A distributed Brillouin optical fiber sensor temperature extraction method and device based on DeepLabv3+
By using a network model based on DeepLabv3+, the problems of time consumption and insufficient accuracy in data processing of the BOTDA system were solved, achieving fast and accurate temperature extraction, expanding the measurement range and enhancing the system's adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA SHIP DEV & DESIGN CENT
- Filing Date
- 2026-03-18
- Publication Date
- 2026-07-10
AI Technical Summary
Existing Brillouin optical time domain analyzer (BOTDA) systems have long data processing times when dealing with massive amounts of sensor data, and the curve fitting method has high requirements for signal-to-noise ratio, resulting in poor temperature extraction accuracy.
A network model based on DeepLabv3+ is adopted. By constructing a training dataset, image preprocessing and supervised training, the mapping relationship between Brillouin gain spectrum image and temperature distribution is established. Multi-scale features are extracted using the void space pyramid pooling module to achieve fast and accurate temperature extraction.
It effectively reduces the time cost of temperature extraction, improves the tolerance to noise, extends the measurement distance, enhances the generalization ability of the model, avoids the time-consuming curve fitting process, and improves the accuracy of temperature extraction.
Smart Images
Figure CN122368720A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of distributed optical fiber sensing technology, specifically to a method and apparatus for temperature extraction using a distributed Brillouin optical fiber sensor based on DeepLabv3+. Background Technology
[0002] Distributed fiber optic sensing systems generate massive amounts of data along the fiber optic cable during operation, with sensing parameters hidden within this data. How to quickly and accurately extract these sensing parameters from the data has become a pressing problem. The Brillouin Optical Time-Domain Analyzer (BOTDA), based on the stimulated Brillouin scattering effect in optical fibers, obtains temperature and stress information at any point on the fiber by detecting the Brillouin frequency shift (BFS) of the light at that point. Due to its advantages of high measurement accuracy, long measurement distance, and high spatial resolution, it is widely used in pipeline and optical cable laying and facility health monitoring.
[0003] The Brillouin optical time-domain analyzer, as a type of distributed Brillouin fiber optic sensing system, features long sensing distance, high measurement accuracy, and excellent spatial resolution. This technology inputs pump and probe beams with a certain optical frequency difference to both ends of the fiber. The two beams interact within the fiber under test via Brillouin interaction, achieving optical power transfer. The Brillouin interaction equation is shown in equation (1): (1) in, This represents the damping coefficient of the optical fiber. The gradient is related to the time-averaged electric field; the light frequency cannot cause its change. It is caused by the slow response time of molecular displacement in the material, and therefore can only respond to the envelope formed by the interference of the pump and probe light. By solving this equation, the Brillouin Gain Spectrum (BGS) received by the probe light during transmission can be obtained: (2) in, Γ1 represents the Brillouin gain coefficient, ΔΩ represents the detuning between BFS and the pump-probe frequency difference, and Γ1 represents the attenuation rate of the acoustic field in the fiber. The generation of Brillouin scattering is highly related to the thermodynamic quantities of the environment in which the fiber is located, such as temperature and density. This is reflected in the BGS as the location of the Brillouin gain peak, that is, BFS is strongly correlated with variables such as ambient temperature and strain on the fiber. By scanning the pump-probe frequency difference within a certain range, the gain of the probe light at different modulation frequencies can be obtained, and the peak frequency of this gain spectrum is BFS. BFS is linearly related to the ambient temperature and strain at the measurement point, as shown in equation (3): (3) in, , , representing the temperature and strain sensitivity of BFS in standard single-mode fiber, respectively. Although similar sensitivities can be found for various fibers, due to the thermal strain in the fiber coating, the two parameters can be completely different from the above values in some cases. Equation (3) is also the core principle of the BOTDA system. However, the massive data processing process that results in this is time-consuming, limiting the improvement of the dynamic response capability of the BOTDA system. At the same time, the curve fitting method for extracting temperature has a high requirement for the BGS signal-to-noise ratio. When the signal-to-noise ratio is low, irrelevant redundant information will have a significant impact on the fitting accuracy during the curve fitting process, resulting in a deterioration in the accuracy of Brillouin temperature extraction based on curve fitting. Summary of the Invention
[0004] This invention provides a method and apparatus for temperature extraction from a distributed Brillouin fiber optic sensor based on DeepLabv3+, which can be applied to the data processing in BOTDA to accurately and quickly extract temperature information from BOTDA measurement data.
[0005] In a first aspect, the present invention provides a method for temperature extraction using a distributed Brillouin fiber optic sensor based on DeepLabv3+, comprising: Brillouin gain spectrum images under different temperature, linewidth and noise conditions were generated by simulation, and each image was labeled with a corresponding temperature label to form a training dataset. The Brillouin gain spectrum image was normalized, and the temperature labels were standardized. Construct a DeepLabv3+ network model, wherein the DeepLabv3+ network model adopts an encoder-decoder structure, the encoder includes a holed spatial pyramid pooling module for extracting multi-scale features, and the decoder is used for feature fusion and upsampling; The DeepLabv3+ network model is trained under supervision using the training dataset to optimize network parameters, establish a mapping relationship between the Brillouin gain spectrum image and the temperature distribution, and then obtain the temperature distribution along the optical fiber from the measured Brillouin gain spectrum image through the trained DeepLabv3+ network model.
[0006] In some instances, the construction of the training dataset includes the following parameter settings: temperature range: -2°C to 102°C, with a step size of 1°C; Brillouin linewidth range: 25 MHz to 80 MHz, with a step size of 1 MHz; noise conditions: including noisy BGS images with no noise and a signal-to-noise ratio of 10 dB.
[0007] In some instances, the Brillouin gain spectrum image is normalized by scaling the pixel values to the [0,1] interval; the temperature labels are normalized by mapping the temperature values to the [0,1] interval.
[0008] In some instances, the loss function used during training is the mean squared error loss function, the optimizer is the Adam optimizer, and the initial learning rate is set to 1×10⁻⁶. -4 .
[0009] In some instances, the measured Brillouin gain spectrum image comes from a Brillouin optical time-domain analysis system, and the image format is a distance-frequency two-dimensional matrix.
[0010] Secondly, the present invention provides a temperature extraction device based on a distributed Brillouin fiber optic sensor using DeepLabv3+, comprising: The dataset construction module is used to generate Brillouin gain spectrum images under different temperature, linewidth and noise conditions through simulation, and to label each image with the corresponding temperature label to form a training dataset. The preprocessing module is used to normalize the Brillouin gain spectrum image and standardize the temperature labels. The model building module is used to build the DeepLabv3+ network model. The DeepLabv3+ network model adopts an encoder-decoder structure. The encoder includes a holed spatial pyramid pooling module for extracting multi-scale features, and the decoder is used for feature fusion and upsampling. The training module is used to supervise the training of the DeepLabv3+ network model using the training dataset, optimize the network parameters, establish the mapping relationship between the Brillouin gain spectrum image and the temperature distribution, and then obtain the temperature distribution along the optical fiber from the measured Brillouin gain spectrum image through the trained DeepLabv3+ network model.
[0011] In some instances, the construction of the training dataset includes the following parameter settings: temperature range: -2°C to 102°C, with a step size of 1°C; Brillouin linewidth range: 25 MHz to 80 MHz, with a step size of 1 MHz; noise conditions: including noisy BGS images with no noise and a signal-to-noise ratio of 10 dB.
[0012] In some instances, the Brillouin gain spectrum image is normalized by scaling the pixel values to the [0,1] interval; the temperature labels are normalized by mapping the temperature values to the [0,1] interval.
[0013] In some instances, the loss function used during training is the mean squared error loss function, the optimizer is the Adam optimizer, and the initial learning rate is set to 1×10⁻⁶. -4 .
[0014] In some instances, the measured Brillouin gain spectrum image comes from a Brillouin optical time-domain analysis system, and the image format is a distance-frequency two-dimensional matrix.
[0015] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: The network model enhances the dilated spatial pyramid pooling module, utilizing dilated convolutions with different dilation rates to acquire convolutional features at multiple scales. Simultaneously, it uses image-level features to output accurate temperature values after passing through the output layer, enabling temperature extraction in the BOTDA sensing system. Compared to traditional methods, the network model's temperature extraction process avoids the time-consuming curve fitting process, effectively reducing the time cost of a single measurement and even enabling dynamic BOTDA measurements. It also exhibits better noise tolerance, effectively extending the measurement distance and sensing range of the BOTDA sensing system. Furthermore, the DeepLabv3+ model demonstrates good generalization ability after training. Through data label normalization during network training, the network can adapt to BOTDA systems with different scanning windows, offering the advantage of ease of use. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 The BGS simulation image acquired by BOTDA is provided in the embodiment of the present invention, wherein (a) is a noise-free BGS trajectory map, and (b) is a BGS trajectory map with a signal-to-noise ratio of 19 dB. Figure 2 This is a flowchart of the DeepLabv3+ network Brillouin temperature extraction process provided in an embodiment of the present invention; Figure 3 This is the DeepLabv3+ network structure provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of the device provided in an embodiment of the present invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] In the following description, specific embodiments of the invention will be illustrated with reference to steps and symbols performed by one or more computers, unless otherwise stated. Therefore, these steps and operations will be referred to several times as being performed by a computer, and computer execution as referred to herein includes operations by a computer processing unit representing electronic signals of data in a structured format. This operation transforms the data or maintains it at a location in the computer's memory system, which can be reconfigured or otherwise alter the operation of the computer in a manner well known to those skilled in the art. The data structure maintained by the data is the physical location of the memory, which has specific characteristics defined by the data format. However, the principles of the invention described above are not intended to be limiting, and those skilled in the art will understand that many of the following steps and operations can also be implemented in hardware.
[0020] The terms "module" or "unit" as used herein can be considered as software objects executing on the computing system. Different components, modules, engines, and services described herein can be considered as implementations on the computing system. The apparatus and methods described herein are preferably implemented in software, but can also be implemented in hardware, both of which are within the scope of this invention.
[0021] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.
[0022] In this embodiment of the invention, a method for temperature extraction based on a distributed Brillouin fiber optic sensor using DeepLabv3+ is provided, which needs to consider the following points: 1. The design of the training dataset for DeepLabv3+ network models is very important. The design of the dataset directly affects the model's learning, generalization and final performance. It needs to be representative, diverse and high-quality. It is also necessary to consider the balance of various training data to prevent the model from being biased towards common categories when making predictions.
[0023] 2. Preprocessing the input data for DeepLabv3+ network models is a crucial step in ensuring the effectiveness of network training and the stability of results. For the temperature extraction task, the input data needs to be preprocessed and regularized, and the labels need to be normalized during supervised learning to guarantee the effectiveness of network training.
[0024] 3. The use of Mean Squared Error (MSE) as the loss function during DeepLabv3+ network model training directly impacts the network's training performance and final temperature extraction capability. During training, it's crucial to consider whether to penalize large error points and the sensitivity to smaller error points. This invention uses the MSE loss function as the objective function for supervised learning, and its expression is:
[0025] in, The number of training samples. For the first The true temperature of each sample The temperature value predicted by the model.
[0026] 4. For BOTDA measurement data, the impact of the DeepLabv3+ network structure on temperature extraction performance needs to be studied in depth. For example, the input layer needs to be adapted to the number of data points in the frequency sweep window, the convolutional layer needs to continuously adjust the convolutional kernel size to match the image features, and the output layer needs to match the output temperature label.
[0027] The present invention applies Deeplabv3+ to temperature extraction in a BOTDA sensing system. The key technical points are as follows: First, regarding the setup of the training dataset, considering the generalization ability of the network model after training, the training dataset should be constructed using simulation data. The network needs to learn the BGS features when variables such as temperature and linewidth change over a wide range to achieve accurate temperature extraction. According to equation (2), the Brillouin gain coefficient and phonon lifetime depend on the type of fiber under test. The impact of temperature changes on the BGS spectrum mainly lies in the change of BFS. Corresponding to the range of BFS offset δv, the temperature offset ΔT range is set to -2 °C ~ 102 °C, with a step size of 1 °C. The Brillouin linewidth in standard single-mode fiber is approximately 30 MHz. The Brillouin linewidth range in the equation is set to 25 MHz ~ 80 MHz, with a step size of 1 MHz. To increase the robustness of the network, a noisy BGS with a signal-to-noise ratio of 10 dB is added to the training set. Subsequently, these BGS are stacked in a two-dimensional image format with distance as the row and frequency as the column: each row corresponds to a fiber distance sampling point, each column corresponds to a frequency sampling point, and the pixel value is the Brillouin gain amplitude at that (distance, frequency) position. The amplitude is attenuated according to the loss coefficient along the fiber. The number of pixels N in the final distance-frequency shift image is N = frequency sampling interval × distance window length / distance sampling interval.
[0028] Secondly, during supervised learning, the images used to construct the training set need to be labeled with appropriate temperature. Since the training images are distance-frequency maps, the features required to extract temperature depend on the position around the peak of the gain spectrum at each measurement point. By setting appropriate temperature labels for each different temperature condition, the correct temperature label can be set for each training image during supervised learning, which helps the DeepLabv3+ model learn the mapping relationship between image features and temperature, thus achieving accurate temperature extraction.
[0029] Before inputting images into the DeepLabv3+ network model, image preprocessing is typically required. Image preprocessing increases data diversity, thus helping the model generalize better to new data. Image normalization, a general-purpose image preprocessing technique, ensures that the input image data has a uniform numerical range and distribution, helping to avoid biases between different data features and accelerating the model's convergence process. Normalizing image pixel values to the [0, 1] range is a common image preprocessing technique, which involves dividing each pixel value of the image by 255 (i.e., the maximum pixel value of the image) to scale it to between 0 and 1. This method is suitable for most cases. Simultaneously, due to the large range of label variations in the training data, and considering the transferability of the network model, the labels during training also need to be normalized. This helps the model better utilize the data, reduces overfitting or underfitting problems, and improves the model's adaptability and generalization ability to real-world data.
[0030] The Brillouin gain spectrum mathematically exhibits a Lorentzian shape. Traditional BOTDA temperature extraction converts the measured BFS into a corresponding temperature change by establishing a pre-calibrated relationship between the BFS and temperature. This calibration process is typically performed through the measurement of the temperature coefficient of the fiber under test. This calibration requires high-precision calibration and adjustment to ensure the accuracy of temperature measurements. Subsequently, the measurement data is reconstructed in the spatial and frequency domains. BOTDA systems offer high spatial and temporal resolution, enabling accurate reconstruction of the Brillouin gain over the entire fiber length. These resolutions can be adjusted according to system design and application requirements. Since these reconstructed Brillouin gain spectra exhibit a Lorentzian shape in the frequency domain, curve fitting for each BGS yields information such as the linewidth and gain peak frequency. The system then converts the processed data into a temperature distribution map or curve, displaying the temperature changes at different points within the fiber.
[0031] The distance-frequency plot of the Brillouin gain spectrum can be used as the image input for the DeepLabv3+ model. Because dilated convolution is introduced into the network architecture, the properly trained model can analyze image pixels, that is, the Brillouin gain obtained at a certain point in the optical fiber at a certain modulation frequency. Equation (4) gives the output of the dilated convolution: (4) in, x For the input feature map, y To output the feature map, w For convolution kernel, dilation rate r This determines the stride used when sampling the input signal. Standard convolution is... r The case where it is 1. By adjusting... r It allows for flexible adjustment of the receptive field of the convolution kernel.
[0032] When the DeepLabv3+ model is applied to the temperature extraction process of a Brillouin sensing system, the time-consuming curve fitting process described above is replaced by network model temperature extraction. The process is simplified to inputting the Brillouin gain distribution image into the trained network model (the "black box"), and the network directly outputs the corresponding temperature distribution value. This process has many advantages over traditional curve fitting. First, the network model can automatically learn complex features and patterns in the data without requiring explicit mathematical model assumptions beforehand, making it suitable for complex and nonlinear peak extraction problems. It also has better tolerance for spectral distortion. Second, the network extraction method has better fault tolerance. Since neural networks typically have a certain degree of fault tolerance, they can handle situations with high noise levels or incomplete data because they can learn general feature representations through training, exhibiting stronger robustness.
[0033] In another embodiment of the present invention, a method for temperature extraction using a distributed Brillouin fiber optic sensor based on DeepLabv3+ is provided, comprising: S1: Generate Brillouin gain spectrum images under different temperature, linewidth and noise conditions through simulation, and label each image with the corresponding temperature label to form a training dataset; S2: Normalize the Brillouin gain spectrum image and standardize the temperature labels; S3: Construct the DeepLabv3+ network model, in which the DeepLabv3+ network model adopts an encoder-decoder structure. The encoder contains a hollow spatial pyramid pooling module for extracting multi-scale features, and the decoder is used for feature fusion and upsampling. S4: Supervised training of the DeepLabv3+ network model is performed using the training dataset to optimize network parameters, establish the mapping relationship between the Brillouin gain spectrum image and the temperature distribution, and then obtain the temperature distribution along the optical fiber from the measured Brillouin gain spectrum image through the trained DeepLabv3+ network model.
[0034] Furthermore, in network model training, firstly, when applying Deeplabv3+ to a distributed Brillouin optical time-domain analysis sensing system to extract temperature information, it is necessary to train the network model first. To ensure that the trained network model has good generalization ability, simulation data should be used to construct the training dataset. The model needs to be able to identify the Brillouin gain spectrum (BGS) characteristics when variables such as temperature and linewidth change significantly, so as to achieve accurate temperature extraction. According to formula (2), the Brillouin gain coefficient and phonon lifetime depend on the type of fiber being tested. Temperature change mainly affects the center frequency of the BGS spectrum line, which is the corresponding Brillouin frequency shift (BFS), with a range of -2 MHz to 102 MHz and a step size of 1 MHz. In standard single-mode fiber, the Brillouin linewidth is usually about 30 MHz, so the range of linewidth change is set to 25 MHz to 80 MHz with a step size of 1 MHz. After traversing these variables, a total of 105 × 56 = 5880 different Brillouin gain spectra can be obtained. Add corresponding temperature labels to these BGS, which can be derived from the linear relationship between Brillouin frequency shift and temperature. Figure 1 The distance-frequency distribution plots of the Brillouin gain in the BOTDA system simulation are given respectively. Figure 1 (a) contains no noise. Figure 1 (b) The signal-to-noise ratio of the BGS trajectory at the fiber end is 15 dB. It can be seen that the Brillouin gain reaches its maximum value at the BFS, forming a significant gain peak. This feature can be used as a learning object for neural networks to establish its mapping relationship with the temperature output.
[0035] Before inputting image data into the DeepLabv3+ network, image preprocessing is typically required to help the model generalize better to new datasets. Image normalization is a common preprocessing technique that gives the input image data a uniform numerical range and distribution, reducing bias between different data features and accelerating model convergence. Normalizing image pixel values to the [0, 1] interval involves dividing each pixel value by 255 (the maximum possible pixel value in the image), scaling it to between 0 and 1. This method is suitable for most cases. To enhance the network's robustness, the training set should also include noisy BGS samples with a signal-to-noise ratio of 10dB. The training set contains a total of 11760 different BGS, each corresponding to a temperature label. The training set BGS data is then input into the network model, enabling the network to establish a non-linear mapping between the input BGS and the output temperature label during training. Various network learning parameters, such as the learning rate and number of training epochs, need to be adjusted during this process to ensure the final root mean square error of the network model is less than 1 MHz. After validation on the test set, the network training is complete.
[0036] like Figure 2 The diagram illustrates the flowchart for extracting BGS temperature information using a DeepLabv3+ network model. The network is trained using a simulation dataset, with feedback provided to adjust network parameters based on the training results. After training, the network model establishes a mapping between the input BGS image and its corresponding temperature label, enabling it to output the temperature distribution from the input BGS image. Subsequently, the BGS distribution measured using a BOTDA device is preprocessed using the same method as the training data and input into the network. The error between the extracted temperature and the actual temperature is calculated, thus evaluating the network model's performance.
[0037] Furthermore, in the temperature extraction of experimental data, after the network training is completed, a distributed Brillouin time-domain analysis system needs to be built to verify the network performance using experimental data. In the experiment, the length of the optical fiber under test is 20 km. Approximately 100 m of the fiber's tail is wound down and placed in a temperature chamber to be heated to different temperatures. Using the experimentally measured 3D-BGS distribution, the data is preprocessed and then input into the network model. After calibration of the fiber temperature coefficient, the extraction process from the Brillouin gain spectrum distribution to the temperature distribution along the fiber can be directly completed, eliminating the cumbersome curve fitting process. The root mean square error of the final experimental temperature extraction result can reach below 1 ℃. During the experiment, it is necessary to avoid the non-local effects caused by excessive input optical power; the input power should be kept below -13 dBm. After verification with experimental data, the trained network model can be transferred to various Brillouin optical time-domain analysis systems with different system parameters after simple temperature coefficient adjustments.
[0038] like Figure 3 The diagram shows the structure of the DeepLabv3+ network model. It combines a deep convolutional neural network with encoder and decoder modules. The deep convolutional neural network serves as the backbone network to extract features from the input image. The decoder upsamples and fuses the extracted features to generate the final semantic segmentation result. DeepLabv3+ employs a global pooling layer and a low-level feature fusion module, which helps improve segmentation accuracy.
[0039] In another embodiment of the present invention, to facilitate better implementation of the method provided in the embodiments of the present invention, an apparatus based on the above method is also provided. The meanings of the terms used are the same as in the above method, and specific implementation details can be found in the description of the method embodiments.
[0040] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of the device provided in an embodiment of the present invention, wherein the device may include a dataset construction module 401, a preprocessing module 402, a model construction module 403, and a training module 404, wherein: The dataset construction module 401 is used to generate Brillouin gain spectrum images under different temperature, linewidth and noise conditions through simulation, and to label each image with the corresponding temperature label to form a training dataset. The preprocessing module 402 is used to normalize the Brillouin gain spectrum image and standardize the temperature labels. Model building module 403 is used to build the DeepLabv3+ network model. The DeepLabv3+ network model adopts an encoder-decoder structure. The encoder contains a holed spatial pyramid pooling module for extracting multi-scale features, and the decoder is used for feature fusion and upsampling. Training module 404 is used to supervise the training of the DeepLabv3+ network model using the training dataset, optimize the network parameters, establish the mapping relationship between the Brillouin gain spectrum image and the temperature distribution, and then obtain the temperature distribution along the optical fiber from the measured Brillouin gain spectrum image through the trained DeepLabv3+ network model.
[0041] In some instances, the training dataset was constructed with the following parameter settings: temperature range: -2°C to 102°C, with a step size of 1°C; Brillouin linewidth range: 25 MHz to 80 MHz, with a step size of 1 MHz; noise conditions: including noisy BGS images with and without noise and with a signal-to-noise ratio of 10 dB.
[0042] In some instances, normalizing the Brillouin gain spectrum image involves scaling the pixel values to the [0,1] interval; temperature labels are normalized by mapping temperature values to the [0,1] interval.
[0043] In some instances, the loss function used during training is the mean squared error loss function, the optimizer is the Adam optimizer, and the initial learning rate is set to 1×10⁻⁶. -4 .
[0044] In some instances, the measured Brillouin gain spectrum images were obtained from the Brillouin optical time-domain analysis system, and the image format was a distance-frequency two-dimensional matrix.
[0045] The above provides a detailed description of a temperature extraction method and apparatus based on a distributed Brillouin fiber optic sensor using DeepLabv3+, as provided in the embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for temperature extraction using a distributed Brillouin fiber optic sensor based on DeepLabv3+, characterized in that, include: Brillouin gain spectrum images under different temperature, linewidth and noise conditions were generated by simulation, and each image was labeled with a corresponding temperature label to form a training dataset. The Brillouin gain spectrum image was normalized, and the temperature labels were standardized. Construct a DeepLabv3+ network model, wherein the DeepLabv3+ network model adopts an encoder-decoder structure, the encoder includes a holed spatial pyramid pooling module for extracting multi-scale features, and the decoder is used for feature fusion and upsampling; The DeepLabv3+ network model is trained under supervision using the training dataset to optimize network parameters, establish a mapping relationship between the Brillouin gain spectrum image and the temperature distribution, and then obtain the temperature distribution along the optical fiber from the measured Brillouin gain spectrum image through the trained DeepLabv3+ network model.
2. The method according to claim 1, characterized in that, The construction of the training dataset includes the following parameter settings: temperature range: -2°C to 102°C, step size 1°C; Brillouin linewidth range: 25 MHz to 80 MHz, step size 1 MHz; noise conditions: including noise-free and noisy BGS images with a signal-to-noise ratio of 10 dB.
3. The method according to claim 2, characterized in that, The Brillouin gain spectrum image is normalized by scaling the pixel values to the [0,1] interval; the temperature label is normalized by mapping the temperature values to the [0,1] interval.
4. The method according to claim 3, characterized in that, The loss function used during training is the mean squared error loss function, the optimizer is the Adam optimizer, and the initial learning rate is set to 1×10. -4 .
5. The method according to claim 1, characterized in that, The measured Brillouin gain spectrum image was obtained from the Brillouin optical time-domain analysis system, and the image format was a distance-frequency two-dimensional matrix.
6. A temperature extraction device based on a distributed Brillouin fiber optic sensor using DeepLabv3+, characterized in that, include: The dataset construction module is used to generate Brillouin gain spectrum images under different temperature, linewidth and noise conditions through simulation, and to label each image with the corresponding temperature label to form a training dataset. The preprocessing module is used to normalize the Brillouin gain spectrum image and standardize the temperature labels. The model building module is used to build the DeepLabv3+ network model. The DeepLabv3+ network model adopts an encoder-decoder structure. The encoder includes a holed spatial pyramid pooling module for extracting multi-scale features, and the decoder is used for feature fusion and upsampling. The training module is used to supervise the training of the DeepLabv3+ network model using the training dataset, optimize the network parameters, establish the mapping relationship between the Brillouin gain spectrum image and the temperature distribution, and then obtain the temperature distribution along the optical fiber from the measured Brillouin gain spectrum image through the trained DeepLabv3+ network model.
7. The apparatus according to claim 6, characterized in that, The construction of the training dataset includes the following parameter settings: temperature range: -2°C to 102°C, step size 1°C; Brillouin linewidth range: 25 MHz to 80 MHz, step size 1 MHz; noise conditions: including noise-free and noisy BGS images with a signal-to-noise ratio of 10 dB.
8. The apparatus according to claim 7, characterized in that, The Brillouin gain spectrum image is normalized by scaling the pixel values to the [0,1] interval; the temperature label is normalized by mapping the temperature values to the [0,1] interval.
9. The apparatus according to claim 8, characterized in that, The loss function used during training is the mean squared error loss function, the optimizer is the Adam optimizer, and the initial learning rate is set to 1×10. -4 .
10. The apparatus according to claim 6, characterized in that, The measured Brillouin gain spectrum image was obtained from the Brillouin optical time-domain analysis system, and the image format was a distance-frequency two-dimensional matrix.