A method and system for synergistically regulating pump and polarization states to achieve multi-spectral switching

Abstract

The application discloses a kind of method and system for realizing multispectral switching by synergic regulation pump and polarization state, belong to ultrafast fiber laser and automatic control technical field.It includes: the output signal of ultrafast fiber laser is input residual neural network and obtains the set solution of spectrum corresponding voltage and pump power;Optimal set solution corresponding to different spectral state is searched using adaptive gradient optimization algorithm;According to the search result, feedback is given to electric control polarization controller and pump, and the automatic switching of spectral state is realized;If the disappearance of spectral state is monitored, perform adaptive recovery algorithm, and randomly extract in set solution to recover;If recovery fails, search through random search algorithm until recovery is successful.The application solves the problem that passive mode-locked fiber laser based on nonlinear polarization rotation cannot quickly switch pulse state and is difficult to stabilize, while realizing full automation control of fiber laser, improving the generality of the algorithm, and the structure is simple and easy to realize.

Description

Technical Field

[0001] This invention relates to the field of ultrafast fiber lasers and automatic control technology, and in particular to a method and system for achieving multispectral switching by synergistically controlling pump and polarization states. Background Technology

[0002] Fiber lasers are widely used in thermal imaging, spectral analysis, optical communication, precision sensing, biomedicine, materials processing, and environmental monitoring due to their small size, low cost, excellent heat dissipation, and high energy conversion efficiency. In practical applications, fiber lasers often switch between different operating states, such as continuous light, Q-switching, and mode-locking, to meet different requirements for energy, pulse width, spectrum, and repetition frequency.

[0003] Existing multi-state switching typically relies on manual adjustment of intracavity parameters, such as pump power, polarization state, and saturable absorber / polarization element settings. These parameters are strongly coupled and sensitive to environmental drift, resulting in low switching efficiency, poor repeatability, and high dependence on operator experience. While some automated solutions can quickly acquire a single target state, they struggle to simultaneously achieve rapid and repeatable switching across multiple states. Furthermore, most solutions only adjust one type of parameter, lacking coordinated optimization of pump power and polarization state, leading to insufficient stability in complex cavity configurations or during long-term operation.

[0004] Therefore, there is an urgent need for a method and system that can jointly regulate pump power and intracavity polarization state, and combine intelligent identification and online optimization for rapid switching of multispectral states, while still possessing high responsiveness, strong robustness and full automation under environmental disturbances and device aging conditions. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for achieving multispectral switching by synergistically controlling pump and polarization states. This method can solve the polarization control problem, incomplete automation, and inability to quickly switch the spectral state of passive mode-locked lasers.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0007] In a first aspect, the present invention provides a method for achieving multispectral switching by synergistically controlling pumping and polarization states, comprising:

[0008] Based on the spectral images under different electric polarization controller voltages and pump powers, a neural network is trained to obtain a recognition model. The input of the recognition model is the spectral image, and the output is multiple feasible set solutions corresponding to the pump power and the voltage of each channel of the electric polarization controller, and the corresponding spectral state labels.

[0009] The target spectral image is input into the recognition model to obtain the set solution and spectral state. An adaptive gradient optimization algorithm is used to search for the optimal combination of pump power and voltage of each channel of the electric polarization controller corresponding to the target spectral state in the set solution.

[0010] Based on the search results, a control signal is generated and fed back to the electronically controlled polarization controller and pump to perform fully automatic spectral state switching; the spectral state is switched between continuous light, Q-switched light, and fundamental frequency mode-locked light.

[0011] After the spectral state is successfully switched, the spectral monitoring mode is entered to identify the spectral state in real time and monitor whether the spectral state is lost.

[0012] If a loss of spectral state is detected, an adaptive recovery algorithm is executed to restore the spectral state to its original state. If the recovery of the spectral state is successful, the system returns to the spectral monitoring mode. If the recovery fails, a random search algorithm is executed until the spectral state is restored and the system returns to the spectral monitoring mode.

[0013] Preferably, the neural network is a residual neural network.

[0014] Preferably, the step of training a neural network to obtain a recognition model based on spectral images under different electronically controlled polarization controller voltages and pump powers includes:

[0015] The laser's output spectral images are collected in real time under different electronically controlled polarization controller voltages and pump powers using a spectral analyzer.

[0016] The acquired spectral image set is binarized and median filtered;

[0017] The processed spectral image set is divided into a training set, a test set, and a validation set.

[0018] The residual neural network structure consists of: an input layer, a convolutional layer, a batch normalization layer, a max pooling layer, four residual modules, an adaptive average pooling layer, and an output layer.

[0019] The constructed residual neural network is trained using the training set until convergence to obtain the recognition model. During the training process, gradient descent is used for optimization, and the loss function is calculated as follows: , in, For loss function, , Indicates an index. This represents the total number of categories, where each category refers to a spectral state. Indicates the first The true label of the class, Indicates the first Predicted probability of class and These are the weighting coefficients. This represents the total number of samples in the training set. and They represent the first The actual pump power value and actual channel voltage value of each sample. and These represent the target pump power and the target channel voltage, respectively.

[0020] Preferably, the adaptive gradient optimization algorithm uses the Adam algorithm.

[0021] Preferably, the step of using an adaptive gradient optimization algorithm to search for the optimal combination of pump power and voltage of each channel of the electrically controlled polarization controller corresponding to the target spectral state in the ensemble solution includes:

[0022] Let the initialization set be R = [V1, V2, V3, V4, P], and the initial momentum be... and variance Where V1, V2, V3, and V4 represent the four-channel voltages of the electric polarization controller, and P represents the pump power.

[0023] The momentum and variance for each iteration are calculated as follows:

[0024] ,

[0025] ,

[0026] in, For the number of iterations, For the first Momentum of the next iteration For the first Momentum of the next iteration , All are coefficients. For the first The variance of the next iteration For the first The variance of the next iteration Indicates the first In the next iteration, the loss function For the current set of solutions The gradient;

[0027] Based on the momentum and variance after iteration, update the ensemble solution as follows:

[0028] ,

[0029] in, For the first The solution set obtained after the nth iteration For the first The solution set at the next iteration For learning rate, It is a constant;

[0030] Set threshold and number of iterations ,like Or the number of iterations exceeds Stop optimization and obtain the optimal pump power and voltage combination of each channel of the electronically controlled polarization controller.

[0031] Preferably, if spectral state loss is detected, the initial values ​​of the pump power adjustment step size and the single-channel voltage adjustment step size of the electric polarization controller are reset to 1 and 0.1V, respectively, and the threshold is adjusted based on the Adam algorithm. and number of iterations The algorithm was modified to re-employ the Adam algorithm to search for the optimal combination of pump power and voltage of each channel of the electronically controlled polarization controller corresponding to the target spectral state.

[0032] Preferably, the recovery failure refers to,

[0033] If the cumulative adjustment step size of the pump power is greater than the maximum value of the pump power adjustment step size The cumulative voltage adjustment step size of the single channel of the electronically controlled polarization controller is greater than the maximum value. If so, the recovery will fail.

[0034] Preferably, the implementation process of the random search algorithm is as follows:

[0035] The initial conditions for random search are set, including setting the initial voltage values ​​of all four channels of the electronically controlled polarization controller to 0V, the step size to 0.1V, the pump power to the intermediate power value P0, the maximum number of searches to M, and a similarity threshold. ;

[0036] Within the parameter space, uniform random sampling is performed on the set solution R0=[V1, V2, V3, V4, P0] within a finite range according to a set step size, with the number of samples per round being Ns;

[0037] The sampled set solution R k The input is an electronically controlled polarization controller and a pumping system, and the laser output spectrum image is acquired by a spectrometer;

[0038] The recognition model built using a residual neural network performs state recognition on each spectral image, compares the results with the target spectral state, and calculates the similarity.

[0039] If the similarity calculated from the set solution is greater than the similarity threshold in the current sampling round... If the number of searches exceeds the maximum number of searches M, the search will end.

[0040] Secondly, the present invention provides a system for achieving multispectral switching by synergistically controlling pump and polarization state, for realizing the above-mentioned method for achieving multispectral switching by synergistically controlling pump and polarization state, the system comprising:

[0041] The model training module is used to train a neural network based on spectral images under different electric polarization controller voltages and pump powers to obtain a recognition model. The input of the recognition model is the spectral image, and the output is multiple feasible set solutions for the corresponding pump power and voltage combinations of each channel of the electric polarization controller, as well as the corresponding spectral state labels.

[0042] The optimization module is used to input the target spectral image into the recognition model to obtain the set solution and spectral state, and to use an adaptive gradient optimization algorithm to search for the optimal combination of pump power and voltage of each channel of the electric polarization controller corresponding to the target spectral state in the set solution.

[0043] The switching module is used to generate control signals based on the search results of the optimization module, and feed them back to the electronically controlled polarization controller and the pump to perform fully automatic spectral state switching; the spectral state is switched between continuous light, Q-switched and fundamental frequency mode-locked.

[0044] The monitoring module is used to enter the spectral monitoring mode after the spectral state switch is successful, and to identify the spectral state in real time and monitor whether the spectral state is lost.

[0045] The recovery module is used to, based on the monitoring results of the monitoring module, if the spectral state is detected to be lost, execute an adaptive recovery algorithm to restore the spectral state to its original state; if the spectral state is successfully restored, return to the spectral monitoring mode; if the restoration fails, execute a random search algorithm until the spectral state is restored and return to the spectral monitoring mode.

[0046] By employing the above-mentioned technical means, the beneficial effects achieved by the present invention are as follows:

[0047] This invention, based on residual neural network learning, adaptive optimization algorithm, and random search algorithm, achieves fully automated and rapid switching of spectral states in passive mode-locked fiber lasers with nonlinear polarization control rotation, and also realizes image recognition of the laser spectrum. Attached Figure Description

[0048] Figure 1 This is a flowchart of a method for achieving multispectral switching by synergistic regulation of pumping and polarization state, provided in a preferred embodiment of the present invention.

[0049] Figure 2This is a schematic diagram of a residual neural network architecture provided in a preferred embodiment of the present invention;

[0050] Figure 3 This is a flowchart of the Adam optimization algorithm provided in a preferred embodiment of the present invention;

[0051] Figure 4 This is a flowchart of a random search algorithm provided in a preferred embodiment of the present invention;

[0052] Figure 5 This is a schematic diagram of the fundamental frequency mode-locked spectrum results provided in a preferred embodiment of the present invention;

[0053] Figure 6 This is a schematic diagram of the fundamental frequency mode-locked pulse width result provided in a preferred embodiment of the present invention;

[0054] Figure 7 This is a schematic diagram of continuous spectrum results provided by a preferred embodiment of the present invention;

[0055] Figure 8 This is a schematic diagram of Q-switched spectral results provided by a preferred embodiment of the present invention; Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0057] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0058] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0059] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0060] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0061] It should be emphasized here that the step markers mentioned below are not a limitation on the order of the steps, but should be understood as meaning that the steps can be executed in the order mentioned in the embodiments, or in a different order than in the embodiments, or several steps can be executed simultaneously.

[0062] This invention provides a method for synergistically controlling pump and polarization states to achieve multispectral switching. It employs a combination of an electrically controlled polarization controller and a residual neural network to achieve fully automatic identification and rapid control of the spectral state of a fiber laser. This invention acquires the spectral image of the laser output and inputs it into a trained residual neural network model for identification. This allows for accurate determination of the current spectral state. Furthermore, an adaptive gradient optimization algorithm generates corresponding control parameters, which drive the electrically controlled polarization controller and pump system to perform real-time adjustments. This enables rapid switching between mode-locking, Q-switching, and continuous wave states, allowing the laser to operate stably in the target state.

[0063] Based on the above inventive concept, this invention provides a method for achieving multispectral switching by synergistically controlling pumping and polarization states, see [link to relevant documentation]. Figure 1 The specific implementation process is as follows:

[0064] Step 1: The spectral images output by the laser under different voltages and pump powers of the electric polarization controller are acquired in real time by a spectral analyzer and transmitted to the computing center via a LAN interface. The spectral images are processed to form a dataset. A recognition model is obtained after training based on a residual neural network. The input of the recognition model is the spectral image output by the laser, and the output is multiple feasible set solutions for the corresponding pump power and voltage combinations of each channel of the electric polarization controller and the corresponding spectral state labels.

[0065] Step 2: Input the target spectral image into the above recognition model to obtain the set solution. The set solution includes all possible combinations of pump power and voltage of each channel of the electronic polarization controller that can achieve the target spectrum. Use an adaptive gradient optimization algorithm to search for the optimal combination of pump power and voltage of each channel of the electronic polarization controller corresponding to the target spectral state in the set solution.

[0066] Step 3: Generate control signals based on search results and feed them back to the electronically controlled polarization controller and pump to achieve fully automatic spectral state switching; the switchable spectral states include continuous light, Q-switching, and fundamental mode-locking.

[0067] Step 4: If the ideal state is achieved, then enter the spectral monitoring mode. The spectral monitoring mode continuously scans the current spectrum and uses the recognition model of the residual neural network to identify the spectral state, thereby monitoring whether the spectral state is lost.

[0068] Step 5: If the loss of spectral state is detected, execute the adaptive recovery algorithm to see if the original state can be recovered;

[0069] Step 6: If the spectral state is successfully restored, return to the spectral monitoring mode; otherwise, execute the random search algorithm until the spectral state is restored and return to the spectral monitoring mode.

[0070] Traditional neural networks work by passing the output data of each layer to the next, while residual neural networks use fast connection techniques to directly pass input data to the output data. They do not directly study the correspondence between input and output, but rather focus on learning the so-called "residual" between input and output, improving the model's speed and accuracy. For example... Figure 2 As shown, the architecture of the residual neural network consists of an input layer, a convolutional layer, a batch normalization layer, a max pooling layer, four residual modules, and an adaptive average pooling layer; this is used to obtain the correspondence between angle and spectrum, enabling the model to perform spectral image recognition.

[0071] Training a recognition model based on a residual neural network specifically includes the following steps:

[0072] Step 1.1: Binarize and median filter the acquired spectral image set. The specific formula for the binarization algorithm is as follows:

[0073] ,

[0074] and Typically represents the pixel bits in an image. This is the binarized image. These are the pixel values ​​of the original image. It is a threshold.

[0075] The specific formula for the median filtering algorithm is as follows:

[0076] ,

[0077] It is a threshold. These are the pixel values ​​of the image after median filtering. This indicates that the median value of the neighboring pixels is taken. and It is the offset relative to the coordinates, and its value ranges from [-]. , ], Indicates at point Pixels are selected within the neighborhood of , and the size of the neighborhood is determined by the parameter. Decide.

[0078] Step 1.2: Based on the processed spectral image set, divide it into a training set, a test set, and a validation set, with a division ratio of 8:1:1.

[0079] Step 1.3: Define the residual neural network structure: input layer, convolutional layer, batch normalization layer, max pooling layer, four residual modules, adaptive average pooling layer, and output layer;

[0080] Step 1.4: Set the loss function and optimizer. The loss function is set according to the following formula for the spectral features:

[0081] ,

[0082] For loss function, , Indicates an index. This represents the total number of categories, where each category refers to a spectral state. Indicates the first The true label of the class, Indicates the first Predicted probability of class and These are the weighting coefficients. The total number of samples, and They represent the first The actual pump power value and actual channel voltage value of each sample. and These represent the target pump power and the target channel voltage, respectively.

[0083] The optimizer uses the gradient descent method, and the specific formula is as follows:

[0084] ,

[0085] in and This represents the four-channel voltage and pump power of the electric polarization controller. and These represent the loss function and the learning rate, which determines the rate of descent, respectively.

[0086] Step 1.5: Train the residual neural network using the training set. In each training epoch, calculate the training loss until the loss value converges, then stop training to obtain the recognition model.

[0087] Step 1.6: Evaluate the trained recognition model using the test set and validation set.

[0088] To address the challenge of rapidly searching for spectral states, a preferred embodiment of this invention proposes an adaptive gradient optimization algorithm, employing the Adam algorithm (see [link]). Figure 3 The specific steps are as follows:

[0089] Step 2.1: Based on the set solution obtained from the recognition model, initialize the set solution as R=[V1, V2, V3, V4, P], and determine the momentum of each parameter. and variance Where V1, V2, V3, and V4 represent the four-channel voltages of the electric polarization controller, and P represents the pump power.

[0090] Step 2.2: Calculate the momentum and variance for each iteration. The momentum formula is as follows:

[0091] ,

[0092] variable For the number of iterations, For the first Momentum of the next iteration For the first Momentum of the next iteration For coefficients, Indicates the first In the next iteration, the loss function For the current set of solutions The gradient.

[0093] The loss function is a weighted combination of the cross-entropy loss function and the parameter bias penalty term.

[0094] The specific formula for variance is as follows:

[0095] ,

[0096] For the first The variance of the next iteration For the first The variance of the next iteration, variables For coefficients;

[0097] Step 2.3: Update the ensemble solution based on the momentum and variance after iteration, using the following formula:

[0098] ,

[0099] For the first The set obtained after the iteration is unified. For the first The set solution at the next iteration, variables For learning rate, It is a constant.

[0100] Step 2.4: Set the threshold and number of iterations ,like Or the number of iterations exceeds Stop optimization.

[0101] After the algorithm iteration is completed, the final optimization result is input into the electrically controlled polarization controller and the pump.

[0102] In step 4 of this invention, if a loss of spectral state is detected, the initial values ​​of the pump power adjustment step size and the four-channel voltage adjustment step size of the electric polarization controller are reset to 1 and 0.1V respectively, and the adaptive recovery algorithm is executed.

[0103] It should be noted that the adaptive recovery algorithm is based on the Adam algorithm with adjustments to the threshold. and number of iterations Make modifications, perform a state search within this range, and restore the spectral state.

[0104] Furthermore, the step size adjustment range of the adaptive recovery algorithm is set manually.

[0105] In step 5 of this invention, if the cumulative adjustment step size is greater than the maximum value of the pump power adjustment step size... The maximum value of the single-channel voltage adjustment step size of the electronically controlled polarization controller If the error is not found, the recovery is considered to have failed, and a random search algorithm is executed.

[0106] To address the sensitivity of light polarization state in optical fibers to environmental influences, a preferred embodiment of this invention proposes a random search algorithm, the process of which is as follows: Figure 4 As shown. The specific steps are as follows:

[0107] Step 4.1: Set the initial conditions for random search. Set the initial voltage values ​​of all four channels of the electrically controlled polarization controller to 0V, the step size to 0.1V, the pump power to the intermediate power value P0, the maximum number of searches to M, and the similarity threshold. .

[0108] Step 4.2: Within the parameter space, perform uniform random sampling within a finite range of R0=[V1, V2, V3, V4, P0] according to the set step size, with the number of samples per round being Ns;

[0109] Step 4.3: Calculate the solution R from the sampled set. k The input is an electronically controlled polarization controller and a pumping system, and the laser output spectrum is acquired by a spectrometer.

[0110] Step 4.4: The recognition model built using the residual neural network performs state recognition on each spectral image, compares the results with the target spectral state, and calculates the similarity. The similarity metric uses cosine similarity, and the specific formula is as follows:

[0111] ;

[0112] in and It is represented as the feature vector of the current spectrum and the target spectrum.

[0113] Step 4.5: If the set solution R is found in the current sampling round, the similarity is calculated. Greater than the similarity threshold If the number of searches exceeds the maximum search count M, the search ends.

[0114] Figure 5 This is a schematic diagram of the fundamental frequency mode-locked spectrum provided by a preferred embodiment of the present invention. The spectral center is located at approximately 1562 nm, the bandwidth is approximately 3.82 nm, and obvious Kelly sidebands are visible on both sides of the spectrum, indicating that the laser is in a stable fundamental frequency mode-locked state.

[0115] Figure 6 This represents the autocorrelation test results corresponding to the fundamental frequency mode-locked pulse. The solid line represents Sech. 2 The function fitting curve, with the hollow circle representing experimental measurement data, shows a pulse width of 1.293 ps, further validating the result. Figure 5 The mode-locked pulse characteristics corresponding to the spectrum shown.

[0116] Figure 7 This is a schematic diagram of a continuous light spectrum provided by a preferred embodiment of the present invention. The spectral lines show narrow band peaks, without obvious modulation and sideband structure, indicating that the laser is operating in a continuous light state.

[0117] Figure 8 This is a schematic diagram of the Q-switched spectral results provided by a preferred embodiment of the present invention. The appearance of periodic modulation peaks in the spectrum reflects a strong energy modulation effect within the cavity, indicating that the laser is in a Q-switched operating state.

[0118] Based on the above inventive concept, the present invention also provides a system for achieving multispectral switching by synergistically controlling pump and polarization state, for realizing the above-mentioned method for achieving multispectral switching by synergistically controlling pump and polarization state, the system comprising:

[0119] The model training module is used to train a neural network based on spectral images under different electric polarization controller voltages and pump powers to obtain a recognition model. The input of the recognition model is the spectral image, and the output is multiple feasible set solutions for the corresponding pump power and voltage combinations of each channel of the electric polarization controller, and the corresponding spectral state labels.

[0120] The optimization module is used to input the target spectral image into the recognition model to obtain the set solution and spectral state, and to use an adaptive gradient optimization algorithm to search for the optimal combination of pump power and voltage of each channel of the electric polarization controller corresponding to the target spectral state in the set solution.

[0121] The switching module is used to generate control signals based on the search results of the optimization module, and feed them back to the electronically controlled polarization controller and the pump to perform fully automatic spectral state switching; the spectral state is switched between continuous light, Q-switched and fundamental frequency mode-locked.

[0122] The monitoring module is used to enter the spectral monitoring mode after the spectral state switch is successful, and to identify the spectral state in real time and monitor whether the spectral state is lost.

[0123] The recovery module is used to, based on the monitoring results of the monitoring module, if the spectral state is detected to be lost, execute an adaptive recovery algorithm to restore the spectral state to its original state; if the spectral state is successfully restored, return to the spectral monitoring mode; if the restoration fails, execute a random search algorithm until the spectral state is restored and return to the spectral monitoring mode.

[0124] It is worth noting that this embodiment corresponds to the above method embodiment. The implementation methods of the above method embodiments are all applicable to this system embodiment and can achieve the same or similar technical effects, so they will not be described in detail here.

[0125] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for achieving multispectral switching by synergistically controlling pump and polarization states, characterized in that, include: Based on the spectral images under different electric polarization controller voltages and pump powers, a neural network is trained to obtain a recognition model. The input of the recognition model is the spectral image, and the output is multiple feasible set solutions for the corresponding pump power and voltage combinations of each channel of the electric polarization controller, as well as the corresponding spectral state labels. The trained neural network yields a recognition model, including: The laser's output spectral images are collected in real time using a spectral analyzer under different electronically controlled polarization controller voltages and pump powers. The acquired spectral image set is binarized and median filtered; The processed spectral image set is divided into a training set, a test set, and a validation set. The residual neural network structure consists of: an input layer, a convolutional layer, a batch normalization layer, a max pooling layer, four residual modules, an adaptive average pooling layer, and an output layer. The constructed residual neural network is trained using the training set until convergence to obtain the recognition model. During the training process, gradient descent is used for optimization, and the loss function is calculated as follows: ， in, For the loss function, , Indicates an index. This represents the total number of categories, where each category refers to a spectral state. Indicates the first The true label of the class, Indicates the first Predicted probability of class and These are the weighting coefficients. This represents the total number of samples in the training set. and They represent the first The actual pump power value and actual channel voltage value of each sample. and These represent the target pump power and the target channel voltage, respectively. The target spectral image is input into the recognition model to obtain the set solution and spectral state. An adaptive gradient optimization algorithm is used to search for the optimal combination of pump power and voltage of each channel of the electric polarization controller corresponding to the target spectral state in the set solution. Based on the search results, a control signal is generated and fed back to the electronically controlled polarization controller and pump to perform fully automatic spectral state switching; the spectral state is switched between continuous light, Q-switched light, and fundamental frequency mode-locked light. After the spectral state is successfully switched, the spectral monitoring mode is entered to identify the spectral state in real time and monitor whether the spectral state is lost. If a loss of spectral state is detected, an adaptive recovery algorithm is executed to restore the spectral state to its original state. If the recovery of the spectral state is successful, the system returns to the spectral monitoring mode. If the recovery fails, a random search algorithm is executed until the spectral state is restored and the system returns to the spectral monitoring mode.

2. The method for achieving multispectral switching by synergistic control of pump and polarization state according to claim 1, characterized in that, The neural network is a residual neural network.

3. The method for achieving multispectral switching by synergistic control of pump and polarization state according to claim 1, characterized in that, The adaptive gradient optimization algorithm uses the Adam algorithm.

4. The method for achieving multispectral switching by synergistic control of pump and polarization state according to claim 3, characterized in that, The adaptive gradient optimization algorithm is used to search for the optimal combination of pump power and voltage of each channel of the electrically controlled polarization controller corresponding to the target spectral state in the ensemble solution, including: Let the initialization set be R = [V1, V2, V3, V4, P], and the initial momentum be... and variance Where V1, V2, V3, and V4 represent the four-channel voltages of the electric polarization controller, and P represents the pump power. The momentum and variance for each iteration are calculated as follows: ， ， in, For the number of iterations, For the first Momentum of the next iteration For the first Momentum of the next iteration , All are coefficients. For the first The variance of the next iteration For the first The variance of the next iteration Indicates the first In the next iteration, the loss function For the current set of solutions The gradient; Based on the momentum and variance after iteration, update the ensemble solution as follows: ， in, For the first The solution set obtained after the nth iteration For the first The solution set at the next iteration For learning rate, It is a constant; Set threshold and number of iterations ,like Or the number of iterations exceeds Stop optimization and obtain the optimal pump power and voltage combination of each channel of the electronically controlled polarization controller.

5. The method for achieving multispectral switching by synergistic control of pump and polarization state according to claim 4, characterized in that, If spectral state loss is detected, the initial values ​​of the pump power adjustment step size and the single-channel voltage adjustment step size of the electric polarization controller are reset to 1 and 0.1V, respectively. Based on the Adam algorithm, the threshold is adjusted... and number of iterations The algorithm was modified to re-employ the Adam algorithm to search for the optimal combination of pump power and voltage of each channel of the electronically controlled polarization controller corresponding to the target spectral state.

6. The method for achieving multispectral switching by synergistic control of pump and polarization state according to claim 5, characterized in that, The recovery failure refers to, If the cumulative adjustment step size of the pump power is greater than the maximum value of the pump power adjustment step size The cumulative voltage adjustment step size of the single channel of the electronically controlled polarization controller is greater than the maximum value. If so, the recovery will fail.

7. The method for achieving multispectral switching by synergistic control of pump and polarization state according to claim 6, characterized in that, The implementation process of the random search algorithm is as follows: The initial conditions for random search are set, including setting the initial voltage values ​​of all four channels of the electronically controlled polarization controller to 0V, the step size to 0.1V, the pump power to the intermediate power value P0, the maximum number of searches to M, and a similarity threshold. ; Within the parameter space, uniform random sampling is performed on the set solution R0=[V1, V2, V3, V4, P0] within a finite range according to a set step size, with the number of samples per round being Ns; The sampled set solution R k The input is an electronically controlled polarization controller and a pumping system, and the laser output spectrum image is acquired by a spectrometer; The recognition model built using a residual neural network performs state recognition on each spectral image, compares the results with the target spectral state, and calculates the similarity. If the similarity calculated from the set solution is greater than the similarity threshold in the current sampling round... If the number of searches exceeds the maximum number of searches M, the search will end.

8. A system for synergistically controlling pump and polarization state to achieve multispectral switching, characterized in that, The system for implementing the method of synergistic regulation of pumping and polarization state to achieve multispectral switching as described in any one of claims 1 to 7, the system comprising: The model training module is used to train a neural network based on spectral images under different electric polarization controller voltages and pump powers to obtain a recognition model. The input of the recognition model is the spectral image, and the output is multiple feasible set solutions for the corresponding pump power and voltage combinations of each channel of the electric polarization controller, and the corresponding spectral state labels. The optimization module is used to input the target spectral image into the recognition model to obtain the set solution and spectral state, and to use an adaptive gradient optimization algorithm to search for the optimal combination of pump power and voltage of each channel of the electric polarization controller corresponding to the target spectral state in the set solution. The switching module is used to generate control signals based on the search results of the optimization module, and feed them back to the electronically controlled polarization controller and the pump to perform fully automatic spectral state switching; the spectral state is switched between continuous light, Q-switched and fundamental frequency mode-locked. The monitoring module is used to enter the spectral monitoring mode after the spectral state switch is successful, and to identify the spectral state in real time and monitor whether the spectral state is lost. The recovery module is used to, based on the monitoring results of the monitoring module, if the spectral state is detected to be lost, execute an adaptive recovery algorithm to restore the spectral state to its original state; if the spectral state is successfully restored, return to the spectral monitoring mode; if the restoration fails, execute a random search algorithm until the spectral state is restored and return to the spectral monitoring mode.

Images