Optical fiber signal processing method based on deep learning optimization

Abstract

The invention discloses an optical fiber signal processing method based on deep learning optimization. According to the invention, the CNN architecture provides strong feature extraction and representation capabilities for optical fiber signal processing. Through the combination of the two convolution layers and the two pooling layers, the model can extract features from simple to complex from original signals layer by layer. The convolution operation utilizes the characteristics of local connection and weight sharing to effectively capture the spatial correlation in the signal, and the pooling operation reduces the data dimension through down-sampling and increases the robustness of the model to the position change at the same time. The hierarchical feature extraction process enables the model to learn the essential features of the signal, and lays a solid foundation for subsequent classification or regression tasks. In addition, the full-connection layer flattens the feature map after convolution and pooling and carries out feature fusion, so that the recognition capability of the model to a complex signal mode is further enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical fiber signal processing, and specifically is an optical fiber signal processing method based on deep learning optimization. Background Art

[0002] With the rapid development of fiber-optic communication technology, the importance of fiber-optic signal processing in communication systems has become increasingly prominent. Traditional fiber-optic signal processing methods mainly rely on artificially designed feature extraction and signal analysis techniques. These methods often show problems such as low efficiency and insufficient accuracy when facing complex and changeable fiber-optic signals. In particular, with the dramatic increase in data transmission volume and the increase in signal complexity, traditional methods have been unable to meet the needs of high-speed and high-precision signal processing. In recent years, deep learning technology has achieved remarkable results in image recognition, natural language processing and other fields. Its powerful automatic feature extraction and model self-learning capabilities provide new ideas for fiber-optic signal processing.

[0003] However, traditional processing methods rely on artificially designed features and are unable to fully capture the complexity and diversity of fiber optic signals. For high-dimensional, nonlinear fiber optic signals, traditional feature extraction methods often cannot effectively identify key features, resulting in insufficient signal processing accuracy. At the same time, existing technologies usually use simple optimization algorithms in the model training process. These algorithms converge slowly when processing large-scale data, are prone to falling into local optimal solutions, and are sensitive to the selection of hyperparameters, requiring a lot of experience and trial and error. Summary of the invention

[0004] The purpose of the present invention is to provide a method for optical fiber signal processing based on deep learning optimization in order to solve the above-mentioned problems.

[0005] The technical solution adopted by the present invention is as follows: a method for processing optical fiber signals based on deep learning optimization, the method comprising the following steps:

[0006] S1: Data acquisition and preprocessing: collect a large amount of raw signal data from the optical fiber transmission system, including normal signals and abnormal signals; perform denoising and normalization operations on the data to ensure data quality; divide the data set into training set, validation set and test set, with a ratio of 70%, 15% and 15%;

[0007] S2: Use Fourier transform to extract the frequency domain features of the signal as one of the input features of deep learning, and design a convolutional neural network (CNN) structure to learn the spatial features of the signal;

[0008] S3: Convolutional neural network (CNN) is selected as the deep learning model, which is suitable for processing signal data with spatial correlation;

[0009] S4: Model training: Use the training set data to train the CNN model, use the Adam optimization algorithm, and set the learning rate to 0.001; use the cross entropy loss function to measure the difference between the model output and the true label; set an early stopping mechanism to stop training when the performance of the validation set no longer improves to avoid overfitting;

[0010] S5: Use the test set data to evaluate the trained CNN model, calculate the performance indicators, adjust the model hyperparameters based on the evaluation results, and consider using data enhancement techniques to improve the generalization ability of the model;

[0011] S6: Use ensemble learning methods to improve overall performance;

[0012] S7: Deploy the trained CNN model to the real-time signal processing system to ensure that the model can efficiently process real-time data and meet real-time requirements;

[0013] S8: Monitor the performance of the CNN model in real time during processing, record processing delay and accuracy indicators, and collect real-time feedback data for continuous optimization and updating of the model;

[0014] S9: Apply this method to an actual fiber-optic communication system to verify its effectiveness and stability, and expand the method according to application requirements, after which the entire fiber-optic signal processing process based on deep learning optimization can be completed.

[0015] In a preferred embodiment, in step S1, first, according to the characteristics of the optical fiber signal and actual needs, a suitable sampling rate or downsampling rate is determined by analyzing the frequency components and change rates of the signal to ensure that important information is not lost; then, a random sampling or stratified sampling method is used to select representative samples from the original data. Random sampling is suitable for situations where the data distribution is relatively uniform, while stratified sampling divides the data into several layers according to certain characteristics of the signal, and then randomly extracts samples from each layer, which is suitable for situations where the data distribution is uneven; then, the extracted samples are further processed, including denoising and normalization to ensure data quality. Denoising uses wavelet transform and filter methods, and normalization scales the data to the same scale for model processing; finally, the processed samples are divided into training set, validation set and test set to ensure that the data distribution in each set is similar, so as to facilitate model training and evaluation; through such a data sampling or downsampling process, while ensuring the representativeness of the data, the computational burden is effectively reduced, providing an efficient and high-quality data foundation for subsequent deep learning model training.

[0016] In a preferred embodiment, in step S2, the convolution layer uses multiple convolution kernels to extract local features, and the pooling layer reduces the dimension and abstracts the features; by stacking multiple convolution layers and pooling layers, the model can automatically learn the multi-level spatial features of the signal; then, a fully connected layer is added to the end of the CNN to combine and abstract the learned features, and the activation function of the final output layer outputs the results according to the task requirements; throughout the design process, it is necessary to consider the size and number of the convolution kernels, the selection parameters of the activation function, and the depth and width of the model to ensure that the model can effectively learn and extract signal features.

[0017] In a preferred embodiment, in step S3, the convolutional neural network (CNN) deep learning model comprises a CNN architecture of two convolutional layers, two pooling layers and one fully connected layer; each component has its own specific calculation formula and parameter definition, which together constitute the mathematical basis of the CNN model, and are used to extract features from input signals and perform classification or regression tasks;

[0018] in:

[0019] The input layer receives the preprocessed signal data in the form of a two-dimensional matrix, including the conversion of time series data into image form;

[0020] Convolution layer 1 uses 32 3x3 convolution kernels to perform convolution operations to extract local features.

[0021] The calculation formula is: (W*X+b); where W is the convolution kernel weight, X is the input data, and b is the bias term; the number of convolution kernels is 32, the convolution kernel size is 3x3, the activation function is ReLU, and the formula is f(x)=\max(0,x)f(x)=max(0,x)

[0022] Pooling layer 1 uses 2x2 maximum pooling for downsampling to reduce the data dimension. The calculation formula is: max(X i,j ,X i+1,j ,X i,j+1 ,X i+1,j+1 ), where X i,j Represents the pixel value in the pooled area; the pooling kernel size is 2x2, and the pooling type is maximum pooling;

[0023] Convolutional layer 2 uses 64 3x3 convolution kernels for convolution operations to further extract more complex features. The calculation formula is the same as that of convolutional layer 1; the number of convolution kernels is 64, the convolution kernel size is 3x3, and the activation function is ReLU;

[0024] Pooling layer 2 uses 2x2 maximum pooling for downsampling; the calculation formula is the same as pooling layer 1; the parameter definition is the same as pooling layer 1;

[0025] The fully connected layer flattens the feature map after convolution and pooling, and performs feature fusion through the fully connected layer. The calculation formula is: W T X+b, where W is the weight matrix of the fully connected layer, X is the flattened feature vector, b is the bias term, and the number of neurons is 128;

[0026] The output layer uses the softmax activation function to output the category probability according to the task requirements. The calculation formula is: Among them, zi is the i-th element of the fully connected layer output, and the number of output categories is determined by the specific task. For multiple classifications, it is the number of categories.

[0027] In a preferred embodiment, in step S4, the preprocessed training set data is input into the CNN model during the training of the CNN model using the training set data, the output is calculated by forward propagation, and then the model parameters are updated by back propagation.

[0028] Adam is an adaptive learning rate optimization algorithm that can adjust the learning rate for each parameter separately. The calculation formula includes: t =β 1 m t-1 +(1-β 1 ) t ;

[0029]

[0030]

[0031] Among them, m t and v t are respectively the momentum and variance at time t, g t is the gradient, θ t is a parameter, α is the learning rate, β 1 and β 2 is a hyperparameter, a small constant to avoid division by zero;

[0032] The cross entropy loss function is used to measure the difference between the model output and the true label. The calculation formula is:

[0033]

[0034] Among them, p(xi) is the probability distribution of the true label, and q(xi) is the probability distribution predicted by the model;

[0035] The monitoring indicator is the loss or accuracy of the validation set; the stopping condition is that the monitoring indicator has not improved for multiple consecutive epochs; patience is the number of epochs tolerated, and training is stopped if there is no improvement after this number.

[0036] In a preferred embodiment, in step S5, the test set data is data that the model has never seen and can objectively reflect the performance of the model; then the various performance indicators of the model are calculated, including accuracy, recall rate, and F1 score. These indicators can comprehensively evaluate the performance of the model in the classification task; then the error cases of the model are analyzed to find out the shortcomings of the model, and through error analysis, the model is improved in a targeted manner to improve its performance; finally, according to the evaluation results, the model parameters or structure are adjusted for further optimization. Model evaluation is a continuous process that requires continuous iteration and improvement.

[0037] In a preferred embodiment, in step S6, a reinforcement learning mechanism is introduced to enable the model to adjust parameters according to feedback in real-time processing to achieve self-optimization; this includes defining a reward function that evaluates the performance of the model based on the accuracy and real-time indicators of signal processing; when processing signals, the model adjusts its internal parameters, including the weights and biases of CNN, according to the feedback of the reward function to maximize the long-term cumulative rewards; this reinforcement learning mechanism enables the model to adaptively optimize its performance in a constantly changing signal environment.

[0038] In a preferred embodiment, in step S7, the input interface needs to convert the real-time signal into the format required by the model, and the output interface needs to convert the output result of the model into a format available to the system; then, the model deployment is tested and verified to ensure the stability and reliability of the model in the actual environment, which includes testing the processing speed and accuracy performance indicators of the model, and verifying the performance of the model in different scenarios; finally, the model deployment is optimized and adjusted according to the test and verification results.

[0039] In a preferred embodiment, in step S8, a real-time data acquisition system is designed, which is responsible for continuously monitoring the results of optical fiber signal processing and system performance indicators, including signal quality, processing delay, and bit error rate; these real-time feedback data are transmitted to a data storage and analysis platform; then the model is learned online or periodically retrained using the collected real-time feedback data; online learning means that the model is fine-tuned according to the real-time feedback data without interrupting service to quickly adapt to new signal characteristics or environmental changes; this is achieved through incremental learning and online gradient descent methods; periodic retraining is to use the accumulated feedback data to comprehensively retrain the model within a certain time interval to update the model's weights and parameters and improve its long-term performance.

[0040] In a preferred embodiment, in step S9, a long-term stability test is performed to observe the performance changes of the model under long-term operation; this includes monitoring the processing delay, resource consumption, and failure rate indicators of the model to ensure the stability and reliability of the model in practical applications; during the verification process, detailed performance data and analysis results are collected to evaluate whether the model meets the predetermined performance indicators; if the model performance does not meet expectations, it is necessary to go back to the previous steps, such as data preprocessing and model training, for targeted optimization and adjustment.

[0041] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0042] 1. In the present invention, the CNN architecture provides powerful feature extraction and representation capabilities for fiber optic signal processing. Through the combination of two convolutional layers and two pooling layers, the model can extract features from simple to complex from the original signal layer by layer. The convolution operation uses the characteristics of local connection and weight sharing to effectively capture the spatial correlation in the signal, while the pooling operation reduces the data dimension by downsampling while increasing the model's robustness to position changes. This hierarchical feature extraction process enables the model to learn the essential characteristics of the signal, laying a solid foundation for subsequent classification or regression tasks. In addition, the fully connected layer flattens the feature maps after convolution and pooling and performs feature fusion, further enhancing the model's ability to recognize complex signal patterns.

[0043] 2. In the present invention, the Adam optimization algorithm is used to adjust the learning rate for each parameter separately, which improves the speed and stability of model training and avoids the problem of local optimal solution. The use of the cross entropy loss function accurately measures the difference between the model output and the true label, and guides the model to update parameters in the right direction. The early stopping mechanism is set to effectively prevent overfitting and ensure the generalization ability of the model on unknown data. Through these training strategies, the CNN model can gradually improve its performance and ultimately achieve the goal of efficient and accurate processing of optical fiber signals. This not only optimizes the model parameters, but also ensures the practical application value of the model by monitoring the performance of the validation set. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0046] Reference Figure 1 ,

[0047] Example:

[0048] A method for processing optical fiber signals based on deep learning optimization, the method comprising the following steps:

[0049] S1: Data collection and preprocessing, collect a large amount of raw signal data from the optical fiber transmission system, including normal signals and abnormal signals. Perform denoising and normalization operations on the data to ensure data quality; divide the data set into training set, validation set and test set, with a ratio of 70%, 15%, and 15%;

[0050] S2: Use Fourier transform to extract the frequency domain features of the signal as one of the input features of deep learning, and design a convolutional neural network (CNN) structure to learn the spatial features of the signal;

[0051] S3: Convolutional neural network (CNN) is selected as the deep learning model, which is suitable for processing signal data with spatial correlation;

[0052] S4: Model training: Use the training set data to train the CNN model, use the Adam optimization algorithm, and set the learning rate to 0.001; use the cross entropy loss function to measure the difference between the model output and the true label; set an early stopping mechanism to stop training when the performance of the validation set no longer improves to avoid overfitting;

[0053] S5: Use the test set data to evaluate the trained CNN model and calculate performance indicators such as accuracy, recall, and F1 score. According to the evaluation results, adjust the model hyperparameters, such as learning rate and batch size, and consider using data enhancement techniques, such as signal time shift and amplitude scaling, to improve the generalization ability of the model.

[0054] S6: Use ensemble learning methods, such as training multiple CNN models and performing voting decisions, to improve overall performance;

[0055] S7: Deploy the trained CNN model to the real-time signal processing system to ensure that the model can efficiently process real-time data and meet real-time requirements, such as using GPU to accelerate computing;

[0056] S8: Monitor the performance of the CNN model in real time during the processing, record indicators such as processing delay and accuracy, and collect real-time feedback data for continuous optimization and updating of the model;

[0057] S9: Apply this method to an actual fiber-optic communication system to verify its effectiveness and stability, and expand the method according to application requirements, after which the entire fiber-optic signal processing process based on deep learning optimization can be completed.

[0058] In the step S1, firstly, according to the characteristics of the optical fiber signal and the actual needs, the appropriate sampling rate or downsampling rate is determined, and the decision is made by analyzing the frequency components and the rate of change of the signal to ensure that important information is not lost. Then, a representative sample is selected from the original data by using methods such as random sampling or stratified sampling. Random sampling is suitable for the case where the data distribution is relatively uniform. Stratified sampling divides the data into several layers according to certain characteristics of the signal, and then randomly extracts samples from each layer, which is suitable for the case where the data distribution is uneven. Subsequently, the extracted samples are further processed, including denoising, normalization, etc., to ensure the data quality. Denoising can be performed by wavelet transform, filter and other methods, and normalization can scale the data to the same scale for model processing. Finally, the processed samples are divided into training set, validation set and test set to ensure that the data distribution in each set is similar, so as to facilitate model training and evaluation. Through such a data sampling or downsampling process, the computational burden can be effectively reduced while ensuring the representativeness of the data, and an efficient and high-quality data foundation can be provided for the subsequent deep learning model training.

[0059] In step S2, the convolution layer uses multiple convolution kernels to extract local features, and the pooling layer reduces the dimension and abstracts the features. Through the stacking of multiple convolution layers and pooling layers, the model can automatically learn the multi-level spatial features of the signal. Subsequently, a fully connected layer is added to the end of the CNN to combine and abstract the learned features, and the activation function of the final output layer outputs the results according to the task requirements (such as classification). Throughout the design process, it is necessary to consider parameters such as the size and number of convolution kernels, the choice of activation functions, and the depth and width of the model to ensure that the model can effectively learn and extract signal features.

[0060] In step S3, the convolutional neural network (CNN) deep learning model includes a CNN architecture with two convolutional layers, two pooling layers, and one fully connected layer. Each component has its own specific calculation formula and parameter definition, which together form the mathematical basis of the CNN model, which is used to extract features from input signals and perform classification or regression tasks;

[0061] in:

[0062] The input layer receives the preprocessed signal data, usually in the form of a two-dimensional matrix. For example, time series data can be converted into an image.

[0063] Convolution layer 1 uses 32 3x3 convolution kernels to perform convolution operations to extract local features.

[0064] The calculation formula is: (W*X+b); where W is the convolution kernel weight, X is the input data, and b is the bias term; the number of convolution kernels is 32, the convolution kernel size is 3x3, the activation function is ReLU, and the formula is f(x)=\max(0,x)f(x)=max(0,x)

[0065] Pooling layer 1 uses 2x2 maximum pooling for downsampling to reduce the data dimension. The calculation formula is: max(X i,j ,X i+1,j ,X i,j+1 ,X i+1,j+1 ), where X i,j Represents the pixel value in the pooled area; the pooling kernel size is 2x2, and the pooling type is maximum pooling;

[0066] Convolutional layer 2 uses 64 3x3 convolution kernels to perform convolution operations to further extract more complex features. The calculation formula is the same as that of convolutional layer 1. The number of convolution kernels is 64, the convolution kernel size is 3x3, and the activation function is ReLU;

[0067] Pooling layer 2 uses 2x2 maximum pooling for downsampling. The calculation formula is the same as that of pooling layer 1. The parameter definition is the same as that of pooling layer 1;

[0068] The fully connected layer flattens the feature map after convolution and pooling, and performs feature fusion through the fully connected layer. The calculation formula is: W T X+b, where W is the weight matrix of the fully connected layer, X is the flattened feature vector, b is the bias term, and the number of neurons is 128;

[0069] The output layer uses the softmax activation function to output the category probability according to the task requirements. The calculation formula is: Among them, zi is the i-th element of the fully connected layer output, and the number of output categories is determined by the specific task. Multi-classification is the number of categories.

[0070] In step S4, the pre-processed training set data is input into the CNN model during the training process using the training set data, the output is calculated by forward propagation, and then the model parameters are updated by back propagation.

[0071] Adam is an adaptive learning rate optimization algorithm that can adjust the learning rate for each parameter separately. The calculation formula includes:

[0072] m t =β 1 m t-1 +(1-β 1 ) t ;

[0073]

[0074]

[0075] Among them, m t and v t are respectively the momentum and variance at time t, g t is the gradient, θ t is a parameter, α is the learning rate, β 1 and β 2 is a hyperparameter, a small constant to avoid division by zero;

[0076] The cross entropy loss function is used to measure the difference between the model output and the true label. The calculation formula is:

[0077]

[0078] Among them, p(xi) is the probability distribution of the true label, and q(xi) is the probability distribution predicted by the model;

[0079] The monitoring indicator is the loss or accuracy of the validation set. The stopping condition is that the monitoring indicator does not improve for multiple consecutive epochs (for example, 10). Patience is the number of epochs tolerated, and training is stopped if there is no improvement after this number.

[0080] In step S5, the test set data is data that the model has never seen before and can objectively reflect the performance of the model. Then calculate the various performance indicators of the model, such as accuracy, recall, F1 score, etc. These indicators can comprehensively evaluate the performance of the model on the classification task. Then analyze the error cases of the model to find out the shortcomings of the model. Through error analysis, the model can be improved in a targeted manner to improve its performance. Finally, according to the evaluation results, adjust the model parameters or structure for further optimization. Model evaluation is a continuous process that requires continuous iteration and improvement.

[0081] In step S6, a reinforcement learning mechanism is introduced to enable the model to adjust parameters according to feedback in real-time processing to achieve self-optimization. This includes defining a reward function that evaluates the performance of the model based on indicators such as accuracy and real-time performance of signal processing. When processing signals, the model adjusts its internal parameters, such as CNN weights and biases, based on feedback from the reward function to maximize long-term cumulative rewards. This reinforcement learning mechanism enables the model to adaptively optimize its performance in a constantly changing signal environment.

[0082] In step S7, the input interface needs to convert the real-time signal into the format required by the model, and the output interface needs to convert the output result of the model into a format available to the system. Subsequently, the model deployment is tested and verified to ensure the stability and reliability of the model in the actual environment, which includes testing the performance indicators such as the processing speed and accuracy of the model, and verifying the performance of the model in different scenarios. Finally, based on the test and verification results, the model deployment is optimized and adjusted, which may involve the allocation of hardware resources, optimization of software architecture, etc., to ensure the efficiency and stability of the model in actual applications.

[0083] In step S8, a real-time data acquisition system is designed, which is responsible for continuously monitoring the results of optical fiber signal processing and system performance indicators, such as signal quality, processing delay, bit error rate, etc. These real-time feedback data are transmitted to the data storage and analysis platform. Then, the model is learned online or periodically retrained using these collected real-time feedback data. Online learning means that the model is fine-tuned according to the real-time feedback data without interrupting the service to quickly adapt to new signal characteristics or environmental changes. This can be achieved through incremental learning, online gradient descent and other methods. Periodic retraining is to use the accumulated feedback data to comprehensively retrain the model within a certain time interval to update the weights and parameters of the model and improve its long-term performance.

[0084] In step S9, a long-term stability test is performed to observe the performance changes of the model under long-term operation. This includes monitoring the processing delay, resource consumption, failure rate and other indicators of the model to ensure the stability and reliability of the model in practical applications. During the verification process, detailed performance data and analysis results are collected to evaluate whether the model meets the predetermined performance indicators. If the model performance does not meet expectations, it is necessary to go back to the previous steps, such as data preprocessing, model training, etc., for targeted optimization and adjustment.

[0085] In the present invention, the CNN architecture provides powerful feature extraction and representation capabilities for optical fiber signal processing. Through the combination of two convolutional layers and two pooling layers, the model can extract features from simple to complex from the original signal layer by layer. The convolution operation uses the characteristics of local connection and value sharing to effectively capture the spatial correlation in the signal, while the pooling operation reduces the data dimension by downsampling while increasing the model's robustness to position changes. This hierarchical feature extraction process enables the model to learn the essential characteristics of the signal, laying a solid foundation for subsequent classification or regression tasks. In addition, the fully connected layer flattens the feature map after convolution and pooling and performs feature fusion, further enhancing the model's ability to recognize complex signal patterns.

[0086] In the present invention, the Adam optimization algorithm is used to adjust the learning rate for each parameter separately, which improves the speed and stability of model training and avoids the problem of local optimal solution. The use of the cross entropy loss function accurately measures the difference between the model output and the true label, and guides the model to update parameters in the right direction. The early stopping mechanism is set to effectively prevent overfitting and ensure the generalization ability of the model on unknown data. Through these training strategies, the CNN model can gradually improve its performance and ultimately achieve the goal of efficient and accurate processing of optical fiber signals. Thereby, not only the model parameters are optimized, but also the practical application value of the model is ensured by monitoring the performance of the validation set.

[0087] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0088] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optical fiber signal processing based on deep learning optimization, characterized in that: The method comprises the following steps: S1: Data acquisition and preprocessing: collect a large amount of raw signal data from the optical fiber transmission system, including normal signals and abnormal signals; perform denoising and normalization operations on the data to ensure data quality; divide the data set into training set, validation set and test set, with a ratio of 70%, 15% and 15%; S2: Use Fourier transform to extract the frequency domain features of the signal as one of the input features of deep learning, and design a convolutional neural network (CNN) structure to learn the spatial features of the signal; S3: Convolutional neural network (CNN) is selected as the deep learning model, which is suitable for processing signal data with spatial correlation; S4: Model training: Use the training set data to train the CNN model, use the Adam optimization algorithm, and set the learning rate to 0.001; use the cross entropy loss function to measure the difference between the model output and the true label; set an early stopping mechanism to stop training when the performance of the validation set no longer improves to avoid overfitting; S5: Use the test set data to evaluate the trained CNN model, calculate the performance indicators, adjust the model hyperparameters based on the evaluation results, and consider using data enhancement techniques to improve the generalization ability of the model; S6: Use ensemble learning methods to improve overall performance; S7: Deploy the trained CNN model to the real-time signal processing system to ensure that the model can efficiently process real-time data and meet real-time requirements; S8: Monitor the performance of the CNN model in real time during processing, record processing delay and accuracy indicators, and collect real-time feedback data for continuous optimization and updating of the model; S9: Apply this method to an actual fiber-optic communication system to verify its effectiveness and stability, and expand the method according to application requirements, after which the entire fiber-optic signal processing process based on deep learning optimization can be completed.

2. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In the step S1, firstly, according to the characteristics of the optical fiber signal and actual needs, a suitable sampling rate or downsampling rate is determined by analyzing the frequency components and the rate of change of the signal to ensure that important information is not lost; then, a representative sample is selected from the original data by random sampling or stratified sampling method. Random sampling is suitable for the case where the data distribution is relatively uniform, while stratified sampling divides the data into several layers according to certain characteristics of the signal, and then randomly extracts samples from each layer, which is suitable for the case where the data distribution is uneven; then, the extracted samples are further processed, including denoising and normalization, to ensure data quality. Denoising uses wavelet transform and filter method, and normalization scales the data to the same scale for model processing; finally, the processed samples are divided into training set, validation set and test set to ensure that the data distribution in each set is similar, so as to facilitate model training and evaluation.

3. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In step S2, the convolution layer uses multiple convolution kernels to extract local features, and the pooling layer reduces the dimension and abstracts the features; by stacking multiple convolution layers and pooling layers, the model can automatically learn the multi-level spatial features of the signal; Then, a fully connected layer is added at the end of the CNN to combine and abstract the learned features, and the activation function of the final output layer outputs the results according to the task requirements; Throughout the design process, it is necessary to consider the size and number of convolution kernels, the selection parameters of the activation function, and the depth and width of the model to ensure that the model can effectively learn and extract signal features.

4. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In step S3, the convolutional neural network (CNN) deep learning model includes a CNN architecture of two convolutional layers, two pooling layers and one fully connected layer; each component has its own specific calculation formula and parameter definition, which together constitute the mathematical basis of the CNN model, and are used to extract features from input signals and perform classification or regression tasks; in: The input layer receives the preprocessed signal data in the form of a two-dimensional matrix, including the conversion of time series data into image form; Convolution layer 1 uses 32 3x3 convolution kernels to perform convolution operations to extract local features. The calculation formula is: (W*X+b); where W is the convolution kernel weight, X is the input data, and b is the bias term; the number of convolution kernels is 32, the convolution kernel size is 3x3, the activation function is ReLU, and the formula is f(x)=\max(0,x)f(x)=max(0,x) Pooling layer 1 uses 2x2 maximum pooling for downsampling to reduce the data dimension. The calculation formula is: max(X i,j ,X i+1,j ,X i,j+1 ,X i+1,j+1 ), where X i,j Represents the pixel value in the pooled area; the pooling kernel size is 2x2, and the pooling type is maximum pooling; Convolutional layer 2 uses 64 3x3 convolution kernels for convolution operations to further extract more complex features. The calculation formula is the same as that of convolutional layer 1; the number of convolution kernels is 64, the convolution kernel size is 3x3, and the activation function is ReLU; Pooling layer 2 uses 2x2 maximum pooling for downsampling; the calculation formula is the same as pooling layer 1; the parameter definition is the same as pooling layer 1; The fully connected layer flattens the feature map after convolution and pooling, and performs feature fusion through the fully connected layer. The calculation formula is: W T X+b, where W is the weight matrix of the fully connected layer, X is the flattened feature vector, b is the bias term, and the number of neurons is 128; The output layer uses the softmax activation function to output the category probability according to the task requirements. The calculation formula is: Among them, zi is the i-th element of the fully connected layer output, and the number of output categories is determined by the specific task. For multiple classifications, it is the number of categories.

5. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In step S4, the pre-processed training set data is input into the CNN model during the training process using the training set data, the output is calculated by forward propagation, and then the model parameters are updated by back propagation. Adam is an adaptive learning rate optimization algorithm that can adjust the learning rate for each parameter separately. The calculation formula includes: m t =β1m t-1 +(1-β1)g t ; Among them, m t and v t are respectively the momentum and variance at time t, g t is the gradient, θ t is a parameter, α is the learning rate, β1 and β2 are hyperparameters, which are small constants to avoid division by zero; The cross entropy loss function is used to measure the difference between the model output and the true label. The calculation formula is: Among them, p(xi) is the probability distribution of the true label, and q(xi) is the probability distribution predicted by the model; The monitoring indicator is the loss or accuracy of the validation set; the stopping condition is that the monitoring indicator has not improved for multiple consecutive epochs; patience is the number of epochs tolerated, and training is stopped if there is no improvement after this number.

6. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In step S5, the test set data is data that the model has never seen before and can objectively reflect the performance of the model; then various performance indicators of the model are calculated, including accuracy, recall rate, and F1 score, which can comprehensively evaluate the performance of the model in the classification task; Then analyze the model's error cases to find out the model's shortcomings. Through error analysis, improve the model in a targeted manner to improve its performance. Finally, based on the evaluation results, adjust the model parameters or structure for further optimization. Model evaluation is an ongoing process that requires continuous iteration and improvement.

7. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In step S6, a reinforcement learning mechanism is introduced to enable the model to adjust parameters according to feedback in real-time processing to achieve self-optimization; this includes defining a reward function that evaluates the performance of the model based on the accuracy and real-time indicators of signal processing; when processing signals, the model adjusts its internal parameters, including the weights and biases of CNN, according to the feedback of the reward function to maximize the long-term cumulative rewards.

8. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In step S7, the input interface needs to convert the real-time signal into the format required by the model, and the output interface needs to convert the output result of the model into a format available to the system; then, the model deployment is tested and verified to ensure the stability and reliability of the model in the actual environment, which includes testing the processing speed and accuracy performance indicators of the model, and verifying the performance of the model in different scenarios; finally, the model deployment is optimized and adjusted according to the test and verification results.

9. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In step S8, a real-time data acquisition system is designed, which is responsible for continuously monitoring the results of optical fiber signal processing and system performance indicators, including signal quality, processing delay, and bit error rate; These real-time feedback data are transmitted to the data storage and analysis platform. Then, the model is learned online or retrained periodically using the collected real-time feedback data. Online learning means that the model is fine-tuned based on real-time feedback data without interrupting service to quickly adapt to new signal characteristics or environmental changes.

10. The optical fiber signal processing method based on deep learning optimization according to claim 1, characterized in that: In step S9, a long-term stability test is performed to observe the performance changes of the model under long-term operation; this includes monitoring the processing delay, resource consumption, and failure rate indicators of the model to ensure the stability and reliability of the model in practical applications; During the verification process, detailed performance data and analysis results are collected to evaluate whether the model meets the predetermined performance indicators; if the model performance does not meet expectations, it is necessary to go back to the previous steps, such as data preprocessing and model training, to perform targeted optimization and adjustments.

Images