bottleneck layer processing unit for processing to obtain bottleneck features; wherein, the bottleneck layer processing unit includes at least two bottleneck layer processing sub-units, and the bottleneck layer processing sub-units are capable of performing feature transformation processing on the enhanced features. The bottleneck features are input into the backbone network for fiber optic event detection, and the fiber optic event detection results corresponding to the fiber optic data are obtained.

[0005] Understandably, by extracting the spatiotemporal and frequency domain features of optical fiber data in parallel and using the attention processing unit for adaptive fusion and enhancement, deep collaboration and complementarity of multi-domain information are achieved. This enables not only the effective differentiation of complex optical fiber events with similar spatiotemporal features, but also the capture of weak signal traces existing only in the frequency domain, significantly improving the accuracy and precision of detection. At the same time, it also solves the problem of poor performance of the bottleneck layer processing unit in lightweight small and medium-sized dataset scenarios.

[0006] Optionally, the first processing unit includes a first depthwise separable convolutional unit and a second depthwise separable convolutional unit connected in sequence. The step of inputting the optical fiber data of the optical cable into the first processing unit for processing to obtain spatiotemporal characteristics includes: The fiber data is input into the first depth-separable convolutional unit for processing to obtain intermediate spatiotemporal features; The intermediate spatiotemporal features are input into the second depthwise separable convolutional unit for processing to obtain the spatiotemporal features.

[0007] Understandably, by cascading two levels of depthwise separable convolutional units, a highly efficient dedicated channel for spatiotemporal feature extraction is constructed. This enables multi-level filtering and feature abstraction of fiber optic data from coarse to fine, more fully revealing the spatiotemporal components and patterns related to fiber optic events within the data. Compared to a single convolutional layer, this application significantly enhances the expressive power and discriminative power of spatiotemporal features, laying a solid foundation for subsequent accurate event detection.

[0008] Optionally, the second processing unit includes a downsampling unit, a transformation unit, and a spectrum analysis unit; The step of inputting the fiber optic data into the second processing unit for processing to obtain frequency domain features includes: The fiber optic data is input into the downsampling unit for downsampling processing to obtain fiber optic downsampled data. The fiber optic downsampled data is input into the transformation unit for fast Fourier transform to obtain frequency domain data; The frequency domain data is input into the spectrum analysis unit for log-Mel spectrum analysis to obtain the frequency domain characteristics.

[0009] Understandably, downsampling effectively compresses the data size and improves overall processing efficiency; the Fast Fourier Transform converts the time-domain signal to the frequency domain, laying the foundation for frequency domain feature extraction. Most importantly, log-Mel spectrum analysis further simulates the nonlinear perception of human hearing, optimizing the scaling of the frequency axis, thereby significantly enhancing the ability of frequency domain features to distinguish subtle differences in similar events, thus forming a highly efficient and discriminative frequency domain feature extraction channel.

[0010] Optionally, the bottleneck layer processing unit includes at least three bottleneck layer processing sub-units connected in sequence; The step of inputting the enhanced features into the bottleneck layer processing unit for processing to obtain bottleneck features includes: The enhanced features are input into three sequentially connected bottleneck layer processing units for feature transformation to obtain bottleneck features.

[0011] Understandably, by deeply refining the enhanced features through at least three cascaded bottleneck layer processing units, a multi-level, progressive feature optimization structure is formed, which significantly enhances the ability to abstract and integrate the features corresponding to complex fiber optic events, thereby outputting highly refined and discriminative bottleneck features, providing a stable and reliable high-level information foundation for subsequent event detection.

[0012] Optionally, the bottleneck layer processing unit includes a deformable attention visual transformation processing unit, a feature compression processing unit, and a third depthwise separable convolution unit connected in sequence.

[0013] Understandably, a deformable attention visual transformation processing unit is integrated within each bottleneck layer processing unit, enabling it to dynamically focus on irregular or critical fiber optic event feature regions, effectively enhancing the perception of complex signal patterns. Combined with a feature compression processing unit for efficient feature integration and transfer, and a third depthwise separable convolution unit for lightweight local feature extraction, an optimized balance between computational resources and feature representation capabilities is achieved.

[0014] Optionally, the backbone network includes a fourth depthwise separable convolutional unit, a pooling unit, and a detection unit; The step of inputting the bottleneck features into the backbone network for fiber optic event detection to obtain the fiber optic event detection results corresponding to the fiber optic data includes: The bottleneck features are input into the fourth depthwise separable convolutional unit to obtain convolutional features; The convolutional features are input into the pooling unit for pooling processing to obtain global features; The global features are input into the detection unit to perform fiber optic event detection, and the fiber optic event detection result corresponding to the fiber optic data is obtained.

[0015] Understandably, lightweight, deeply separable convolutional units are used to finally map and fine-tune the bottleneck features, further optimizing their feature representation. Pooling units can aggregate global contextual information to generate global features that are insensitive to scale changes and contain complete event semantics. Finally, the detection unit achieves efficient and accurate fiber optic event detection and classification decisions based on highly refined global features, thereby reliably outputting specific detection results for fiber optic events.

[0016] Optionally, before the step of inputting the optical fiber data of the optical cable into the first processing unit for processing to obtain spatiotemporal characteristics, the method further includes: Obtain the raw fiber data of the optical cable; Convert the raw fiber optic data into a grayscale image; The grayscale image is color-rendered to obtain a heatmap containing several color channels, and the heatmap serves as the fiber optic data.

[0017] Understandably, converting the raw fiber optic data into grayscale images achieves preliminary structuring and visualization of the raw fiber optic data. Further, generating a heatmap containing several color channels through color rendering maps signal intensity or frequency differences to different color dimensions, significantly enhancing the human eye's and machine's ability to perceive subtle changes and patterns in signals. This provides richer and more discriminative input data for subsequent processing, thus laying an important preliminary foundation for improving the accuracy of the entire fiber optic event detection process.

[0018] According to a second aspect of this application, an event detection system for optical cable inspection is provided, the system comprising: The first processing unit is used to process the optical fiber data of the optical cable to obtain spatiotemporal characteristics; The second processing unit is used to process the optical fiber data to obtain frequency domain features; An attention processing unit is used to perform feature enhancement on the spliced ​​frequency domain features and spatiotemporal features to obtain enhanced features; A bottleneck layer processing unit is used to process the enhanced features to obtain bottleneck features; wherein, the bottleneck layer processing unit includes at least two bottleneck layer processing sub-units, and the bottleneck layer processing sub-units are capable of performing feature transformation processing on the enhanced features. The backbone network is used to detect fiber optic events based on the bottleneck characteristics, and to obtain the fiber optic event detection results corresponding to the fiber optic data.

[0019] According to a third aspect of this application, an electronic device is provided, comprising: Memory, used to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements the event detection method for optical cable detection described in the first aspect above.

[0020] According to a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the event detection method for optical cable detection described in the first aspect above.

[0021] Based on any of the above aspects, the event detection method, system, device, and storage medium for optical cable detection provided in this application embodiment can achieve the following effects: A parallel multidimensional feature extraction architecture is constructed to fuse the temporal, spatial, and frequency domain features of optical fiber data, thereby improving the accuracy of detection. This application breaks through the limitations of traditional methods that rely on a single data domain, enabling the effective capture of optical fiber events that are weak or hidden in spatiotemporal features but only show traces in the frequency domain. The complementarity and cross-validation of spatiotemporal and frequency domain features significantly improve the overall accuracy and reliability of detecting various optical fiber events in complex field environments.

[0022] We introduce depthwise separable convolution and log-Mel spectrum analysis as feature extraction tools for fiber optic data, achieving accurate feature extraction from fiber optic data. By utilizing depthwise separable convolution to obtain spatiotemporal features, and performing high-order abstraction of the frequency domain features of fiber optic data from the perspective of human auditory perception bionics, we highlight the key mid-to-low frequency information for fiber optic event identification. This enables us to effectively distinguish between different fiber optic events that are highly similar and difficult to distinguish at the single spatiotemporal feature level, such as knocking and digging, achieving more refined fiber optic event classification. At the same time, we also solve the problem of poor performance of subsequent use of deformable attention visual transformation units in lightweight small to medium-sized dataset scenarios.

[0023] The method employs lightweight units, including depthwise separable convolutional units and deformable attention visual transformation units, to improve the real-time performance of event detection. The use of depthwise separable convolutional units greatly reduces the number of parameters and computational load. The deformable attention visual transformation unit enables adaptive focusing on the features of irregular fiber optic events, resulting in detection efficiency far exceeding that of traditional attention methods. The collaborative work of each lightweight processing unit allows the entire detection method to maintain or even improve the recognition performance of complex events and small target events while significantly reducing computational complexity, thus meeting the stringent real-time requirements of practical event detection applications. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of an event detection method for optical cable inspection provided in this embodiment.

[0026] Figure 2 This is a flowchart illustrating the first processing unit's processing of optical fiber data in this embodiment.

[0027] Figure 3 This is a flowchart for acquiring optical fiber data provided in this embodiment.

[0028] Figure 4 This is a flowchart illustrating the second processing unit's processing of optical fiber data in this embodiment.

[0029] Figure 5 This is a flowchart illustrating the bottleneck characteristics of the backbone network processing provided in this embodiment.

[0030] Figure 6 This is a schematic diagram of the functional modules of an event detection system for optical cable detection provided in this embodiment.

[0031] Figure 7 This is a schematic diagram of the structure of the electronic device provided in this embodiment. Detailed Implementation

[0032] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this application. To better illustrate the following embodiments, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0033] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] Current event detection methods for optical cable inspection suffer from insufficient multi-dimensional feature fusion, high computational complexity, and weak generalization ability, making it difficult to achieve high-precision real-time detection in complex scenarios.

[0036] This embodiment provides a technical solution that can solve the above problems. The specific implementation of this application will be described in detail below with reference to the accompanying drawings.

[0037] like Figure 1 As shown, this embodiment provides an event detection method for optical cable inspection, which can be further divided into the following steps: S100. Input the optical fiber data of the optical cable into the first processing unit for processing to obtain the spatiotemporal characteristics; In this embodiment, the fiber optic data of the optical cable can be acquired and processed using a phase-sensitive optical time domain reflectometer (φ-OTDR). This phase-sensitive ODR includes, but is not limited to, a distributed optical fiber vibration sensing system (DVS) and a distributed optical fiber audio sensing system (DAS). It is understood that the acquired raw fiber optic data needs to be processed to obtain fiber optic data that conforms to the input data format of the first processing unit. This allows the first processing unit to better process fiber optic data in a standard input data format, thereby obtaining accurate spatiotemporal characteristics.

[0038] Understandably, the spatiotemporal features can simultaneously embody both time-domain and spatial-domain features. Time-domain features contain information about dynamic changes over time, while spatial-domain features contain information about spatial distribution and correlation. In fiber optic sensing, these spatiotemporal features typically manifest as the sequence or pattern of fiber optic signal evolution over time at each detection point distributed along the spatial location of the fiber. In this embodiment, complete data on the location, propagation range, and duration of fiber optic events can be captured simultaneously based on spatiotemporal features. Time-domain features reveal the dynamic behavior, rhythm, and duration of fiber optic events; for example, instantaneous impacts or continuous vibrations of the fiber optic cable can be identified through time-domain features. Spatial-domain features can locate the point of occurrence of fiber optic events and determine the scope of influence. The fusion of time-domain and spatial-domain features significantly improves the ability to distinguish event types, the accuracy of event location determination, and the robustness in identifying real fiber optic event patterns in noisy environments.

[0039] Specifically, the first processing unit includes a first depthwise separable convolutional unit and a second depthwise separable convolutional unit connected in sequence. In this embodiment, the first and second depthwise separable convolutional units can be specifically configured as depthwise separable convolutional networks (DWCONV). The depthwise separable convolutional network is a lightweight network that overcomes the shortcomings of traditional methods, such as high computational cost, poor real-time performance, and easy loss of early features, caused by using one-dimensional convolutional neural networks – bidirectional long short-term memory networks (1DCNN-BiLSTM) to extract spatiotemporal features. In actual optical cable inspection, it can improve the processing efficiency of optical fiber data, thereby meeting the real-time requirements of optical fiber event detection.

[0040] like Figure 2 As shown, the step of inputting the optical fiber data of the optical cable into the first processing unit for processing to obtain spatiotemporal characteristics may include the following steps: S110. Input the optical fiber data into the first depth-separable convolutional unit for processing to obtain intermediate spatiotemporal features; S120. The intermediate spatiotemporal features are input into the second depthwise separable convolutional unit for processing to obtain the spatiotemporal features.

[0041] In this embodiment, the fiber optic data is input into a first depthwise separable convolutional unit and a second depthwise separable convolutional unit connected in sequence for processing. This enables the construction of spatiotemporal feature abstractions from local details to global patterns at extremely low computational cost. The first depthwise separable convolutional unit can extract basic intermediate spatiotemporal features from the fiber optic data, while the second depthwise separable convolutional unit can synthesize more complex and discriminative spatiotemporal features from these intermediate features. This not only significantly improves the feature representation capability for complex events but also perfectly meets the requirements of real-time processing and high energy efficiency for fiber optic event detection.

[0042] In this embodiment, the fiber optic data can be input data of size a*b*c, where a*b*c represents the spatial dimension, temporal dimension, and data channel of the fiber optic data. Specifically, the data channel can be a color channel. For example, a can be 640, b can be 256, and c can be 3, meaning the fiber optic data can be specifically of size 640*256*3, where 640 is the spatial dimension data, 256 is the temporal dimension data, and 3 is the color channel data.

[0043] When the fiber optic data is input into the first depthwise separable convolutional unit for processing to obtain intermediate spatiotemporal features, the size of the intermediate spatiotemporal features can be (a / 2)*(b / 2)*d. For example, d can be 16, that is, the size of the intermediate spatiotemporal features can specifically be 320*128*16.

[0044] When the intermediate spatiotemporal features are input into the second depthwise separable convolutional unit for processing to obtain spatiotemporal features, the size of the spatiotemporal features can be (a / 2)*(b / 2)*2d. For example, the size of the spatiotemporal features can specifically be 320*128*32.

[0045] Specifically, such as Figure 3 As shown, before the step of inputting the optical fiber data of the optical cable into the first processing unit for processing to obtain the spatiotemporal characteristics, the method further includes the following steps: A1. Obtain the raw fiber data of the optical cable; A2. Convert the original optical fiber data into a grayscale image; A3. The grayscale image is color-rendered to obtain a heatmap containing several color channels, and the heatmap is used as the optical fiber data.

[0046] In this embodiment, the raw fiber data of the optical cable can be obtained through the aforementioned DVS device. This raw fiber data reflects external disturbances such as vibration and strain experienced along the fiber optic cable, specifically manifested as a multi-dimensional matrix of space, time, and signal strength. It is understood that the raw fiber data contains rich information in the time, space, and frequency domains. However, traditional fiber optic event detection in optical cables typically only uses the time and space domain information of the raw fiber data. This leads to difficulties in identifying the detailed event categories when facing complex optical cable environments, relying solely on spatiotemporal characteristics for detection. Furthermore, for some weak or unclear fiber optic events, their presence may only be detectable in the frequency domain. Therefore, in this embodiment, the three-dimensional information in the time, space, and frequency domains of the raw data is fully utilized to predict fiber optic events, thereby improving the accuracy of fiber optic event prediction and reducing misjudgments or omissions.

[0047] In this embodiment, the original fiber optic data is converted into a grayscale image, i.e., an image with only one color channel. Considering the needs of visualization and the standard format of input data in algorithm processing, the grayscale image needs to be converted into a heatmap with three color channels, and the heatmap is then input as fiber optic data into the first processing unit and the second processing unit for processing.

[0048] S200: Input the optical fiber data into the second processing unit for processing to obtain frequency domain characteristics; In this embodiment, frequency domain features are characteristics obtained by converting the optical fiber signal from the time domain to the frequency domain. These features can be represented as a spectrum or spectrogram. By revealing the unique frequencies of different optical fiber events, they directly reflect the physical nature of these events, thereby significantly improving the accuracy of optical fiber event classification. Simultaneously, frequency domain analysis can effectively separate signals in specific frequency bands from broadband noise, significantly enhancing anti-interference capability and robustness in low signal-to-noise ratio environments. Frequency domain features can supplement the spatiotemporal features in describing optical fiber events. Based on both spatiotemporal and frequency domain features, high-precision, highly adaptive optical fiber event detection can be achieved.

[0049] Specifically, the second processing unit includes a downsampling unit, a transformation unit, and a spectrum analysis unit; like Figure 4 As shown, the step of inputting the fiber optic data into the second processing unit for processing to obtain frequency domain features may include the following steps: S210. Input the optical fiber data into the downsampling unit for downsampling processing to obtain optical fiber downsampled data; In this embodiment, downsampling the fiber optic data enables data simplification and feature enhancement while preserving key information, directly reducing the computational load and memory consumption of subsequent processing and improving real-time performance. Simultaneously, downsampling, as an effective noise reduction method, filters out high-frequency noise and redundant details, making the dominant frequency components and event-related low-frequency features in the fiber optic data more prominent and stable, thereby improving the robustness of subsequent frequency domain feature extraction. Specifically, fiber optic downsampling data is obtained by inputting the fiber optic data into a downsampling unit. For example, the size of the fiber optic downsampling data can be a*(b / 8)*c, i.e., the size of the fiber optic downsampling data can be 640*32*3.

[0050] S220. Input the fiber downsampled data into the transformation unit to perform a fast Fourier transform to obtain frequency domain data; Preferably, the fiber downsampled data can be subjected to real-time fast Fourier transform (RFFT), which can convert the fiber downsampled data from a time-varying waveform into an energy spectrum that varies with frequency, thereby obtaining frequency domain data.

[0051] S230. Input the frequency shift data into the spectrum analysis unit to perform log-Mel spectrum analysis to obtain the frequency domain characteristics.

[0052] In this embodiment, the frequency domain data is input into a spectrum analysis unit for log-Mel spectrum analysis. Specifically, the frequency axis of the frequency domain data is nonlinearly mapped according to the Mel scale formula to simulate the characteristics of human hearing. Then, the spectrum of the frequency domain data is passed through a Mel filter bank with triangular bandpass to divide and integrate the energy of key frequency bands. Finally, the logarithm of the output energy of each filter in the Mel filter bank is taken, thereby compressing the frequency domain data and transforming it into a log-Mel spectrum with lower dimensionality, a dynamic range more consistent with perception, and the ability to highlight the discriminative spectral patterns of different fiber optic events. The log-Mel spectrum is the frequency domain feature. For example, the size of the frequency domain feature can be (a / 2)*(b / 2)*2d. For example, the specific size of the frequency domain feature can be 320*128*32. It is understood that the data dimension and size of the spatiotemporal feature and the frequency domain feature need to be the same to complete the subsequent stitching work.

[0053] S300: The spatiotemporal features and the frequency domain features are concatenated and then input into the attention processing unit for feature enhancement to obtain enhanced features; In this embodiment, the spatiotemporal features and the frequency domain features need to be concatenated before being input into the attention processing unit, thereby enabling better feature fusion of the spatiotemporal features and the frequency domain features. Specifically, the size of the concatenated feature is (a / 2)*(b / 2)*4d, for example, the size of the concatenated feature can be 320*128*64.

[0054] In this embodiment, the attention processing unit can be an efficient multi-scale attention (EMA) unit. Based on the efficient multi-scale attention processing unit, the spliced ​​features after splicing spatiotemporal features and frequency domain features are dynamically assigned feature weights, so that each sub-feature is evenly distributed in each feature group, thereby improving the ability to identify small target optical fiber events with weak signals.

[0055] S400. The enhanced features are input into the bottleneck layer processing unit for processing to obtain bottleneck features; wherein, the bottleneck layer processing unit includes at least two bottleneck layer processing sub-units, and the bottleneck layer processing sub-units are capable of performing feature transformation processing on the enhanced features. In this embodiment, the bottleneck layer processing unit is a highly efficient network structure in deep learning. It plays a crucial role in feature transformation within the bottleneck layer of existing deep learning systems, enabling it to refine and reduce features with minimal computational cost while maintaining almost no loss of representational power. This significantly reduces the number of parameters and computational burden, improves operational efficiency, and promotes deep fusion and enhancement of feature information. In this embodiment, the bottleneck layer processing unit comprises at least two bottleneck layer processing sub-units. These sub-units perform feature transformation on the enhanced features, allowing for more refined refinement and higher-order semantic combination of the enhanced features through a progressive and abstract approach. This enhances the expressive power of the acquired bottleneck features. Furthermore, this stacked structure of multiple bottleneck layer processing sub-units helps maintain high computational efficiency while increasing the depth of the bottleneck layer processing unit, further improving the ability to extract key information.

[0056] Specifically, the bottleneck layer processing unit includes at least three bottleneck layer processing sub-units connected in sequence, and the structure of each bottleneck layer processing sub-unit refers to the standard bottleneck structure in deep learning. The standard bottleneck structure is generally set as a multi-layer network structure of "transformation-compression-expansion"; specifically, in this embodiment, the bottleneck layer processing sub-unit includes a deformable attention visual transformation processing unit, a feature compression processing unit, and a third depthwise separable convolutional unit connected in sequence, wherein the deformable attention visual transformation processing unit undertakes the transformation function of features, the feature compression processing unit undertakes the compression function of features, and the third depthwise separable convolutional unit undertakes the expansion function of features.

[0057] In this embodiment, at least three bottleneck layer processing units can be configured with the same bottleneck layer processing unit structure. Each bottleneck layer processing unit internally includes a Deformable Attention Vision Transformer (DAViT). It is understood that the Vision Transformer (ViT) is a general network structure, and for some specific knowledge, such as the inherent locality and translation invariance of image data in this embodiment, there is no clear definition of prior knowledge. Therefore, if the Vision Transformer is required to learn the corresponding specific knowledge, it needs to be pre-trained with massive amounts of data to learn these visual rules that should be common sense from scratch. Otherwise, overfitting or poor performance is likely to occur on small to medium-sized datasets. The Deformable Attention Vision Transformer is also a type of Vision Transformer and still suffers from the aforementioned shortcomings. In this embodiment, a lightweight processing architecture is desired; therefore, it is necessary to consider reducing the size of the dataset for training the processing units while still ensuring accurate feature extraction. Therefore, in this embodiment, a depthwise separable convolutional unit is introduced to process the fiber data during the aforementioned processing of the deformable attention visual transformation processing unit, thereby avoiding the processing defects of the deformable attention visual transformation processing unit.

[0058] Specifically, before the fiber optic data is input into the bottleneck layer processing unit, a first processing unit has already performed basic feature extraction on the fiber optic data. In this embodiment, the fiber optic data is essentially a heatmap. Based on the inherent visual inductive bias of the depthwise separable convolutional units in the first processing unit, the feature locality and translation invariance in the heatmap can be processed. This allows for the direct extraction of key local patterns and basic features from the fiber optic data, injecting crucial prior information into the deformable attention visual transformation processing unit. This overcomes the fundamental shortcoming of the deformable attention visual transformation processing unit in learning specific prior information on small to medium-sized datasets, enabling it to focus on learning long-range global dependencies between spatiotemporal and frequency domain features. This design allows the deformable attention processing unit to be stably trained and used without massive amounts of data, while simultaneously possessing the efficiency of convolutional data processing and the powerful representational capabilities of visual transformations, achieving an optimal balance between performance and efficiency.

[0059] Understandably, the feature compression processing unit needs to process the features output by the deformable attention visual transformation processing unit in terms of format, specifically by compressing the features on the channels so that the features output by the feature compression processing unit can conform to the input data format of the third depthwise separable convolutional unit.

[0060] The third depthwise separable convolutional unit can also be configured as a depthwise separable convolutional network, and its description is similar to that of the first and second depthwise separable convolutional units in step S100 above, so it will not be repeated here.

[0061] Specifically, the step of inputting the enhanced features into the bottleneck layer processing unit for processing to obtain bottleneck features includes: The enhanced features are input into three sequentially connected bottleneck layer processing units for feature transformation to obtain bottleneck features.

[0062] Specifically, the three bottleneck layer processing units can be a first bottleneck layer processing unit, a second bottleneck layer processing unit, and a third bottleneck layer processing unit; the step of inputting the enhanced features into the three sequentially connected bottleneck layer processing units for feature transformation to obtain bottleneck features specifically includes: The enhanced features are input into the deformable attention visual transformation processing unit and the feature compression processing unit connected in sequence in the first bottleneck layer processing unit to obtain the first input features; the first input features are input into the third depthwise separable convolution unit of the first bottleneck layer processing unit for processing to obtain the first output features; the first input features and the first output features are concatenated to obtain the first intermediate bottleneck features. The first intermediate bottleneck feature is input into the deformable attention visual transformation processing unit and the feature compression processing unit connected in sequence in the second bottleneck layer processing unit to obtain the second input feature; the second input feature is input into the third depthwise separable convolution unit of the second bottleneck layer processing unit for processing to obtain the second output feature; the second input feature and the second output feature are concatenated to obtain the second intermediate bottleneck feature. The second intermediate bottleneck feature is input into the deformable attention visual transformation processing unit and the feature compression processing unit connected in sequence in the third bottleneck layer processing unit to obtain the third input feature; the third input feature is input into the third depthwise separable convolution unit of the third bottleneck layer processing unit for processing to obtain the third output feature; the third input feature and the third output feature are concatenated to obtain the bottleneck feature.

[0063] In this embodiment, the data processing in the first bottleneck layer processing unit can be as follows: The enhanced features are input into the deformable attention visual transformation processing unit to obtain a first transformation feature. For example, the size of the first transformation feature can be (a / 4)*(b / 4)*8d, that is, the size of the first transformation feature can be 160*64*128. The first transformation feature is input into the feature compression processing unit to obtain a first input feature. For example, the size of the first input feature can be (a / 4)*(b / 4)*4d, that is, the size of the first input feature can be 160*64*64. The first input feature is input into the third depthwise separable convolution unit to obtain a first output feature. It is understood that the size of the first output feature can remain unchanged, still being (a / 4)*(b / 4)*4d, that is, 160*64*64. The first output feature and the first input feature are concatenated to obtain a first intermediate bottleneck feature. For example, the size of the first intermediate bottleneck feature can be (a / 4)*(b / 4)*8d, that is, the size of the first intermediate bottleneck feature can be 160*64*128.

[0064] In this embodiment, the data processing by the second bottleneck layer processing unit can be as follows: The first intermediate bottleneck feature is input into the deformable attention visual transformation processing unit to obtain the second transformation feature. For example, the size of the second transformation feature can be (a / 8)*(b / 8)*16d, that is, the size of the second transformation feature can be 80*32*256. The second transformation feature is input into the feature compression processing unit to obtain the second input feature. For example, the size of the second input feature can be (a / 8)*(b / 8)*4d, that is, the size of the second input feature can be 80*32*128. The second input feature is input into the third depthwise separable convolution unit to obtain the second output feature. It is understood that the size of the second output feature can remain unchanged, still being (a / 8)*(b / 8)*4d, that is, 80*32*128. The second output feature and the second input feature are concatenated to obtain the second intermediate bottleneck feature. For example, the size of the second intermediate bottleneck feature can be (a / 4)*(b / 4)*8d, that is, the size of the second intermediate bottleneck feature can be 80*32*256.

[0065] In this embodiment, the data processing by the third bottleneck layer processing unit can be as follows: The second intermediate bottleneck feature is input into the deformable attention visual transformation processing unit to obtain the third transformation feature. For example, the size of the third transformation feature can be (a / 16)*(b / 16)*32d, that is, the size of the third transformation feature can be 40*16*512. The third transformation feature is input into the feature compression processing unit to obtain the third input feature. For example, the size of the third input feature can be (a / 16)*(b / 16)*16d, that is, the size of the third input feature can be 40*16*256. The third input feature is input into the third depthwise separable convolution unit to obtain the third output feature. The size of the third output feature can remain unchanged, still (a / 16)*(b / 16)*16d, that is, 40*16*256. The third output feature is concatenated with the third input feature to obtain the bottleneck feature. For example, the size of the bottleneck feature can be (a / 16)*(b / 16)*32d, that is, the size of the bottleneck feature can be 40*16*512.

[0066] S500. Input the bottleneck features into the backbone network for fiber optic event detection to obtain the fiber optic event detection results corresponding to the fiber optic data.

[0067] Specifically, the backbone network includes a fourth depthwise separable convolutional unit, a pooling unit, and a detection unit; The fourth depthwise separable convolutional unit can also be configured as a depthwise separable convolutional network, and its description is similar to that of the first and second depthwise separable convolutional units in step S100 above, so it will not be repeated here.

[0068] The pooling unit can be an adaptive two-dimensional average pooling unit (Adaptive AvgPool2d), which adaptively downsamples the input features of arbitrary size to achieve dimensionality reduction, expand the receptive field, and enhance feature robustness, outputting output features after unified feature standardization and compression; at the same time, it can significantly reduce the amount of subsequent computation and keep the output features stable to small changes in the position of the input signal, thereby improving the efficiency and generalization ability of event detection when processing fiber optic data.

[0069] The detection unit may include a linear function and a softmax function, which are used to process the output features of the pooling unit sequentially through the linear function and the softmax function to obtain the fiber optic event detection result corresponding to the fiber optic data.

[0070] like Figure 5As shown, the step of inputting the bottleneck features into the backbone network for fiber optic event detection to obtain the fiber optic event detection results corresponding to the fiber optic data may include the following steps: S510. Input the bottleneck features into the fourth depthwise separable convolutional unit to obtain convolutional features; In this embodiment, the bottleneck feature is input into the fourth depthwise separable convolutional unit to further learn the key features that the bottleneck feature can guide the identification of fiber optic events, thereby outputting the convolutional feature. For example, the size of the convolutional feature can be (a / 16)*(b / 16)*64d, that is, the size of the convolutional feature can be 40*16*1024.

[0071] S520. Input the convolutional features into the pooling unit for pooling processing to obtain global features; In this embodiment, preferably, the convolutional features are input into the pooling unit for adaptive two-dimensional average pooling processing to perform standardized feature compression on the convolutional features, thereby obtaining global features. Specifically, the global features can be a one-dimensional vector. For example, in this embodiment, the size of the one-dimensional vector can be 1024*1.

[0072] S530. Input the global features into the detection unit to perform fiber optic event detection and obtain the fiber optic event detection result corresponding to the fiber optic data.

[0073] In this embodiment, the fiber optic event result may include the fiber optic event corresponding to the global feature and the probability of the fiber optic event occurring. Preferably, in this embodiment, possible fiber optic events and corresponding features can be preset. In fiber optic event detection, the global feature is first processed by the Linear function to obtain the fiber optic event corresponding to the global feature. The global feature obtained by the Linear function and the fiber optic event corresponding to the global feature are then calculated by the softmax function to obtain the probability of the global feature occurring corresponding to the fiber optic event, so that the operator can understand the fiber optic situation along the fiber based on the fiber optic event detection result.

[0074] like Figure 6 As shown in the illustration, this application also provides an event detection system for optical cable inspection. Optionally, the system includes: The system comprises a first processing unit 611, a second processing unit 612, an attention processing unit 613, a bottleneck layer processing unit 614, and a backbone network 615, wherein: The first processing unit 611 is used to process the optical fiber data of the optical cable to obtain spatiotemporal characteristics. In this embodiment, the first processing unit 611 can be used to execute Figure 1 For a detailed description of the first processing unit 611, please refer to the description of step S100 shown.

[0075] The first processing unit 611 further includes a first depthwise separable convolution unit and a second depthwise separable convolution unit connected in sequence, used to input the optical fiber data into the first depthwise separable convolution unit for processing to obtain intermediate spatiotemporal features; and input the intermediate spatiotemporal features into the second depthwise separable convolution unit for processing to obtain the spatiotemporal features.

[0076] In this embodiment, the first processing unit 611 can also be used to execute Figure 2 For a detailed description of the first processing unit 611, please refer to the description of steps S110-S120.

[0077] The second processing unit 612 is used to process the optical fiber data to obtain frequency domain features; In this embodiment, the second processing unit 612 can be used to execute Figure 1 For a detailed description of step S200 shown, and the second processing unit 612, please refer to the description of step S200.

[0078] The second processing unit 612 further includes a downsampling unit, a transformation unit, and a spectrum analysis unit; it is used to input the optical fiber data into the downsampling unit for downsampling processing to obtain optical fiber downsampled data; input the optical fiber downsampled data into the transformation unit for fast Fourier transform to obtain frequency domain data; and input the frequency shift data into the spectrum analysis unit for log-Mel spectrum analysis to obtain the frequency domain characteristics.

[0079] In this embodiment, the second processing unit 612 can also be used to execute Figure 4 For a more detailed description of steps S210-S230 shown, please refer to the description of steps S210-S230.

[0080] Attention processing unit 613 is used to perform feature enhancement on the spatiotemporal features and frequency domain features after splicing to obtain enhanced features; In this embodiment, the attention processing unit 613 can be used to perform... Figure 1 For a detailed description of the attention processing unit 613, please refer to the description of step S300 shown.

[0081] The bottleneck layer processing unit 614 is used to process the enhanced features to obtain bottleneck features; wherein, the bottleneck layer processing unit includes at least two bottleneck layer processing sub-units, and the bottleneck layer processing sub-units are capable of performing feature transformation processing on the enhanced features. In this embodiment, the bottleneck layer processing unit 614 can be used to perform... Figure 1 For a detailed description of the bottleneck layer processing unit 614, please refer to the description of step S400 shown.

[0082] The backbone network 615 is used to detect fiber optic events based on the bottleneck characteristics and obtain the fiber optic event detection results corresponding to the fiber optic data.

[0083] In this embodiment, the backbone network 615 can be used to perform... Figure 1 For a detailed description of the backbone network 615, please refer to the description of step S500 shown.

[0084] The backbone network 615 further includes a fourth depthwise separable convolutional unit, a pooling unit, and a detection unit; used to input the bottleneck features into the fourth depthwise separable convolutional unit to obtain convolutional features; input the convolutional features into the pooling unit for pooling processing to obtain global features; input the global features into the detection unit for fiber optic event detection to obtain fiber optic event detection results corresponding to the fiber optic data.

[0085] In this embodiment, the backbone network 615 can also be used to perform... Figure 5 For a detailed description of the backbone network 615, please refer to the description of steps S510-S530 shown.

[0086] The system may further include an optical fiber data acquisition module, which is used to acquire the raw optical fiber data of the optical cable; convert the raw optical fiber data into a grayscale image; and perform color rendering on the grayscale image to obtain a heat map containing several color channels, wherein the heat map serves as the optical fiber data.

[0087] In this embodiment, the fiber optic data acquisition module can be used to perform... Figure 3 For a detailed description of the fiber optic data acquisition module, please refer to the description of steps A1-A3 shown.

[0088] This application also provides an electronic device, the structure of which is as follows: Figure 7As shown, the electronic device includes a memory 711, a processor 712, a communication module 713, and an input / output interface 714, etc. Optionally, the memory 711, the processor 712, the communication module 713, and the input / output interface 714 can be connected and communicate with each other through a bus 715.

[0089] The memory 711 is used to store one or more computer programs and to transfer the code of the computer programs to the processor 712; when the one or more computer programs are executed by the processor 712, an event detection method for optical cable detection in this embodiment of the application is implemented.

[0090] Optionally, the electronic device can be connected to a network via the communication module 713 to communicate with other devices, such as terminals or servers, and to interact with data. The electronic device can be various forms of digital computers, exemplarily such as desktop computers, servers, workbenches, mainframes, or other types of computers. The electronic device can also be various forms of mobile terminals, exemplarily such as smartphones, tablets, wearable devices (such as helmets, glasses, watches, etc.), and other similar mobile terminals.

[0091] Optionally, the electronic device can connect to desired input / output devices, such as a keyboard or display device, via the input / output interface 714. The electronic device itself may have a display device, and other display devices can also be connected externally via the input / output interface 714. Optionally, a storage device, such as a hard disk, can also be connected via the input / output interface 714 to store data from the electronic device, read data from the storage device, or store data from the storage device in the memory 711. It is understood that the input / output interface 714 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to the input / output interface 714 can be a component of the electronic device or an external device connected to the electronic device when needed.

[0092] Optionally, the memory 711 may be a volatile memory and / or a non-volatile memory. The volatile memory may be a random access memory, etc., and the non-volatile memory may be a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, or a flash memory, etc.

[0093] Optionally, the computer program stored in the memory 711 can be divided into one or more modules, which are stored in the memory 711 and executed by the processor 712 to perform the method provided in this embodiment. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device.

[0094] Optionally, the processor 712 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 712 include, but are not limited to, a central processing unit, a graphics processing unit, a digital signal processor, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, and can also be any suitable controller, microcontroller, processor, etc. The processor 712 executes the various methods and processes of this embodiment, exemplarily, such as an event detection method for optical cable detection according to an embodiment of this application.

[0095] Optionally, the bus 715 may include a path for transmitting information. Depending on its function, the bus 715 may be classified as an address bus, a data bus, a control bus, etc.

[0096] In an optional implementation, this application embodiment also provides a computer storage medium storing a computer program thereon, which, when executed by a computer, enables the computer to perform the methods described in the above-described method embodiments. Part or all of the computer program may be loaded and / or installed on the memory 711 of an electronic device. When the computer program is executed by the processor 712, one or more steps of an event detection method for optical cable detection according to an embodiment of this application can be performed.

[0097] Optionally, the computer-readable storage medium may be a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, etc.

[0098] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the technical solution of the present invention, and are not intended to limit the specific implementation of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. An event detection method for optical cable inspection, characterized in that, The method includes: The optical fiber data of the optical cable is input into the first processing unit for processing to obtain spatiotemporal features; wherein, the first processing unit includes a first depthwise separable convolution unit and a second depthwise separable convolution unit connected in sequence; the optical fiber data is a heat map; The fiber optic data is input into the second processing unit for processing to obtain frequency domain characteristics; The spatiotemporal features and the frequency domain features are concatenated and then input into the attention processing unit for feature enhancement to obtain enhanced features. The enhanced features are input into the bottleneck layer processing unit for processing to obtain bottleneck features; wherein, the bottleneck layer processing unit includes at least two bottleneck layer processing sub-units, and the bottleneck layer processing sub-units are capable of performing feature transformation processing on the enhanced features; wherein, the bottleneck layer processing sub-units include a deformable attention visual transformation processing unit, a feature compression processing unit, and a third depthwise separable convolution unit connected in sequence. The bottleneck features are input into the backbone network for fiber optic event detection, and the fiber optic event detection results corresponding to the fiber optic data are obtained.

2. The method according to claim 1, characterized in that, The step of inputting the optical fiber data of the optical cable into the first processing unit for processing to obtain spatiotemporal characteristics includes: The fiber data is input into the first depth-separable convolutional unit for processing to obtain intermediate spatiotemporal features; The intermediate spatiotemporal features are input into the second depthwise separable convolutional unit for processing to obtain the spatiotemporal features.

3. The method according to claim 1, characterized in that, The second processing unit includes a downsampling unit, a transformation unit, and a spectrum analysis unit; The step of inputting the fiber optic data into the second processing unit for processing to obtain frequency domain features includes: The fiber optic data is input into the downsampling unit for downsampling processing to obtain fiber optic downsampled data. The fiber optic downsampled data is input into the transformation unit for fast Fourier transform to obtain frequency domain data; The frequency domain data is input into the spectrum analysis unit for log-Mel spectrum analysis to obtain the frequency domain characteristics.

4. The method according to claim 1, characterized in that, The bottleneck layer processing unit includes at least three bottleneck layer processing sub-units connected in sequence. The step of inputting the enhanced features into the bottleneck layer processing unit for processing to obtain bottleneck features includes: The enhanced features are input into three sequentially connected bottleneck layer processing units for feature transformation to obtain bottleneck features.

5. The method according to any one of claims 1 to 4, characterized in that, The backbone network includes a fourth depthwise separable convolutional unit, a pooling unit, and a detection unit; The step of inputting the bottleneck features into the backbone network for fiber optic event detection to obtain the fiber optic event detection results corresponding to the fiber optic data includes: The bottleneck features are input into the fourth depthwise separable convolutional unit to obtain convolutional features; The convolutional features are input into the pooling unit for pooling processing to obtain global features; The global features are input into the detection unit to perform fiber optic event detection, and the fiber optic event detection result corresponding to the fiber optic data is obtained.

6. The method according to any one of claims 1 to 4, characterized in that, Before the step of inputting the optical fiber data of the optical cable into the first processing unit for processing to obtain spatiotemporal characteristics, the method further includes: Obtain the raw fiber data of the optical cable; Convert the raw fiber optic data into a grayscale image; The grayscale image is color-rendered to obtain a heatmap containing several color channels, and the heatmap serves as the fiber optic data.

7. An event detection system for optical cable inspection, characterized in that, The system includes: The first processing unit is used to process the optical fiber data of the optical cable to obtain spatiotemporal features; wherein, the first processing unit includes a first depthwise separable convolution unit and a second depthwise separable convolution unit connected in sequence; the optical fiber data is a heat map. The second processing unit is used to process the optical fiber data to obtain frequency domain features; An attention processing unit is used to perform feature enhancement on the spatiotemporal features and the frequency domain features after splicing, to obtain enhanced features; A bottleneck layer processing unit is used to process the enhanced features to obtain bottleneck features; wherein, the bottleneck layer processing unit includes at least two bottleneck layer processing sub-units, and the bottleneck layer processing sub-units are capable of performing feature transformation processing on the enhanced features; the bottleneck layer processing sub-units include a deformable attention visual transformation processing unit, a feature compression processing unit, and a third depthwise separable convolution unit connected in sequence. The backbone network is used to detect fiber optic events based on the bottleneck characteristics, and to obtain the fiber optic event detection results corresponding to the fiber optic data.

8. The system according to claim 1, characterized in that, The first processing unit is further configured to: The fiber data is input into the first depth-separable convolutional unit for processing to obtain intermediate spatiotemporal features; The intermediate spatiotemporal features are input into the second depthwise separable convolutional unit for processing to obtain the spatiotemporal features.

9. An electronic device, characterized in that, include: Memory, used to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements the functionality of claim 1. An event detection method for optical cable inspection as described in any one of the six claims.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute as described in claim 1. An event detection method for optical cable inspection as described in any one of the six claims.

Images