A fiber fault detection system and method based on optical time domain reflectometry

Abstract

The application discloses an optical fiber fault detection system and method based on optical time domain reflectometry (OTDR). The method collects reflected signals through an OTDR detection module, decomposes, denoises and reconstructs the signals using wavelet transform, removes high-frequency noise and retains key features, then inputs the processed signals into a convolutional neural network (CNN) model for feature extraction and classification, achieving automatic fault type identification and precise location positioning, wherein the threshold is dynamically adjustable. The system includes a light source module, a circulator, a fiber under test, an OTDR detection module, and a data processing module. The light source module emits dual-wavelength optical pulses to solve the short-range blind area problem of traditional OTDR and enhances the adaptability to complex environments with a dynamic threshold algorithm. Through the intelligent fusion of wavelet transform and CNN, the signal denoising capability, fault detection accuracy and automation level are significantly improved, the artificial analysis cost is reduced, and the detection needs of modern optical fiber networks for high precision and real-time are met.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical fiber communication technology, in particular to an optical fiber fault detection system and method based on optical time domain reflectometry (OTDR). BACKGROUND

[0002] Optical time domain reflectometer (OTDR) is a detection tool widely used in the field of optical fiber communication. It transmits high-energy optical pulses into the optical fiber, and when the optical pulses encounter joints, breakpoints, loss change points and other positions during transmission, reflected light and backscattered light are generated. OTDR receives and analyzes these returned optical signals, measures the round-trip time and optical power change of the optical signals, and accurately measures the length, loss distribution, fault point position and other key parameters of the optical fiber. However, the traditional OTDR has a blind area in short-distance optical fiber measurement, and has limitations in the detection accuracy of weak signals, which is difficult to meet the needs of modern optical fiber networks for high-precision and high-resolution detection. SUMMARY

[0003] In view of the problems in the prior art, the present application provides an optical fiber fault detection method based on optical time domain reflection, characterized in that the method comprises the following steps:

[0004] S1: acquiring a reflected signal by using an OTDR detection module;

[0005] S2: performing wavelet transform denoising processing on the signal;

[0006] S3: inputting the denoised signal into a CNN model for feature extraction;

[0007] S4: outputting the fault type and position by the CNN model;

[0008] The step S2 specifically comprises the following steps:

[0009] S21: performing wavelet transform on the reflected signal, and using the wavelet to decompose the signal;

[0010] S22: after decomposing the signal, removing the high-frequency noise components in the wavelet coefficients that are lower than a certain threshold, and retaining the main signal features;

[0011] S23: reconstructing the signal by inverse wavelet transform to obtain a clear denoised signal as the input of the CNN model.

[0012] Preferably, the step S21 specifically comprises the following steps:

[0013] S211: selecting a wavelet basis function;

[0014] S212: setting the decomposition layer number N;

[0015] S213: decompose the signal into a low frequency part and a high frequency part using a discrete wavelet transform;

[0016] S214: continue to decompose the low frequency part until a set number of layers is reached.

[0017] Preferably, the detection method of the fault point E is as follows:

[0018]

[0019] where CD i is the high frequency component of the i-th layer; when E is greater than a set threshold value, it is considered that the point has a fault.

[0020] Preferably, the detection method of the fault point E is as follows: find the abrupt point by the wavelet modulus maximum detection method; calculate the first derivative of the wavelet coefficient, find the local maximum, and the point where the maximum value is located is the fault point.

[0021] Preferably, the step of determining that the point has a fault when E is greater than a set threshold value further comprises a step of calculating the threshold value; the threshold value is a dynamically adjustable threshold value.

[0022] The application also provides a detection system for implementing the above detection method, which comprises a light source module, a circulator, a fiber to be tested, an OTDR detection module, and a data processing module.

[0023] Preferably, the light source module emits light pulses of two different wavelengths.

[0024] Preferably, the circulator is used to guide the light pulses emitted by the light source module into the fiber to be tested, and guide the returned reflected light signal to the OTDR detection module.

[0025] Preferably, the OTDR detection module comprises a photodetector and a data acquisition system; the photodetector converts the received light signal into an electrical signal, and the data acquisition system measures the time delay and the change in optical power.

[0026] Preferably, the data processing module uses a wavelet transform algorithm to denoise the reflected signal and improve the signal quality; then, the CNN model is used to extract features and classify the processed signal, automatically identify the fault type and calculate the fault position.

[0027] Compared with the prior art, the application has at least the following beneficial effects:

[0028] 1) This invention uses two different wavelengths of light pulses and the reflected signals received by an OTDR, combined with an intelligent signal processing algorithm, to accurately calculate the location of fiber optic faults. This technology effectively solves the blind zone problem of traditional OTDRs in short-distance fiber optic detection and improves fault location accuracy.

[0029] 2) Wavelet transform denoising and convolutional neural network (CNN) fault identification algorithms are introduced to denoise and enhance the reflected signal received by the OTDR, and automatically identify the fault type, such as fiber breakage, excessive joint loss, and fiber bending. This technology reduces the workload of manual analysis and improves detection efficiency.

[0030] 3) Using wavelet denoising combined with fault point identification can more accurately detect fiber optic faults and improve maintenance efficiency. This includes the necessary step of setting a threshold T for fault points. This method combines wavelet analysis and noise estimation to calculate an adaptive threshold (T), which can accurately detect fiber optic fault points and can be combined with CNN for intelligent classification. Compared with the traditional fixed threshold method, this method is more suitable for complex environments, especially when the signal-to-noise ratio varies greatly. The advantages of applying wavelet transform (db4) to OTDR signal analysis are: it can provide both time and frequency domain information simultaneously, and is suitable for non-stationary signals; it can effectively separate noise and abrupt signals, improving fault identification capabilities; and compared with Fourier transform, it is more suitable for abrupt detection (such as fiber breakage, bending, etc.). Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the fiber optic fault detection system of the present invention.

[0032] Figure reference numerals: 100, Fiber optic fault detection system; 10, Light source module; 20, Circulator; 30, Fiber optic cable under test; 40, OTDR detection module; 50, Data processing module.

[0033] The present invention will now be described in further detail. However, the examples described below are merely simplified examples of the present invention and do not represent or limit the scope of protection of the present invention. The scope of protection of the present invention is determined by the claims. Specific Implementation

[0034] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0035] like Figure 1 As shown, the fiber optic fault detection system 100 provided by the present invention includes a light source module 10, a circulator 20, a fiber under test 30, an OTDR detection module 40, and a data processing module 50.

[0036] (I) Light Source Module

[0037] The light source module is one of the key components of the system, responsible for emitting light pulses of two different wavelengths. The light source module uses a tunable pulsed laser, which can adjust the wavelength and pulse width of the emitted light pulses as needed. For different detection requirements, the wavelength selection of the light source module should be determined based on the characteristics of the optical fiber and the receiving range of the OTDR detection module.

[0038] (2) Circulator

[0039] A circulator is used to guide the optical pulses emitted by the light source module into the optical fiber under test and to direct the reflected light signal back to the OTDR detection module. The circulator should have low insertion loss and high isolation to ensure signal quality and transmission efficiency.

[0040] (III) OTDR Detection Module

[0041] The OTDR detection module includes a high-sensitivity photodetector and a high-speed data acquisition system. The photodetector converts the received optical signal into an electrical signal, and the data acquisition system measures the time delay and changes in optical power. This module needs to have high time resolution and high dynamic range to ensure high-precision fault location.

[0042] (iv) Data Processing Unit

[0043] The data processing unit employs wavelet transform algorithms to denoise the reflected signal, improving signal quality. Then, a convolutional neural network (CNN) is used to extract features and classify the processed signal, automatically identifying the fault type and calculating the fault location. This intelligent algorithm significantly reduces manual intervention, improving the efficiency and accuracy of fault diagnosis.

[0044] (V) Optical Fiber Under Test

[0045] The fiber optic cable under test is the object of inspection. This fiber optic network may have different types of faults, such as breaks, excessive connector loss, and fiber bending.

[0046] Example 1: Dual-wavelength pulse fault location

[0047] Assuming the fiber optic cable under test is 1000 meters long, and the light source module emits light pulses with wavelengths of 1310 nm and 1550 nm, the OTDR detection module records the return times of the two wavelengths of light pulses respectively. Assuming the return time of the 1310 nm light pulse is 10 microseconds and the return time of the 1550 nm light pulse is 10.5 microseconds, the distance to the fault point from the light source module is calculated to be 500 meters. This technology effectively solves the blind zone problem of traditional OTDRs in short-distance fiber optic cable testing.

[0048] Specific calculation steps: Assume the light source module emits light pulses, which propagate through optical fiber. When a fault point is encountered, part of the light pulse is reflected back to the OTDR detection module. The OTDR calculates the distance to the fault point based on the return time of the light pulse.

[0049] The calculation formula is as follows:

[0050] Distance S = (c*Δt) / 2;

[0051] Where c is the speed of light in the optical fiber (the speed of light in an optical fiber is typically about 2 × 10⁻⁶). 8 m / s (depending on the refractive index of the optical fiber); Δt is the time difference between the transmitted pulse and the return of the reflected pulse (unit: seconds).

[0052] The formula divides by 2 because the time it takes for the light signal to travel from the light source to the fault point and back is divided by 2 to obtain the one-way distance.

[0053] Assuming the time difference between the returned signal received by the OTDR is 10 microseconds, the distance to the fault point can be calculated as follows:

[0054] Distance S = ((2×10) 8 m / s)×(10×10 -6 s)) / 2=500 meters

[0055] This result indicates that the fault point is 500 meters away from the light source module.

[0056] The above embodiments, 1) address the short-distance blind zone problem: Traditional OTDRs encounter a "blind zone" problem when measuring short-distance optical fibers. The blind zone arises because the propagation time of the optical pulse is too short, and the time difference between the reflected and transmitted signals is very small, making it impossible to accurately measure faults near the light source. However, by using dual-wavelength pulses (e.g., 1310nm and 1550nm), different wavelengths of light can propagate separately, and the time difference between them can improve the measurement accuracy over short distances. Because different wavelengths of light propagate at slightly different speeds in optical fibers, this can be used to improve fault location over short distances.

[0057] 2) Improved positioning accuracy: By using two light pulses of different wavelengths, more information about the characteristics of the optical fiber can be obtained, making the calculation of the fault location more accurate. Specifically, different wavelengths of light have different propagation losses and scattering characteristics in the optical fiber, which provides more references for data analysis, thereby more accurately deduce the fault point.

[0058] 3) Enhanced signal analysis capabilities: By utilizing dual-wavelength technology, the system can receive more data. This data, processed by intelligent algorithms (such as wavelet transform denoising and CNN fault identification), can not only accurately locate the fault point but also automatically determine the fault type based on the characteristics of the reflected light, such as fiber breakage, connector loss, or excessive bending. Compared to traditional single-wavelength systems, the increased signal processing capabilities significantly improve the intelligence level of detection and reduce the complexity of manual analysis.

[0059] 4) Enhanced processing capability for weak signals: When detecting weak signals or long-term fiber optic transmission, the use of dual wavelengths can provide greater signal redundancy. Even in weak reflected signals, more information can be extracted through the signal differences between different wavelengths, thereby improving the detection of weak faults. The greatest advantage of the dual-wavelength pulse method lies in improving the accuracy of short-distance fault location, solving the blind zone problem of traditional OTDRs in short distances, and enabling more accurate fault identification and location through richer signal information. This allows the system to provide higher accuracy and more intelligent analysis when dealing with complex fiber optic networks, especially in the processing of weak signals and fault identification, outperforming traditional single-wavelength OTDRs. However, the introduction of this technology also increases the complexity and cost of the system and places higher demands on hardware and algorithms.

[0060] Compared to traditional calculation methods, the dual-wavelength pulse approach offers significant improvements in specific applications (such as short-range, weak signal detection). Therefore, this calculation method is not entirely the same as conventional methods, but rather offers advantages in terms of greater accuracy and intelligence, making it particularly suitable for modern fiber optic network detection capabilities that demand high precision and resolution.

[0061] Example 2

[0062] OTDR devices receive complex reflected signals during fiber optic fault detection. These signals typically include noise, signal attenuation, scattering effects, and weak fault signals (such as fiber breaks and splice losses). Traditional OTDR detection methods rely on simple signal analysis (such as peak detection and time difference calculation), which have low accuracy in noisy environments and are particularly limited in their ability to identify weak signals and complex fault types.

[0063] Therefore, the goal of this embodiment is to use wavelet transform denoising and convolutional neural network (CNN) fault identification algorithm to process the reflected signal received by OTDR, improve the accuracy of fault location and automatically identify the fault type.

[0064] Before implementing a convolutional neural network (CNN), noise reduction is required for the OTDR reflection signal. Traditional OTDR reflection signals contain noise components, which may originate from fluctuations in the light source, scattering effects of the optical fiber itself, or other external interference. To improve the accuracy and stability of the CNN model, we use wavelet transform to denoise the signal.

[0065] Wavelet transform denoising steps:

[0066] S21: Perform wavelet transform on the reflected signal received by the OTDR, and use a wavelet such as Daubechies (db4) for signal decomposition.

[0067] S22: After decomposing the signal, remove high-frequency noise components below a certain threshold from the wavelet coefficients, while retaining the main signal features;

[0068] S23: Reconstruct the signal through inverse wavelet transform to obtain a clear, denoised signal, which is then used as the input to the CNN.

[0069] The main purpose of performing wavelet transform (such as Daubechies wavelet (db4)) on the reflected signal received by the OTDR is to perform time-frequency analysis of the signal in order to extract effective features, remove noise, and improve fault detection accuracy. This processing method can effectively reduce the interference of noise on the signal, retain useful fault features, and ensure that the CNN model can identify faults more accurately.

[0070] Wavelet Transform (WT) is a time-frequency analysis method that can simultaneously provide information about a signal in both the time and frequency domains, making it particularly suitable for processing non-stationary signals (such as OTDR reflection signals). Compared to the traditional Fourier Transform (FFT), wavelet transform can provide different resolutions across different frequency ranges, which helps in identifying abrupt changes in optical fibers (such as joint loss, breaks, microbending, and other faults).

[0071] The Daubechies wavelet (db4) is a wavelet basis with good compact support and smoothness, suitable for the analysis of fiber optic reflection signals.

[0072] The measurement result from an OTDR is a one-dimensional time-series signal, where the horizontal axis represents propagation time or fiber distance (in meters), and the vertical axis represents the echo signal strength (in dB). Assume the signal acquired by the OTDR is as follows:

[0073] Distance (m) Reflection intensity (dB) 0 -30 10 -25 20 -40 30 -35 40 -45 50 -60 60 -45 70 -35 80 -30 90 -20 100 -60

[0074] ...

[0075] As shown in the table above, a fracture fault (signal drop) occurred at 100 meters.

[0076] The specific steps of signal decomposition S21 are as follows:

[0077] S211: Select wavelet basis functions (such as db4);

[0078] S212: Set the number of decomposition layers N (generally 2 to 4 layers are selected);

[0079] S213: The signal is decomposed into a low-frequency component (Approximation Coefficients, CA) and a high-frequency component (Detail Coefficients, CD) using Discrete Wavelet Transform (DWT);

[0080] S214: Continue to decompose the low-frequency part until the set number of layers is reached.

[0081] Among them, CA2: the low-frequency approximation component of the second layer, representing the overall trend of the signal; CD2: the high-frequency detail component of the second layer, representing medium-frequency components, such as fiber optic connector loss; CD1: the high-frequency detail component of the first layer, representing high-frequency noise, such as environmental interference.

[0082] The Daubechies 4th order wavelet (db4) was chosen as the mother wavelet. Its characteristics include: good time-frequency localization properties; effective decomposition of complex signals; and suitability for signal denoising and fault detection.

[0083] Analyzing the results of wavelet decomposition, we can find that:

[0084] Low-frequency component (CA2): Represents the overall trend of the signal, which is relatively smooth. High-frequency component (CD2, CD1): The location of the fracture fault (100 meters) will appear as a significant abrupt change in CD2 or CD1; thresholding can remove noise and highlight the fault characteristics.

[0085] Fault detection methods:

[0086] Method 1: Calculate the energy of the detail coefficients. If the energy at a certain point is higher than a set threshold, it may be a fault point.

[0087] Fault point

[0088] If E > T (the set threshold), then the point is considered to be faulty; where T is the threshold.

[0089] The process for setting the threshold is as follows:

[0090] A threshold is calculated using wavelet detail coefficient energy analysis combined with noise estimation, specifically designed for fiber optic fault location detection.

[0091] T=μ+β·σ

[0092] Where: μ is the mean of the detail coefficients (CD2); σ is the standard deviation of the detail coefficients; β is an adjustment coefficient (generally taken as 2 to 4, adjusted according to the experiment).

[0093] The steps of this method for calculating the threshold T include: 1. Calculating the mean and standard deviation of the wavelet detail coefficients (CD2); 2. Setting the threshold T, points exceeding T are considered fault points; 3. Applicable to abrupt change detection and can effectively filter noise.

[0094] Below is a practical example of threshold T calculation based on the above fiber optic fault detection: This threshold is not a fixed value, but a dynamically adjustable threshold.

[0095] (1) OTDR reflected signal

[0096] Suppose the fiber optic reflected signal obtained in a certain OTDR test is as follows (unit: dB):

[0097]

[0098] The fiber optic cable broke at a distance of 100 meters, causing the signal to drop sharply to -9.5 dB. The signal remained stable at other locations, with a normal signal level of approximately -5 dB.

[0099] (2) Wavelet Transform (Daubechies db4)

[0100] The signal is decomposed using Daubechies (db4) wavelet decomposition to obtain the detail coefficients (CD2).

[0101]

[0102] At 100 meters, CD2 = -3.5, which is much higher than at other locations, indicating that there may be a fault there; while at other locations, CD2 does not change much and is within the normal noise range.

[0103] (3) Calculate the threshold T

[0104] Calculate the mean and standard deviation of CD2:

[0105] μ=mean({0.1,0.2,0.1,-0.1,0.0,-3.5,0.2,0.1,0.0,-0.1,0.1})=-0.27

[0106] σ=std({0.1,0.2,0.1,-0.1,0.0,-3.5,0.2,0.1,0.0,-0.1,0.1})=1.06

[0107] Set β = 3, calculate the threshold:

[0108] T=μ+β·σ=-0.27+3×1.06=2.91

[0109] (4) Determine the fault point

[0110] At 100 meters, CD2 = -3.5, which is significantly lower than -2.91 (exceeding the threshold), indicating a fault point. The CD2 values ​​at other points are all within the threshold range and are considered normal signals.

[0111] This method combines wavelet analysis and noise estimation to calculate an adaptive threshold (T), enabling accurate detection of fiber optic fault points and allowing for intelligent classification using CNNs. Compared to traditional fixed threshold methods, this approach is more suitable for complex environments, especially when the signal-to-noise ratio varies significantly.

[0112] Method 2: Using modulus maxima detection: Finding abrupt change points using wavelet modulus maxima detection.

[0113] Calculate the first derivative of the wavelet coefficients; find the local maxima (fault points).

[0114] Through wavelet decomposition and analysis, we can observe that:

[0115] 1) Wavelet decomposition successfully extracted the high-frequency anomaly at the fault point (100 meters away);

[0116] 2) The detail coefficient (CD2) shows abrupt changes in the signal and can help identify fiber loss and breaks;

[0117] 3) Fault points can be automatically identified through threshold energy analysis or modulus maximum detection.

[0118] Convolutional Neural Network (CNN) Structure

[0119] Once the signal has undergone wavelet denoising, it can be fed into a CNN for feature extraction and fault classification. CNN is a powerful deep learning model, particularly suitable for processing signal data with spatial local dependencies, such as time series signals (OTDR reflection signals).

[0120] CNN model architecture:

[0121] 1. Input Layer: The input layer receives the OTDR reflection signal after wavelet transform denoising, typically inputting it into the CNN network in the form of a one-dimensional time series. The signal length can be adjusted according to the specific application, with a common length being several thousand data points.

[0122] 2. Convolutional Layer: A convolutional layer contains multiple convolutional kernels used to extract local features from a signal. Each kernel performs a convolution operation on the time series through a sliding window, identifying feature patterns in the signal, such as peaks, reflection points, and abrupt changes in the signal.

[0123] The kernel size can be set to 3×3 or 5×5, the stride is 1, and the number of kernels can be 32 or 64, which can be adjusted as needed.

[0124] The purpose of convolution is to extract important local features from the original signal, such as the peak position of the reflected light and the reflection amplitude.

[0125] 3. Activation layer: The ReLU (Rectified Linear Unit) activation function is used to map the output of the convolutional layer to a nonlinear space, improving the network's ability to fit complex signals.

[0126] 4. Pooling Layer: Max pooling is used to reduce the size of the feature map, thereby reducing computational complexity and helping to prevent overfitting. A commonly used pooling operation is 2×2 max pooling.

[0127] 5. Fully Connected Layer: The feature maps extracted by the convolutional layers are further processed by the fully connected layer. The fully connected layer flattens the features output by the convolutional layer into a one-dimensional vector, and performs a weighted summation using a weight matrix to finally output the predicted fault type.

[0128] 6. Output Layer: The output layer uses the Softmax activation function to convert the output of the CNN model into probabilities of different fault types. Different fault types, such as fiber breakage, excessive connector loss, and fiber bending, will output a probability value representing the likelihood of that fault type.

[0129] Training and optimization

[0130] To train the CNN model, a large number of OTDR reflection signal samples need to be collected, including normal signals and different types of fault signals. These signal samples are preprocessed (including wavelet transform denoising) and then divided into training and test sets.

[0131] During training: (1) The Cross-Entropy Loss Function is used to measure the difference between the predicted results and the true labels. (2) The Adam Optimizer (Adaptive Moment Estimation) is used to optimize the network parameters, adjusting hyperparameters such as learning rate and batch size. (3) Data Augmentation techniques are used to generate more training samples, such as by shifting, scaling, and adding noise on the time axis to enhance the robustness of the model.

[0132] Fault identification and location

[0133] After training, the CNN model can perform real-time analysis of new OTDR reflection signals. After wavelet denoising and feature extraction, the model can automatically identify the fault type in the signal and determine the location of the fault.

[0134] Fault identification process:

[0135] S1: The OTDR detection module collects reflected signals.

[0136] S2: Perform wavelet transform denoising on the signal.

[0137] S3: Input the denoised signal into the CNN model for feature extraction.

[0138] S4: The CNN model outputs the fault type and location. For example, if the model identifies a significant spike in the signal, it may indicate a fiber optic break; if it detects a large and flat amplitude in the reflected signal, it may indicate excessive connector loss.

[0139] By combining CNN and wavelet transform, the fiber optic fault detection system of this invention has made significant improvements in the following aspects: (1) Improved accuracy: Through the deep learning capability of CNN, it can automatically learn to extract complex features from the original signal and identify various types of faults, avoiding the limitations of relying on manual analysis and experience judgment in traditional methods. (2) Noise suppression: Wavelet transform denoising can effectively remove noise in the reflected signal, especially common interference signals in fiber optic systems, enhancing the signal quality and providing a cleaner input for CNN. (3) Automatic fault type identification: The CNN model can identify different types of faults (such as fiber breakage, connector loss, etc.) and provide accurate fault locations, reducing the workload of manual analysis and improving work efficiency. (4) Processing speed: The parallel computing capability of CNN accelerates the fault diagnosis process, significantly improving the detection speed and meeting the real-time detection requirements of modern fiber optic networks.

[0140] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0141] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

[0142] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

Claims

1. A fiber optic fault detection method based on optical time-domain reflectometry, characterized in that: The method includes the following steps: S1: Acquire the reflected signal using the OTDR detection module; S2: Perform wavelet transform denoising on the signal; S3: Input the denoised signal into the CNN model for feature extraction; S4: The CNN model outputs the fault type and location; In steps S3 and S4, the CNN model, through its deep learning capabilities, is able to automatically learn to extract features from the denoised signal to identify the fault type and provide an accurate location. Step S2 specifically includes the following steps: S21: Perform wavelet transform on the reflected signal, and use the wavelet to decompose the signal; S22: After decomposing the signal, remove high-frequency noise components below a certain threshold from the wavelet coefficients, while retaining the main signal features; S23: Reconstruct the signal through inverse wavelet transform to obtain a clear, denoised signal, which is then used as the input to the CNN model; The method further includes a step of detecting fault points after step S2 and before step S3; the specific steps are as follows: Calculate the i-th level high-frequency components of wavelet decomposition The sum of squares; ; Wherein, the CD i Here, E represents the high-frequency component of the i-th layer; when E is greater than a set threshold T, the point is considered to have a fault; N is the number of decomposition layers. Calculate the high-frequency components of the second layer mean and standard deviation ; Calculate the set threshold T, the value of which is based on the second layer high-frequency components. The dynamic adjustment threshold is determined by the statistical characteristics; the set threshold T is based on the formula: Calculated; where It is the adjustment coefficient; When E is greater than the set threshold T, it is determined that there is a fault point in the signal, and the denoised signal is input into the CNN model for feature extraction.

2. The detection method according to claim 1, characterized in that: Step S21 specifically includes the following steps: S211: Select wavelet basis functions; S212: Set the number of decomposition levels N; S213: The signal is decomposed into low-frequency and high-frequency components using discrete wavelet transform; S214: Continue to decompose the low-frequency part until the set number of layers is reached.

3. A detection system implementing any one of the detection methods as described in claims 1-2, characterized in that: The system includes a light source module, a circulator, an optical fiber under test, an OTDR detection module, and a data processing module.

4. The detection system according to claim 3, characterized in that: The light source module emits two different wavelengths of light pulses.

5. The detection system according to claim 3, characterized in that: The circulator is used to guide the light pulse emitted by the light source module into the optical fiber under test, and guide the returned reflected light signal to the OTDR detection module.

6. The detection system according to claim 3, characterized in that: The OTDR detection module includes a photodetector and a data acquisition system; the photodetector converts the received optical signal into an electrical signal, and the data acquisition system measures the time delay and optical power change.

7. The detection system according to claim 3, characterized in that: The data processing module uses wavelet transform algorithm to denoise the reflected signal and improve signal quality; then, it uses the CNN model to extract features and classify the processed signal, automatically identify the fault type and calculate the fault location.

Images