DAS data processing method and device, electronic equipment and storage medium
Patent Information
- Application Number
- CN202610805146.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-05
- Publication Date
- 2026-10-16
AI Technical Summary
然而，露天矿区的环境条件复杂，光纤在布设过程中经常出现弯曲、挤压以及局部应力集中等现象
[0007] Furthermore, by sequentially connecting the initialization convolution, orientation awareness, time-frequency dual branch, channel topology attention, multi-scale residual fusion, and boundary awareness output modules, the bad sector characteristics can be comprehensively characterized from multiple perspectives such as space, orientation, time-frequency, and channel correlation, which is conducive to the accurate identification of bad sectors.
Smart Images
Figure CN122925925A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a DAS data processing method, apparatus, electronic device, and storage medium. Background Technology
[0002] Distributed Acoustic Sensing (DAS) technology utilizes optical fiber as the sensing medium to achieve continuous, distributed monitoring of vibration or acoustic signals along the fiber's length. It offers advantages such as long-distance operation, high sensitivity, and real-time response, and has therefore been widely used in safety monitoring and resource exploration in open-pit mines. However, the complex environmental conditions in open-pit mines often lead to bending, compression, and localized stress concentration in the optical fiber during deployment. These factors result in significant interference in the acquired DAS data, specifically manifested as abnormal signal energy in some fiber channels, severe waveform distortion, and a significant decrease in cross-correlation with adjacent channels. These channels are commonly referred to as "bad channels."
[0003] Interference caused by bad sectors leads to low availability of DAS data, so how to accurately identify bad sectors is an urgent problem to be solved. Summary of the Invention
[0004] The purpose of this application is to provide at least one DAS data processing method, apparatus, electronic device and storage medium, providing a method for accurately identifying bad sectors, which can at least effectively remove interference from bad sectors in DAS data and improve the availability of DAS data.
[0005] To address the aforementioned technical problems, at least one embodiment of this application provides a DAS data processing method, comprising: acquiring raw DAS data, the raw DAS data including DAS data from multiple channels; inputting the raw DAS data into a bad sector identification model, processing the raw DAS data using the bad sector identification model to generate a bad sector probability map and a boundary gradient map; determining bad sectors in the multiple channels based on the bad sector probability map and the boundary gradient map; and removing the DAS data corresponding to the bad sector from the raw DAS data to obtain target DAS data; wherein, the bad sector identification model includes, in sequence: an initialization convolution module, used to extract local features from the raw DAS data to obtain... A first feature map; a direction-aware convolution module, used to process the first feature map to capture the unidirectional propagation characteristics of bad channels along the fiber direction, to obtain a second feature map; a time-frequency dual-branch extraction module, used to process the second feature map to extract the time-domain baseline drift characteristics and frequency-domain harmonic and spike characteristics of bad channels, to obtain a third feature map; a channel topology attention module, used to process the third feature map to construct the correlation of DAS data under different channels, to obtain a fourth feature map; a multi-scale residual fusion module, used to process the fourth feature map to fuse the features of receptive fields at different scales, to obtain a fifth feature map; and a boundary-aware output module, used to generate a bad channel probability map and a boundary gradient map based on the fifth feature map.
[0006] This paper presents a processing method for removing bad sectors from raw DAS data, covering the entire process of bad sector identification and removal. This method effectively removes interference from bad sectors in DAS data and improves the usability of DAS data.
[0007] Furthermore, by sequentially connecting the initialization convolution, orientation awareness, time-frequency dual branch, channel topology attention, multi-scale residual fusion, and boundary awareness output modules, the bad sector characteristics can be comprehensively characterized from multiple perspectives such as space, orientation, time-frequency, and channel correlation, which is conducive to the accurate identification of bad sectors.
[0008] Furthermore, a bad sector probability map and a boundary gradient map are generated simultaneously, enabling bad sector identification to not only provide probabilities but also accurately locate boundaries, thereby improving the accuracy of bad sector removal.
[0009] In some optional embodiments, the orientation-aware convolution module includes: a main orientation convolution submodule, used to process the first feature map using a first convolution kernel along the fiber direction to obtain a first orientation feature map; a reverse dilation convolution submodule, used to process the first feature map using a second convolution kernel along the opposite direction of the fiber to obtain a second orientation feature map; and an orientation attention weighting submodule, used to perform a weighted summation of the first orientation feature map and the second orientation feature map to obtain the second feature map.
[0010] The principal direction convolution is processed along the fiber direction to capture the core region of bad channels, while the reverse dilation convolution is processed along the opposite direction of the fiber to capture stress attenuation characteristics. The combination of the two can describe the asymmetric distribution of bad channels along the fiber direction. By using weighted summation to adapt to the laying scenario, the model's adaptability is improved, which in turn helps to improve the accuracy of identifying asymmetric bad channels caused by extrusion and bending.
[0011] In some optional embodiments, the time-frequency dual-branch extraction module includes: a time-domain branching submodule, used to perform convolution processing on the second feature map through multiple cascaded convolutional layers to obtain a time-domain feature map; a frequency-domain branching submodule, used to perform Fourier transform on the second feature map along the time dimension to obtain an amplitude spectrum; and then convolve the amplitude spectrum to obtain a frequency-domain feature map; and an adaptive fusion submodule, used to perform weighted fusion of the time-domain feature map and the frequency-domain feature map to obtain the third feature map.
[0012] Multiple cascaded convolutional layers can extract time-domain anomalies such as baseline drift, while frequency-domain anomalies such as harmonics and spikes can be extracted by obtaining the amplitude spectrum through Fourier transform and then convolving it. The adaptive fusion submodule can balance the weights of time-domain and frequency-domain features to be applicable to different types of bad channels, which is beneficial for decoupling and fusing time-frequency information and improving the ability to distinguish complex bad channels.
[0013] In some optional embodiments, the channel topology attention module includes: a position encoding submodule, used to superimpose a position embedding vector onto the third feature map to obtain a position feature map; and a relative position attention submodule, used to calculate a relative position attention weight matrix between different channels based on the position feature map, wherein the attention weight matrix applies a preset bias term according to the spatial distance between channels, and the position feature map is weighted and recombined using the attention weight matrix to obtain the fourth feature map.
[0014] Attention weights with bias applied based on the distance between channels are strengthened to enhance the correlation between adjacent channels. This is beneficial for subsequent identification of continuous bad channels (such as a section of optical fiber being buried) based on the fourth feature map. At the same time, it helps to suppress isolated noise and improve spatial consistency.
[0015] In some optional embodiments, the multi-scale residual fusion module includes: a multi-scale convolution submodule, used to process the fourth feature map in parallel using multiple dilated convolutions with different dilation rates; a fusion submodule, used to concatenate the feature maps output by the multi-scale convolution submodule in the channel dimension, and then reduce the number of channels of the concatenated result to be the same as the number of channels of the fourth feature map; and a residual connection submodule, used to add the output of the fusion submodule to the fourth feature map element by element to obtain the fifth feature map.
[0016] Using parallel convolutions with different dilation rates can simultaneously capture bad sector morphologies of different sizes, which helps to enhance the model's accuracy in identifying bad sectors at different scales and improve the accuracy of the output bad sector probability map.
[0017] In some optional embodiments, the boundary-aware output module includes: a probability branch submodule, used to reduce the number of channels in the fifth feature map to 1 and use a first activation function to output the probability that each channel is a bad channel, thereby obtaining the bad channel probability map; and a gradient branch submodule, used to reduce the number of channels in the fifth feature map to 2 and use a second activation function to output the gradient estimates of each channel along the forward and reverse directions of the optical fiber, thereby obtaining the boundary gradient map.
[0018] At least one embodiment of this application also provides a DAS data processing apparatus, comprising: an acquisition module for acquiring raw DAS data, the raw DAS data including DAS data from multiple channels; a processing module for inputting the raw DAS data into a bad sector identification model, and processing the raw DAS data using the bad sector identification model to generate a bad sector probability map and a boundary gradient map; a determination module for determining bad sectors in the multiple channels based on the bad sector probability map and the boundary gradient map; and a removal module for removing the DAS data corresponding to the bad sector from the raw DAS data to obtain target DAS data; wherein the bad sector identification model includes, in sequence: an initialization convolution module for extracting the raw DAS data. The algorithm generates a first feature map by processing local features in the optical fiber; a direction-aware convolution module processes the first feature map to capture the unidirectional propagation features of bad channels along the fiber direction, resulting in a second feature map; a time-frequency dual-branch extraction module processes the second feature map to extract the time-domain baseline drift features and frequency-domain harmonic and spike features of bad channels, resulting in a third feature map; a channel topology attention module processes the third feature map to construct the correlation of DAS data under different channels, resulting in a fourth feature map; a multi-scale residual fusion module processes the fourth feature map to fuse the features of receptive fields at different scales, resulting in a fifth feature map; and a boundary-aware output module generates a bad channel probability map and a boundary gradient map based on the fifth feature map.
[0019] At least one embodiment of this application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the DAS data processing method described above.
[0020] At least one embodiment of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the DAS data processing method described above.
[0021] At least one embodiment of this application also provides a computer program product, including a computer program that, when executed by a processor, implements the DAS data processing method described above. Attached Figure Description
[0022] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.
[0023] Figure 1 This is a flowchart of a DAS data processing method provided in one embodiment of this application; Figure 2 This is a schematic diagram of a DAS data processing apparatus provided in another embodiment of this application. Detailed Implementation
[0024] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
[0025] Furthermore, the accompanying drawings are merely illustrative of this application and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0026] To facilitate understanding of the embodiments of this application, relevant content regarding DAS data processing will be introduced first.
[0027] In safety monitoring and resource exploration in open-pit mines, distributed fiber optic sensing systems are widely used due to their long-distance monitoring, high sensitivity, and real-time response. Existing distributed fiber optic technology primarily involves deploying fiber optic lines within the mining area to sense physical signals such as vibration and sound in the surrounding environment, thereby monitoring the underground structure of the mine. However, the environment in open-pit mines is complex, and fiber optic cables often experience bending and compression during deployment, frequently resulting in significant interference in the acquired signals. Furthermore, environmental and equipment limitations can introduce other irrelevant noise. These problems severely impact data accuracy and subsequent imaging results.
[0028] Currently, although some technologies have attempted to preprocess the acquired DAS data, such as using fully connected neural networks (FCNN) or convolutional neural networks (CNN) for bad channel identification, these methods often struggle to effectively remove specific types of interference (bad channels) caused by excessive bending and compression of optical fibers. This results in low usability of the DAS data, making it difficult to meet the needs of high-precision monitoring of underground structures in mining areas. Therefore, a more effective signal processing method is urgently needed to improve the effective removal of bad channel interference from DAS data.
[0029] To address the aforementioned technical problems, this application proposes a DAS data processing method. The implementation details of the DAS data processing method in this embodiment are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution.
[0030] Example 1: The DAS data processing method of this embodiment can be applied to electronic devices with communication, computing, and data storage capabilities. Its specific process can be as follows: Figure 1 As shown, it includes: S101-S104.
[0031] S101, Obtain raw DAS data, which includes DAS data from multiple channels.
[0032] The embodiments of this application do not limit how DAS data is acquired. For example, when an electronic device is directly connected to an optical fiber, it receives DAS data transmitted through that optical fiber to obtain the raw DAS data. Alternatively, other devices may collect the raw DAS data and then transmit it to the electronic device executing this method.
[0033] In this process, each sampling point on the optical fiber corresponds to one channel, and multiple sampling points correspond to multiple channels.
[0034] This application does not impose any restrictions on the environment in which the optical fiber corresponding to the original DAS data is laid. For example, the optical fiber may be laid in an open-pit mine.
[0035] S102, input the original DAS data into the bad sector identification model, and use the bad sector identification model to process the original DAS data to generate a bad sector probability map and a boundary gradient map.
[0036] In some examples, the bad sector identification model includes an initialization convolution module, a direction-aware convolution module, a time-frequency dual-branch extraction module, a channel topology attention module, a multi-scale residual fusion module, and a boundary-aware output module connected in sequence.
[0037] In some examples, the initialization convolution module is used to extract local features from the raw DAS data to obtain the first feature map.
[0038] The embodiments of this application do not limit the size of the convolution kernel used to initialize the convolution module. For example, the size of the convolution kernel used to initialize the convolution module may be 3×3 or 5×3.
[0039] When the initialization convolution module performs convolution processing on the original DAS data using convolution kernels, the specific stride of the convolution kernel is not limited in the embodiments of this application. For example, the stride is 1.
[0040] In some examples, the orientation-aware convolutional module is used to process the first feature map to capture the unidirectional propagation characteristics of bad channels along the fiber direction, resulting in a second feature map.
[0041] In some examples, the orientation-aware convolution module includes: a main orientation convolution submodule, which processes the first feature map using a first convolution kernel along the fiber direction to obtain a first orientation feature map; a reverse dilation convolution submodule, which processes the first feature map using a second convolution kernel along the opposite direction of the fiber to obtain a second orientation feature map; and an orientation attention weighting submodule, which performs a weighted summation of the first orientation feature map and the second orientation feature map to obtain a second feature map.
[0042] The embodiments of this application do not limit the size of the first convolution kernel. For example, the size of the first convolution kernel is 1×7, where "1" corresponds to the time dimension and "7" corresponds to the channel dimension (i.e., the spatial dimension).
[0043] The first convolution kernel along the fiber direction refers to the first convolution kernel that moves along the fiber direction during convolution processing.
[0044] "Along the fiber direction" refers to the direction opposite to "along the fiber direction," and correspondingly, the first and second convolution kernels move in opposite directions.
[0045] The embodiments of this application do not limit the size of the second convolution kernel. For example, the size of the second convolution kernel is 1×9, where "1" corresponds to the time dimension and "9" corresponds to the channel dimension. In some examples, the dilation rate of the second convolution kernel is greater than 1. The embodiments of this application do not limit the specific value of the dilation rate of the second convolution kernel. For example, the dilation rate of the second convolution kernel is 2.
[0046] The weights used by the directional attention weighting submodule when performing a weighted summation of the first directional feature map and the second directional feature map are weights determined after training. These weights are adjusted during the training phase to adapt to the corresponding deployment scenario (e.g., open-pit coal mine).
[0047] The principal direction convolution is processed along the fiber direction to capture the core region of bad channels, while the reverse dilation convolution is processed along the opposite direction of the fiber to capture stress attenuation characteristics. The combination of the two can describe the asymmetric distribution of bad channels along the fiber direction. By using weighted summation to adapt to the laying scenario, the model's adaptability is improved, which in turn helps to improve the accuracy of identifying asymmetric bad channels caused by extrusion and bending.
[0048] In some examples, the time-frequency dual-branch extraction module is used to process the second feature map to extract the time-domain baseline drift features and frequency-domain harmonic and spike features of the bad channel, thus obtaining the third feature map.
[0049] In some examples, the time-frequency dual-branch extraction module includes: a time-domain branch submodule, which performs convolution processing on the second feature map through multiple cascaded convolutional layers to obtain a time-domain feature map; a frequency-domain branch submodule, which performs Fourier transform on the second feature map along the time dimension to obtain an amplitude spectrum; and then convolves the amplitude spectrum to obtain a frequency-domain feature map; and an adaptive fusion submodule, which performs weighted fusion of the time-domain feature map and the frequency-domain feature map to obtain a third feature map.
[0050] The embodiments of this application do not limit the specific number of convolutional layers in a series. For example, the series of convolutional layers may consist of two convolutional layers. The embodiments of this application also do not limit the size of the convolutional kernel in each convolutional layer. For example, the kernel size of each layer in the series of convolutional layers may be 3×3.
[0051] The embodiments of this application do not limit the size of the convolution kernel used when convolving the amplitude spectrum. For example, the size is 3×3.
[0052] Multiple cascaded convolutional layers can extract time-domain anomalies such as baseline drift, while frequency-domain anomalies such as harmonics and spikes can be extracted by obtaining the amplitude spectrum through Fourier transform and then convolving it. The adaptive fusion submodule can balance the weights of time-domain and frequency-domain features to be applicable to different types of bad channels, which is beneficial for decoupling and fusing time-frequency information and improving the ability to distinguish complex bad channels.
[0053] In some examples, the channel topology attention module is used to process the third feature map to construct the correlation of DAS data under different channels, resulting in a fourth feature map.
[0054] In some examples, the channel topology attention module includes: a position encoding submodule, used to overlay position embedding vectors onto the third feature map to obtain a position feature map; and a relative position attention submodule, used to calculate the relative position attention weight matrix between different channels based on the position feature map. The attention weight matrix applies a preset bias term according to the spatial distance between channels, and uses the attention weight matrix to perform weighted recombination of the position feature map to obtain a fourth feature map.
[0055] During training, the position encoding submodule learns the position embedding vectors corresponding to different spatial locations (channels).
[0056] Attention weights with bias applied based on the distance between channels are strengthened to enhance the correlation between adjacent channels. This is beneficial for subsequent identification of continuous bad channels (such as a section of optical fiber being buried) based on the fourth feature map. At the same time, it helps to suppress isolated noise and improve spatial consistency.
[0057] In some examples, the multi-scale residual fusion module is used to process the fourth feature map to fuse features from receptive fields at different scales to obtain the fifth feature map.
[0058] In some examples, the multi-scale residual fusion module includes: a multi-scale convolution submodule, used to process the fourth feature map in parallel using multiple dilated convolutions with different dilation rates; a fusion submodule, used to concatenate the feature maps output by the multi-scale convolution submodule along the channel dimension, and then reduce the number of channels of the concatenated result to be the same as the number of channels of the fourth feature map; and a residual connection submodule, used to add the output of the fusion submodule to the fourth feature map element by element to obtain the fifth feature map.
[0059] The embodiments of this application do not limit the specific types of expansion rates. For example, the different expansion rates include three expansion rates: 1, 2, and 4.
[0060] The embodiments of this application do not impose limitations on how to reduce the number of channels in the stitched result to the same number of channels as the fourth feature map. For example, a 1×1 convolution can be used to reduce the number of channels in the stitched result to the same number of channels as the fourth feature map.
[0061] Using parallel convolutions with different dilation rates can simultaneously capture bad sector morphologies of different sizes, which helps to enhance the model's accuracy in identifying bad sectors at different scales and improve the accuracy of the output bad sector probability map.
[0062] In some examples, the boundary-aware output module is used to generate a bad road probability map and a boundary gradient map based on the fifth feature map.
[0063] In some examples, the boundary-aware output module includes: a probability branch submodule, which reduces the number of channels in the fifth feature map to 1 and uses the first activation function to output the probability that each channel is a bad channel, thus obtaining a bad channel probability map; and a gradient branch submodule, which reduces the number of channels in the fifth feature map to 2 and uses the second activation function to output the gradient estimates of each channel along the forward and reverse directions of the optical fiber, thus obtaining a boundary gradient map.
[0064] The embodiments of this application do not limit the specific activation functions used for the first and second activation functions. For example, the first activation function may be the Sigmoid function, and the second activation function may be the Tanh function.
[0065] The embodiments of this application do not limit how the number of channels in the fifth feature map is reduced to 1 or 2. For example, a 1×1 convolution kernel can be used to reduce the number of channels in the fifth feature map.
[0066] S103, based on the bad sector probability map and boundary gradient map, identifies bad sectors in multiple channels.
[0067] The bad sector probability map includes the bad sector probability corresponding to the data at each time point for each channel. The boundary gradient map includes the two gradients in the forward and reverse directions of the optical fiber corresponding to the data at each time point for each channel.
[0068] The embodiments of this application do not limit how to determine bad sectors in multiple channels based on bad sector probability maps and boundary gradient maps.
[0069] In some examples, determining bad channels among multiple channels based on bad channel probability maps and boundary gradient maps may include: selecting candidate bad channels from multiple channels based on the bad channel probability map; for consecutive candidate bad channels, determining the left and right boundaries from the consecutive candidate bad channels using the boundary gradient map, and then selecting the final bad channel from the consecutive candidate bad channels; and for isolated candidate bad channels, if the channel-level bad channel probability of the candidate bad channel is greater than a first probability threshold, it is determined to be a bad channel; otherwise, the positive boundary strength and negative boundary strength of the candidate bad channel are checked, and if both are lower than a preset gradient threshold, it is determined to be a noise channel rather than a bad channel, and it is removed from the candidate bad channels.
[0070] In some examples, selecting candidate bad channels from multiple channels based on the bad channel probability map may include: for each channel, aggregating its bad channel probability values at all time points (e.g., taking the average or maximum value) to obtain the channel-level bad channel probability of that channel; setting a second probability threshold, and marking channels with channel-level bad channel probabilities greater than or equal to the second probability threshold as candidate bad channels, which may form continuous channel intervals in space.
[0071] The absolute values of the positive and negative gradients of each channel at all time points are aggregated (e.g., averaged) to obtain the positive and negative boundary strengths of that channel.
[0072] S104, Remove the DAS data corresponding to bad sectors from the original DAS data to obtain the target DAS data.
[0073] This approach provides a processing method from raw DAS data to bad sector removal, covering the entire process of bad sector identification and removal, effectively removing interference from bad sectors in DAS data and improving the usability of DAS data.
[0074] Furthermore, by sequentially connecting the initialization convolution, orientation awareness, time-frequency dual branch, channel topology attention, multi-scale residual fusion, and boundary awareness output modules, the bad sector characteristics can be comprehensively characterized from multiple perspectives such as space, orientation, time-frequency, and channel correlation, which is conducive to the accurate identification of bad sectors.
[0075] Furthermore, a bad sector probability map and a boundary gradient map are generated simultaneously, enabling bad sector identification to not only provide probabilities but also accurately locate boundaries, thereby improving the accuracy of bad sector removal.
[0076] In some cases, after obtaining the target DAS data, adaptive filtering can be applied to the target DAS data to remove other noise.
[0077] In some cases, the denoised DAS data can also be used for correlation analysis and imaging.
[0078] Example 2: Another embodiment of this application relates to a DAS data processing device. The implementation details of the DAS data processing device of this embodiment are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution. The schematic diagram of the DAS data processing device of this embodiment can be seen as follows: Figure 2As shown, it includes: an acquisition module 21 for acquiring raw DAS data, which includes DAS data from multiple channels; a processing module 22 for inputting the raw DAS data into a bad sector identification model, processing the raw DAS data using the bad sector identification model, and generating a bad sector probability map and a boundary gradient map; a determination module 23 for determining bad sectors in multiple channels based on the bad sector probability map and the boundary gradient map; and a removal module 24 for removing the DAS data corresponding to the bad sectors from the raw DAS data to obtain the target DAS data; wherein, the bad sector identification model includes, in sequence: an initialization convolution module for extracting local features from the raw DAS data to obtain a first feature. The diagram shows: a direction-aware convolution module, used to process the first feature map to capture the unidirectional propagation characteristics of bad channels along the fiber direction, resulting in a second feature map; a time-frequency dual-branch extraction module, used to process the second feature map to extract the time-domain baseline drift characteristics and frequency-domain harmonic and spike characteristics of bad channels, resulting in a third feature map; a channel topology attention module, used to process the third feature map to construct the correlation of DAS data under different channels, resulting in a fourth feature map; a multi-scale residual fusion module, used to process the fourth feature map to fuse the features of receptive fields at different scales, resulting in a fifth feature map; and a boundary-aware output module, used to generate a bad channel probability map and a boundary gradient map based on the fifth feature map.
[0079] This paper presents a processing method for removing bad sectors from raw DAS data, covering the entire process of bad sector identification and removal. This method effectively removes interference from bad sectors in DAS data and improves the usability of DAS data.
[0080] Furthermore, by sequentially connecting the initialization convolution, orientation awareness, time-frequency dual branch, channel topology attention, multi-scale residual fusion, and boundary awareness output modules, the bad sector characteristics can be comprehensively characterized from multiple perspectives such as space, orientation, time-frequency, and channel correlation, which is conducive to the accurate identification of bad sectors.
[0081] Furthermore, a bad sector probability map and a boundary gradient map are generated simultaneously, enabling bad sector identification to not only provide probabilities but also accurately locate boundaries, thereby improving the accuracy of bad sector removal.
[0082] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed in this application; however, this does not mean that other units are absent in this embodiment.
[0083] Example 3: Another embodiment of this application relates to an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the DAS data processing methods in the above embodiments.
[0084] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.
[0085] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.
[0086] Example 4: Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.
[0087] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0088] Example 5: Another embodiment of this application relates to a computer program product, including a computer program. When executed by a processor, the computer program implements the method embodiments described above.
[0089] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.
[0090] Furthermore, although the steps of the method in this application are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.
[0091] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware.
[0092] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.
Claims
1. A DAS data processing method, characterized in that, include: Acquire raw distributed acoustic wave sensor (DAS) data, which includes DAS data from multiple channels; The original DAS data is input into the bad sector identification model, and the bad sector identification model is used to process the original DAS data to generate a bad sector probability map and a boundary gradient map. Based on the bad sector probability map and the boundary gradient map, bad sectors in the plurality of channels are determined; Remove the DAS data corresponding to the bad sector from the original DAS data to obtain the target DAS data; The bad sector identification model includes the following components connected in sequence: The convolution module is initialized to extract local features from the original DAS data to obtain the first feature map; A direction-aware convolution module is used to process the first feature map to capture the unidirectional propagation characteristics of bad channels along the fiber direction to obtain a second feature map. The time-frequency dual-branch extraction module is used to process the second feature map to extract the time-domain baseline drift features and frequency-domain harmonic and spike features of the bad channel, and obtain the third feature map. The channel topology attention module is used to process the third feature map to construct the correlation of DAS data under different channels and obtain the fourth feature map. A multi-scale residual fusion module is used to process the fourth feature map to fuse features of receptive fields at different scales to obtain a fifth feature map. The boundary-aware output module is used to generate a bad road probability map and a boundary gradient map based on the fifth feature map.
2. The method according to claim 1, characterized in that, The orientation-aware convolutional module includes: The main direction convolution submodule is used to process the first feature map using a first convolution kernel along the fiber direction to obtain a first direction feature map; The reverse dilation convolution submodule is used to process the first feature map with a second convolution kernel in the opposite direction of the optical fiber to obtain a second-direction feature map. The directional attention weighting submodule is used to perform a weighted summation of the first directional feature map and the second directional feature map to obtain the second feature map.
3. The method according to claim 1, characterized in that, The time-frequency dual-branch extraction module includes: The temporal branch submodule is used to perform convolution processing on the second feature map through multiple cascaded convolutional layers to obtain a temporal feature map; The frequency domain branching submodule is used to perform a Fourier transform on the second feature map along the time dimension to obtain the amplitude spectrum; then, the amplitude spectrum is convolved to obtain the frequency domain feature map. An adaptive fusion submodule is used to perform weighted fusion of the time-domain feature map and the frequency-domain feature map to obtain the third feature map.
4. The method according to claim 1, characterized in that, The channel topology attention module includes: The position encoding submodule is used to superimpose a position embedding vector onto the third feature map to obtain a position feature map; The relative position attention submodule is used to calculate the relative position attention weight matrix between different channels based on the position feature map. The attention weight matrix applies a preset bias term according to the spatial distance between the channels, and the position feature map is weighted and recombined using the attention weight matrix to obtain the fourth feature map.
5. The method according to claim 1, characterized in that, The multi-scale residual fusion module includes: A multi-scale convolution submodule is used to process the fourth feature map in parallel using multiple dilated convolutions with different dilation rates; The fusion submodule is used to concatenate the feature maps output by the multi-scale convolution submodule in the channel dimension, and then reduce the number of channels of the concatenated result to be the same as the number of channels of the fourth feature map. The residual connection submodule is used to add the output of the fusion submodule to the fourth feature map element by element to obtain the fifth feature map.
6. The method according to claim 1, characterized in that, The boundary sensing output module includes: The probability branch submodule is used to reduce the number of channels in the fifth feature map to 1 and use the first activation function to output the probability that each channel is a bad channel, so as to obtain the bad channel probability map. The gradient branch submodule is used to reduce the number of channels in the fifth feature map to 2 and use the second activation function to output the gradient estimates of each channel along the forward and reverse directions of the optical fiber to obtain the boundary gradient map.
7. A DAS data processing device, characterized in that, include: The acquisition module is used to acquire raw distributed acoustic wave sensor (DAS) data, which includes DAS data from multiple channels. The processing module is used to input the original DAS data into the bad sector identification model, and use the bad sector identification model to process the original DAS data to generate a bad sector probability map and a boundary gradient map. The determination module is used to determine bad channels in the plurality of channels based on the bad channel probability map and the boundary gradient map; The removal module is used to remove the DAS data corresponding to the bad sector from the original DAS data to obtain the target DAS data; The bad sector identification model includes the following components connected in sequence: The convolution module is initialized to extract local features from the original DAS data to obtain the first feature map; A direction-aware convolution module is used to process the first feature map to capture the unidirectional propagation characteristics of bad channels along the fiber direction to obtain a second feature map. The time-frequency dual-branch extraction module is used to process the second feature map to extract the time-domain baseline drift features and frequency-domain harmonic and spike features of the bad channel, and obtain the third feature map. The channel topology attention module is used to process the third feature map to construct the correlation of DAS data under different channels and obtain the fourth feature map. A multi-scale residual fusion module is used to process the fourth feature map to fuse features of receptive fields at different scales to obtain a fifth feature map. The boundary-aware output module is used to generate a bad road probability map and a boundary gradient map based on the fifth feature map.
8. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the DAS data processing method as described in any one of claims 1 to 6.
9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the DAS data processing method according to any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the DAS data processing method according to any one of claims 1 to 6.