CN118232155A - Solid mode-locked laser control method and system based on twin convolutional neural network - Google Patents

Abstract

The application relates to the technical field of laser pulse characteristic measurement, and provides a solid mode-locked laser control method and system based on a twin convolutional neural network, wherein the control method comprises the steps of training the twin convolutional neural network; acquiring one or more of a second time sequence signal, a second radio frequency spectrum signal or a second spectrum signal; converting the extracted features into a second transient feature sequence; inputting the second characteristic parameter into a second sub-network to obtain a second characteristic parameter; comparing to obtain Euclidean distance; determining that the real-time working state of the solid mode-locked laser does not comprise the stable mode-locked state in response to the Euclidean distance not being in the preset range; and adjusting the position of the X-cavity mirror assembly to enable the X-cavity mirror assembly to be in a stable mode locking state. The control method integrates the automatic analysis and adjustment functions of the convolutional neural network, can reduce manual intervention, improves the degree of automation and the stability, and reduces the maintenance cost and the labor cost. The output performance of the solid mode-locked laser can be continuously adjusted and optimized through real-time monitoring and adjustment.

Description

Solid-state mode-locked laser control method and system based on twin convolutional neural network

Technical Field

The present application relates to the technical field of laser pulse characteristic measurement, and in particular to a solid-state mode-locked laser control method and system based on a twin convolutional neural network.

Background technique

Solid-state mode-locked laser light sources play an important role in scientific research, medical diagnosis and treatment, material processing, communication technology, and lidar. Its stable pulse output and high power characteristics make it an ideal choice for ultrafast optical experiments and precision processing. In scientific research, solid-state mode-locked lasers are used to observe ultrafast dynamic processes such as the vibration and electronic behavior of atoms and molecules. In the medical field, solid-state mode-locked lasers can be used in ophthalmic surgery, skin treatment, and cancer treatment, with the characteristics of high efficiency and precision. In addition, solid-state mode-locked lasers also play an important role in material processing and can be used in fields such as micromachining, laser marking, and cutting. In communication technology, solid-state mode-locked lasers can be used in high-speed data transmission and fiber-optic communication systems. Its advantages include stable pulse output, high power, excellent beam quality, and compact design, which make it widely applicable and reliable in various application scenarios.

However, during the design and construction of a solid-state mode-locked laser light source, the cavity type and cavity length are initially determined according to actual needs or experimental conditions, and the beam waist radius/diameter of the crystal and saturated absorber position is calculated using the ABCD transmission matrix, and the cavity mirror with a suitable curvature is selected to further determine the position and size of the stable zone to guide the experiment. In practice, the position of a key cavity mirror is manually adjusted based on the experimental phenomenon. The range of the mode-locked stable zone is different for crystals doped with different ions, especially for crystals with a small emission cross-section, and the alignment conditions are very harsh. If the cavity mirror position is not adjusted to reach the range of the mode-locked stable zone, it is difficult for the entire mode-locked cavity to form a stable mode-locked laser output or even unable to output laser.

Therefore, how to adjust and determine the position of the key cavity mirror to form a stable solid-state laser mode-locked laser output is a technical problem that is difficult to solve at present.

Summary of the invention

The present application provides a solid-state mode-locked laser control method and system based on a twin convolutional neural network to solve the technical problem that it is difficult to adjust the key cavity mirror to reach the mode-locked stable range during the construction of the solid-state mode-locked laser.

The first aspect of the present application provides a solid-state mode-locked laser control method based on a twin convolutional neural network, comprising: establishing a database and a twin convolutional neural network; wherein the database comprises a first transient feature sequence and a first feature parameter corresponding to the first transient sequence, and the twin convolutional neural network comprises a first subnetwork and a second subnetwork; the twin convolutional neural network is trained using the database; wherein the first subnetwork and the second subnetwork are trained in the same manner; the input of the first subnetwork is the first transient feature sequence, the output of the first subnetwork is the first feature parameter, the first transient feature sequence is one or more of the first timing signal, the first radio frequency spectrum signal or the first spectrum signal of the solid-state mode-locked laser under different working states, obtained after feature extraction, and the working states include a non-light-emitting state, a continuous light output state, a Q-switched state, an incompletely mode-locked state and a stable mode-locked state. A method of the present invention comprises the following steps: obtaining a second timing signal, a second radio frequency spectrum signal or a second spectrum signal of a solid-state mode-locked laser; obtaining one or more of a second timing signal, a second radio frequency spectrum signal or a second spectrum signal of the solid-state mode-locked laser; wherein the second timing signal, the second radio frequency spectrum signal or the second spectrum signal are respectively real-time timing signals, real-time radio frequency spectrum signals or real-time spectrum signals of the solid-state mode-locked laser; converting one or more of the second timing signal, the second radio frequency spectrum signal or the second spectrum signal into a second transient feature sequence after feature extraction; inputting the second transient feature sequence into a second subnetwork to obtain a second feature parameter; comparing the second feature parameter with the first feature parameter to obtain a Euclidean distance; in response to the Euclidean distance being not within a preset range, determining that the real-time working state of the solid-state mode-locked laser does not include a stable mode-locked state; adjusting the position of an X-cavity mirror assembly in the solid-state mode-locked laser according to the real-time working state to put the solid-state mode-locked laser in a stable mode-locked state.

In some feasible implementations, the solid-state mode-locked laser control method based on a twin convolutional neural network further includes: in response to the Euclidean distance being within a preset range, determining that the real-time operating state of the solid-state mode-locked laser is a stable mode-locked state.

In some feasible implementations, the first sub-network and the second sub-network both include a first convolutional layer, a first pooling layer, a second convolutional layer, and a second pooling layer connected in sequence; wherein the first convolutional layer and the second convolutional layer both use a Relu activation function.

The first aspect of the present application provides a solid-state mode-locked laser control method based on a twin convolutional neural network, which determines the working state and performance parameters of the solid-state mode-locked laser by performing high-precision analysis and identification on the input timing signal, radio frequency spectrum or spectrum signal, and realizes precise adjustment of the key cavity mirror position state. Due to the high efficiency and parallel processing capability of the convolutional neural network, the output of the solid-state mode-locked laser can be monitored and adjusted in real time, so that it can achieve a stable mode-locked state and adapt to environmental changes and fluctuations in working conditions. The control method integrates the automated analysis and adjustment functions of the convolutional neural network, which can reduce manual intervention, improve the automation and stability of the system, and reduce maintenance costs and labor costs. Through real-time monitoring and adjustment, the output performance of the solid-state mode-locked laser can be continuously adjusted and optimized to achieve the best working state, improve energy efficiency and output quality.

The second aspect of the present application provides a solid-state mode-locked laser control system based on a twin convolutional neural network, which adopts the solid-state mode-locked laser control method based on a twin convolutional neural network provided in the first aspect. The solid-state mode-locked laser control system based on a twin convolutional neural network includes: a pump source, which is configured to generate light; wherein the light carries energy; an X-cavity mirror assembly, which is arranged on the optical path of the light; a laser crystal, which is arranged in the X-cavity mirror assembly, is configured to absorb energy and generate laser light; wherein the X-cavity mirror assembly is configured to reflect the light to the laser crystal, receive and reflect the laser converted by the laser crystal; a detector, which is arranged on the output optical path of the X-cavity mirror assembly, is configured to receive the laser reflected by the X-cavity mirror assembly; a data acquisition card, which is connected to the detector, is configured to collect data in the laser; a processor, which is connected to the data acquisition card, is configured to process data, determine and output the real-time working state of the solid-state mode-locked laser; a state controller, which is connected to the processor, is configured to adjust the position of the X-cavity mirror assembly according to the real-time working state in response to the real-time working state not including the stable mode-locked state, so that the solid-state mode-locked laser is in a stable mode-locked state.

In some feasible implementations, the X-cavity mirror assembly includes a first cavity mirror, a second cavity mirror, a third cavity mirror, a semiconductor saturated absorption mirror and an output mirror; wherein the laser crystal is arranged between the first cavity mirror and the second cavity mirror.

In some feasible implementations, adjusting the position of the X-cavity mirror assembly according to the real-time working status includes:

The setting positions of the second cavity mirror and the output mirror are adjusted according to the real-time working status.

In some feasible implementations, the solid-state mode-locked laser control system based on the twin convolutional neural network also includes: a first lens and a second lens, and the first lens and the second lens are sequentially arranged between the pump source and the first cavity mirror.

In some feasible implementations, the data acquisition card includes a first acquisition module and a second acquisition module, wherein the first acquisition module is configured to acquire a first timing signal, a first radio frequency spectrum signal or a first spectrum signal of a solid-state mode-locked laser in different working states, and the working state includes one of no light output, continuous light output state, Q-switched state, incomplete mode-locked state and stable mode-locked state; the second acquisition module is configured to acquire one or more of a second timing signal, a second radio frequency spectrum signal or a second spectrum signal of the solid-state mode-locked laser; the second timing signal, the second radio frequency spectrum signal or the second spectrum signal are respectively a real-time timing signal, a real-time radio frequency spectrum signal or a real-time spectrum signal of the solid-state mode-locked laser; the processor includes: an establishment module, configured to establish a database and a twin convolutional neural network; wherein the twin convolutional neural network includes a first subnetwork and a second subnetwork; the database includes a first Transient characteristic parameter, the first transient characteristic parameter is obtained after feature extraction of one or more of the first timing signal, the first radio frequency spectrum signal or the first spectrum signal; the training module is configured to train the twin convolutional neural network using a database; wherein the training method of the first subnetwork and the second subnetwork is the same; the input of the first subnetwork is the first transient characteristic parameter, and the output of the first subnetwork is the first characteristic parameter; the input of the second subnetwork is the second transient characteristic parameter; the second transient characteristic parameter is obtained after feature extraction of one or more of the second timing signal, the second radio frequency spectrum signal or the second spectrum signal, and the output of the second subnetwork is the second characteristic parameter; the comparison module is configured to compare the Euclidean distance between the second characteristic parameter and the first characteristic parameter; the determination module is configured to determine that the real-time working state of the solid-state mode-locked laser does not include a stable mode-locked state in response to the Euclidean distance not being within a preset range.

In some feasible implementations, the determination module is further configured to determine that the real-time operating state of the solid-state mode-locked laser is a stable mode-locked state in response to the Euclidean distance being within a preset range.

In some feasible implementations, the first sub-network and the second sub-network both include a first convolutional layer, a first pooling layer, a second convolutional layer, and a second pooling layer connected in sequence; wherein the first convolutional layer and the second convolutional layer both use a Relu activation function.

The solid-state mode-locked laser control system based on a twin convolutional neural network provided in the second aspect of the present application adopts the solid-state mode-locked laser control method based on a twin convolutional neural network provided in the first aspect. Therefore, the beneficial technical effects that can be achieved can be referred to the first aspect and will not be repeated here.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

FIG1 is a flow chart of a solid-state mode-locked laser control method based on a twin convolutional neural network provided in an embodiment of the present application;

FIG2 is a schematic diagram of the structure of a twin convolutional neural network provided in an embodiment of the present application;

3 is a schematic diagram of a first spectrum signal, a first timing signal, and a first radio frequency spectrum signal of a solid-state mode-locked laser in different states provided by an embodiment of the present application;

4 is a schematic diagram of a second spectrum signal, a second timing signal, and a second radio frequency spectrum signal provided in an embodiment of the present application;

FIG5 is a data processing flow chart of a specific implementation method provided in an embodiment of the present application;

FIG6 is a schematic diagram of the structure of a solid-state mode-locked laser control system based on a twin convolutional neural network provided in an embodiment of the present application.

Graphic marking:

100-solid-state mode-locked laser control system based on twin convolutional neural network; 10-pump source; 20-X cavity mirror assembly; 21-first cavity mirror; 22-second cavity mirror; 23-third cavity mirror; 24-semiconductor saturated absorption mirror; 25-output mirror; 30-laser crystal; 40-detector; 50-data acquisition card; 60-processor; 70-state controller; 80-first lens; 90-second lens.

Detailed ways

The technical solutions in the embodiments of the present application will be described clearly below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of the present application, other embodiments obtained by ordinary technicians in this field without making creative work all belong to the protection scope of the present application.

In the following, the terms "first", "second", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

In addition, in the present application, directional terms such as "upper", "lower", "inner" and "outer" are defined relative to the orientation of the components in the drawings. It should be understood that these directional terms are relative concepts. They are used for relative description and clarification, and they can change accordingly according to the changes in the orientation of the components in the drawings.

In order to solve the above technical problems, the embodiment of the present application provides a solid mode-locked laser control method based on a twin convolutional neural network. Referring to Figure 1, the solid mode-locked laser control method based on a twin convolutional neural network provided in the embodiment of the present application is applied to a solid mode-locked laser control system based on a twin convolutional neural network, and the control method can be implemented by the following steps S100-S800.

Step S100: Establish a database and a twin convolutional neural network.

Among them, step S100 can be implemented by establishing a database and establishing a twin convolutional neural network, and can specifically include the following steps S101 and S102.

Step S101: Establish a database.

Step S102: Establish a twin convolutional neural network.

There is no fixed execution order for step S101 and step S102, and they may be executed in the order of step S101 to step S102, or in the order of step S102 to step S101.

In some feasible implementations, step S101 may be implemented by the following steps S101a to S101b.

Specifically, the twin convolutional neural network is a one-dimensional network, and the input data is one-dimensional data.

Step S101a: Obtain training data.

The training data is one or more of the first timing signal, the first radio frequency spectrum signal or the first spectrum signal of the solid state mode-locked laser in different working states, and the first characteristic parameter corresponding to one or more of the first timing signal, the first radio frequency spectrum signal or the first spectrum signal. The working state includes one of the state of no light output, the state of continuous light output, the Q-switched state, the state of incomplete mode-locking and the state of stable mode-locking.

Specifically, the signals in the continuous light output state, Q-switched state, incomplete mode-locked state and stable mode-locked state are collected when the X-cavity mirror assembly of the solid mode-locked laser is in the stable range, and the signals in the non-light output state are collected when the X-cavity mirror assembly of the solid mode-locked laser is in the unstable range.

For example, 10,000 groups can be recorded in different working states.

At the beginning of acquiring training data, a solid-state mode-locked laser control system based on a twin convolutional neural network can be built to facilitate the processing and feedback of the collected data.

Step S101b: extract features from the training data to obtain a database.

Specifically, one or more of the acquired first time series signal, the first radio frequency spectrum signal or the first spectrum signal is framed and feature extracted to obtain a first transient feature sequence to establish a database, wherein the first transient feature sequence corresponds to the first feature parameter.

The principle of feature extraction is not specifically limited in this application.

For example, when extracting features from the first time series signal, principles such as root mean square, kurtosis, and skewness of the time domain waveform may be used.

When extracting features from the first radio frequency spectrum signal, spectrum features such as power spectrum density, frequency distribution, etc. may be used.

When extracting features from the first spectral signal, principles such as peak wavelength, wavelength range, and spectral width may be used.

The embodiments of the present application do not limit feature extraction, and any of the above-mentioned feature extraction principles may be adopted, or other principles other than those introduced above may be adopted.

In some feasible implementations, before performing feature extraction on one or more of the first time series signal, the first radio frequency spectrum signal, or the first spectral signal, the first time series signal, the first radio frequency spectrum signal, or the first spectral signal may be preprocessed, and the preprocessing may include operations such as denoising and normalization.

Step S200: Use the database to train the twin convolutional neural network.

Specifically, the twin convolutional neural network includes a first subnetwork and a second subnetwork. In the process of training the twin convolutional neural network, the training methods of the first subnetwork and the second subnetwork are the same. The training method can be any training method, and the embodiment of the present application does not specifically limit the training method of the twin convolutional neural network.

After the training is completed, the inputs of the first sub-network and the second sub-network can both be the first transient feature sequence, and the outputs of the first sub-network and the second sub-network can both be the first feature parameter.

In conjunction with Figures 2, 3 and 4, the first subnetwork and the second subnetwork may both include a first convolutional layer, a first pooling layer, a second convolutional layer and a second pooling layer connected to each other. The first convolutional layer and the second convolutional layer may both use a Relu activation function, and the inputs of the first subnetwork and the second subnetwork are both one-dimensional data.

Step S300: obtaining one or more of a second timing signal, a second radio frequency spectrum signal or a second optical spectrum signal of a solid state mode-locked laser.

After training the twin convolutional neural network, the real-time output performance of the solid-state mode-locked laser is further evaluated by collecting real-time signals. The real-time signals include a second timing signal, a second radio frequency spectrum signal or a second spectral signal, which are respectively the real-time timing signal, the real-time radio frequency spectrum signal and the real-time spectral signal of the solid-state mode-locked laser.

Step S400: converting one or more of the second time series signal, the second radio frequency spectrum signal or the second optical spectrum signal into a second transient feature sequence after feature extraction.

In step S400, before the real-time signal is framed for feature extraction, the real-time signal may be preprocessed, wherein the preprocessing may include denoising and normalization. This ensures the accuracy of the real-time signal, facilitates feature extraction and reduces measurement errors. The preprocessed signal is feature extracted and converted into a second transient feature sequence.

The feature extraction principle of the real-time signal is the same as that of the above-mentioned training data, thereby ensuring the uniformity of the data.

For example, when the first time series signal uses the root mean square to extract features, the second time series signal also uses the root mean square to extract features.

Step S500: inputting the second transient feature sequence into the second sub-network to obtain a second feature parameter.

Specifically, the second sub-network serves the real-time signal. After the real-time signal is collected and pre-processed and feature extracted to obtain the second transient feature sequence, it is input into the second sub-network to obtain the second feature parameter. In other words, the input of the second sub-network is the second transient feature sequence, and the output of the second sub-network is the second feature parameter.

It can be understood that the first sub-network is used to output raw data, and the second sub-network is used to output real-time data, so as to facilitate the subsequent determination of the real-time working state of the solid-state mode-locked laser by comparing the real-time data with the raw data. The first sub-network and the second sub-network form a training and comparison model.

Among them, the specific preprocessing method of the real-time signal is the same as the first time series signal, the first radio frequency spectrum signal, and the first optical spectrum signal.

Step S600: Compare the second characteristic parameter with the first characteristic parameter to obtain the Euclidean distance.

The second characteristic parameter is compared with the first characteristic parameter to obtain the Euclidean distance, and the real-time working state of the solid-state mode-locked laser is determined by the Euclidean distance.

Specifically, the data in the database are compared with the real-time signal, the characteristic parameters of the two sub-networks are compared, and the Euclidean distance is used to evaluate the similarity between the real-time signal and the corresponding database frames. The twin convolutional neural network is adjusted according to the results of the similarity evaluation to further improve the performance and stability of the twin convolutional neural network. The Euclidean distance can be used to determine the laser state in which the current solid-state mode-locked laser is working.

It is worth noting that if the real-time signal includes multiple types, for example, the collected real-time signal includes a second timing signal and a second radio frequency spectrum signal, then when the second characteristic parameter is output, two second characteristic parameters are output, namely the timing second characteristic parameter and the radio frequency second characteristic parameter. When comparing, the timing second characteristic parameter is compared with the timing first characteristic parameter output by the first sub-network to obtain the timing Euclidean distance, and the radio frequency second characteristic parameter is compared with the radio frequency first characteristic parameter output by the first sub-network to obtain the radio frequency Euclidean distance. In other words, the real-time signal is compared with the data in the corresponding first sub-network to ensure the output accuracy of the network.

Step S700: In response to the Euclidean distance not being within a preset range, determining that the real-time operating state of the solid state mode-locked laser does not include a stable mode-locked state.

If the Euclidean distance is not within the preset range, it means that the real-time working state of the solid-state mode-locked laser is not a stable mode-locked state. The preset range represents the reference range of the Euclidean distance of the solid-state mode-locked laser in the stable mode-locked state. The specific value of the preset range can be set according to the actual use requirements of the solid-state mode-locked laser, and the embodiment of the present application does not make specific limitations.

Step S800: adjusting the position of the X-cavity mirror assembly in the solid-state mode-locked laser according to the real-time working status, so that the solid-state mode-locked laser is in a stable mode-locked state.

When it is determined that the solid mode-locked laser is not in a stable mode-locked state, the position of the X-cavity mirror assembly is adjusted to put the solid mode-locked laser in a stable mode-locked state.

That is to say, when the twin convolutional neural network evaluates that the solid-state mode-locked laser is in the state of no light output, continuous light output, Q-switched state, or incomplete mode-locked state, it is necessary to adjust the position of one or two key cavity mirrors in the X cavity mirror assembly so that the solid-state mode-locked laser is within the stable mode-locked region. In this way, the solid-state mode-locked laser can achieve stable output.

In addition, when the solid-state mode-locked laser is out of adjustment due to external disturbances or other factors, this control method can be used to achieve self-feedback adjustment.

In some feasible implementations, the control method further includes step S900.

Step S900: In response to the Euclidean distance being within a preset range, determining that the real-time working state of the solid state mode-locked laser is a stable mode-locked state.

At this time, when it is determined that the solid-state mode-locked laser is in a stable mode-locked state, there is no need to adjust the X-cavity mirror assembly.

Specifically, the control method provided in the embodiment of the present application determines the working state and performance parameters of the solid-state mode-locked laser by performing high-precision analysis and identification on the input timing signal, radio frequency spectrum signal or spectrum signal, and realizes precise adjustment of the position state of the key cavity mirror. Due to the high efficiency and parallel processing capability of the convolutional neural network, the control system can monitor and adjust the output of the solid-state mode-locked laser in real time, so that it can achieve a stable mode-locked state and adapt to environmental changes and fluctuations in working conditions. The control method integrates the automated analysis and adjustment functions of the convolutional neural network, which can reduce manual intervention, improve the automation and stability of the control system, and reduce maintenance costs and labor costs. Through real-time monitoring and adjustment, the output performance of the solid-state mode-locked laser can be continuously adjusted and optimized to achieve the best working state, improve energy efficiency and output quality.

In a specific implementation, referring to FIG5 , the solid-state mode-locked laser control method based on a twin convolutional neural network provided in an embodiment of the present application can be implemented by the following steps S1 to S7.

Step S1: Acquire spectrum, time series, and radio frequency spectrum signals.

This step may correspond to the above-mentioned step S300.

Step S2: Data collection.

Step S3: Input data and processing, denoising, normalization, etc.

This step may be a preprocessing process for the real-time signal.

Step S4: Construction and training of twin convolutional neural network.

This step may be part of the above-mentioned steps S100 and S200.

It is worth noting that in this implementation, the reason why step S300 is executed first and then step S100 and step S200 can be understood as a process of continuously acquiring real-time signals. It is precisely by continuously acquiring real-time signals that real-time adjustment of the unstable locking state is achieved.

Step S5: Feature comparison.

Step S6: Similarity analysis.

Step S5 and step S6 may correspond to the above-mentioned step S600.

Step S7: Generate control signal and adjust in real time.

This step may correspond to the above-mentioned step S800.

Corresponding to the embodiment of the aforementioned control method, the present application also provides an embodiment of a solid-state mode-locked laser control system 100 based on a twin convolutional neural network. The solid-state mode-locked laser control system 100 based on a twin convolutional neural network adopts the solid-state mode-locked laser control method based on a twin convolutional neural network provided in the above embodiment.

6 , the solid-state mode-locked laser control system 100 based on the twin convolutional neural network includes a pump source 10 , an X-cavity mirror assembly 20 , a laser crystal 30 , a detector 40 , a data acquisition card 50 , a processor 60 and a state controller 70 .

The pump source 10 is used to generate light, which carries energy. Specifically, the pump source 10 is used to provide energy to excite atoms or ions in the laser crystal 30, so that the atoms or ions are in a higher energy state and are ready to emit stimulated radiation.

The X-cavity mirror assembly 20 is disposed on the optical path of the light, and the X-cavity mirror assembly 20 includes a first cavity mirror 21 , a second cavity mirror 22 , a third cavity mirror 23 , a semiconductor saturated absorption mirror 24 and an output mirror 25 .

Specifically, the first cavity mirror 21, the second cavity mirror 22, the third cavity mirror 23, the semiconductor saturated absorber mirror 24 and the output mirror 25 form a typical X-shaped mode-locked cavity. Based on the ABCD transmission matrix, a solid mode-locked laser is designed to build a resonant cavity with two arms of equal length, so that there is no interval or split between the two stable regions, and the stable region range is relatively large. Among them, the semiconductor saturated absorber mirror 24 is a key component for realizing and stabilizing the mode-locked operation. The semiconductor saturated absorber mirror 24 can start the mode-locked operation through its nonlinear absorption characteristics. When the laser intensity is low, the semiconductor saturated absorber mirror 24 exhibits high absorption, so that the noise and unstable components in the laser cavity are absorbed. As the laser intensity increases to a certain threshold, the absorption rate of the semiconductor saturated absorber mirror 24 decreases and the transmittance increases, thereby promoting the formation of pulses with higher intensity and starting the mode-locked process. The semiconductor saturated absorber mirror 24 can also help maintain the stability of the mode-locked state. Due to its saturated absorption characteristics, it can reduce absorption when the laser pulse reaches a certain intensity, allowing high-intensity pulses to propagate stably in the cavity. This ability to dynamically adjust absorption helps to suppress pulse fluctuations in the cavity and maintain the stability and consistency of the pulse.

The laser crystal 30 is arranged in the X-cavity mirror assembly 20, and the laser crystal 30 is used to receive the energy in the light and convert it into laser. Specifically, the positions of the first cavity mirror 21, the second cavity mirror 22 and the laser crystal 30 are adjusted, and the laser crystal 30 is placed near the center of the calculated stable range. In this way, the laser generated by the laser crystal 30 being excited can be directly sent to the first cavity mirror 21 and the second cavity mirror 22, and reflected in the X-type mode-locked cavity through the first cavity mirror 21 and the second cavity mirror 22.

Among them, the laser crystal 30, as a laser medium, can absorb the energy of the pump source 10, and the electrons are excited to a high energy state. Then, they return to the ground state through stimulated radiation, emitting photons with a specific wavelength to form a laser. The function of the X-cavity mirror assembly 20 is to form an optical resonant cavity, which reflects photons and makes them pass through the laser medium multiple times, increasing the probability of stimulated radiation and thereby amplifying the light intensity.

The detector 40 is arranged on the output optical path of the X-cavity mirror assembly 20, and is used to receive the laser reflected by the X-cavity mirror assembly 20. Specifically, the detector 40 is arranged on the output optical path of the output mirror 25, so as to facilitate receiving the laser emitted by the output mirror 25. Among them, the laser received by the detector 40 contains timing signals, radio frequency spectrum signals and spectrum signals, and these signals can characterize the working state of the solid state mode-locked laser.

The data acquisition card 50 is connected to the detector 40 and is used to collect data in the laser, which are timing signals, radio frequency spectrum signals and optical spectrum signals.

In some feasible implementations, the data acquisition card 50 includes a first acquisition module and a second acquisition module. The first acquisition module is used to acquire a first timing signal, a first radio frequency spectrum signal or a first spectrum signal of the solid state mode-locked laser in different working states and a first characteristic parameter corresponding to the first timing signal, the first radio frequency spectrum signal or the first spectrum signal; the working state includes one of a non-light-emitting state, a continuous light output state, a Q-switched state, an incomplete mode-locked state and a stable mode-locked state.

Among them, the signals in the continuous light output state, Q-switched state, incomplete mode-locked state and stable mode-locked state are collected when the X-cavity mirror assembly of the solid mode-locked laser is in the stable range, and the signals in the non-light output state are collected when the X-cavity mirror assembly of the solid mode-locked laser is in the unstable range.

The second acquisition module is configured to acquire a second timing signal, a second radio frequency spectrum signal or a second spectrum signal of the solid mode-locked laser; wherein the second timing signal, the second radio frequency spectrum signal or the second spectrum signal is respectively a real-time timing signal, a real-time radio frequency spectrum signal or a real-time spectrum signal of the solid mode-locked laser.

That is, the first acquisition module is used to execute step S101a in the above control method embodiment. The second acquisition module is used to execute step S300 in the above control method embodiment.

The processor 60 is connected to the data acquisition card 50 and is used for processing data, determining the real-time working state of the solid-state mode-locked laser according to the processed data, and sending the real-time working state to the state controller 70 .

Specifically, the processor 60 includes a building module, a training module, a comparing module and a determining module.

The establishment module is used to establish a twin convolutional neural network. The twin convolutional neural network includes a first sub-network and a second sub-network; that is, the establishment module is used to execute step S102 in the above control method embodiment.

The training module is used to train the twin convolutional neural network using the training library; wherein, the training method of the first subnetwork and the second subnetwork is the same; the input of the first subnetwork is a first transient characteristic parameter, and the first transient characteristic parameter is obtained after feature extraction of one or more of the first time series signal, the first radio frequency spectrum signal or the first spectrum signal; the output of the first subnetwork is the first characteristic parameter; the input of the second subnetwork is a second transient characteristic parameter; the second transient characteristic parameter is obtained after feature extraction of one or more of the second time series signal, the second radio frequency spectrum signal or the second spectrum signal, and the output of the second subnetwork is the second characteristic parameter; that is, the training module is used to execute step S200 in the above-mentioned control method embodiment.

Specifically, the first sub-network and the second sub-network may include a first convolutional layer, a first pooling layer, a second convolutional layer, and a second pooling layer connected in sequence; wherein the first convolutional layer and the second convolutional layer both use a Relu activation function.

The comparison module is used to compare the Euclidean distance between the second characteristic parameter and the first characteristic parameter; that is, the comparison module is used to execute step S600 in the above control method embodiment.

The determination module is used to determine that the real-time working state of the solid state mode-locked laser does not include a stable mode-locked state in response to the Euclidean distance not being within a preset range. That is, the determination module is used to execute step S700 in the above control method embodiment.

The determination module is also used to determine that the real-time working state of the solid state mode-locked laser is a stable mode-locked state in response to the Euclidean distance being within a preset range. That is, the determination module is also used to execute step S900 in the above control method embodiment.

One end of the state controller 70 is connected to the processor 60, and is used to adjust the setting positions of the second cavity mirror 22 and the output mirror 25 in the X-cavity mirror assembly 20 according to the real-time working state when the real-time working state does not include the stable mode-locked state, so as to put the solid-state mode-locked laser in a stable mode-locked state.

Specifically, the position may include parameters such as coordinates and angles.

The other end of the state controller 70 is connected to the second cavity mirror 22 and the output mirror 25. When the solid-state mode-locked laser control system 100 based on the twin convolutional neural network determines that the solid-state mode-locked laser is not in a stable mode-locked state, the second cavity mirror 22 and the output mirror 25 that are driven electrically can be adaptively adjusted. By changing the setting positions of the second cavity mirror 22 and the output mirror 25, the X-shaped mode-locked cavity is placed in a stable mode-locked region. At this time, the solid-state mode-locked laser can achieve stable output, thereby completing the control of the solid-state mode-locked laser.

Continuing to refer to Fig. 6, the solid-state mode-locked laser control system 100 based on a twin convolutional neural network provided in an embodiment of the present application further includes a first lens 80 and a second lens 90, and the first lens 80 and the second lens 90 are sequentially arranged between the pump source 10 and the first cavity mirror 21. By arranging the first lens 80 and the second lens 90, light can be better transmitted to the first cavity mirror 21.

In the solid-state mode-locked laser control system 100 based on the twin convolutional neural network provided in the embodiment of the present application, the twin convolutional neural network can perform high-precision analysis and identification of the input timing signal, radio frequency spectrum signal or spectrum signal, so as to determine the working state and performance parameters of the solid-state mode-locked laser, and realize accurate adjustment of the position state of the second cavity mirror 22 and the output mirror 25. Due to the high efficiency and parallel processing capability of the twin convolutional neural network, the solid-state mode-locked laser control system 100 based on the twin convolutional neural network can monitor and adjust the output of the solid-state mode-locked laser in real time, so that it can achieve a stable mode-locked state and adapt to environmental changes and fluctuations in working conditions. The twin convolutional neural network can dynamically adjust the feedback to the solid-state mode-locked laser control system 100 based on the twin convolutional neural network according to the data collected in real time, so that the solid-state mode-locked laser control system 100 based on the twin convolutional neural network has adaptability, ensuring output stability and excellent performance. The present invention integrates the automated analysis and adjustment functions of the twin convolutional neural network, which can reduce manual intervention, improve the automation and stability of the solid-state mode-locked laser control system 100 based on the twin convolutional neural network, and reduce maintenance costs and labor costs. Through real-time monitoring and adjustment, the solid-state mode-locked laser control system 100 based on the twin convolutional neural network can continuously adjust and optimize the output performance of the solid-state mode-locked laser, so that it reaches the optimal working state, and improves energy efficiency and output quality. At the same time, the solid-state mode-locked laser control system 100 based on the twin convolutional neural network uses a multi-dimensional-multi-angle laser database in real time to adaptively adjust the mode-locked pulse output of the solid-state mode-locked laser, improve the stability and robustness of the mode-locked laser, and thus improve the redundant operation of the solid-state mode-locked laser in the implementation.

It should be noted that those skilled in the art will easily think of other embodiments of the present application after considering the specification and practicing the application disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary technical means in the art that are not disclosed in the present application. The specification and examples are only regarded as exemplary, and the true scope of the present application is indicated by the claims.

It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims (10)

1. A solid-state mode-locked laser control method based on a twin convolutional neural network, characterized by comprising: Establishing a database and a twin convolutional neural network; wherein the database includes a first transient feature sequence and a first feature parameter corresponding to the first transient sequence, and the twin convolutional neural network includes a first subnetwork and a second subnetwork; The twin convolutional neural network is trained using the database; wherein the first subnetwork and the second subnetwork are trained in the same manner; the input of the first subnetwork is the first transient feature sequence, the output of the first subnetwork is the first feature parameter, the first transient feature sequence is one or more of the first timing signal, the first radio frequency spectrum signal or the first spectrum signal of the solid-state mode-locked laser under different working states, obtained after feature extraction, and the working state includes one of a non-light-emitting state, a continuous light output state, a Q-switched state, an incompletely mode-locked state and a stable mode-locked state; Acquire one or more of a second timing signal, a second radio frequency spectrum signal or a second spectrum signal of the solid-state mode-locked laser; wherein the second timing signal, the second radio frequency spectrum signal or the second spectrum signal is respectively a real-time timing signal, a real-time radio frequency spectrum signal or a real-time spectrum signal of the solid-state mode-locked laser; Convert one or more of the second time series signal, the second radio frequency spectrum signal or the second optical spectrum signal into a second transient feature sequence after feature extraction; Inputting the second transient feature sequence into the second sub-network to obtain a second feature parameter; Comparing the second characteristic parameter with the first characteristic parameter to obtain a Euclidean distance; In response to the Euclidean distance not being within a preset range, determining that the real-time operating state of the solid-state mode-locked laser does not include the stable mode-locked state; The position of the X-cavity mirror assembly in the solid-state mode-locked laser is adjusted according to the real-time working state, so that the solid-state mode-locked laser is in the stable mode-locked state. 2. The solid-state mode-locked laser control method based on a twin convolutional neural network according to claim 1, characterized in that the solid-state mode-locked laser control method based on a twin convolutional neural network further comprises: In response to the Euclidean distance being within the preset range, determining that the real-time operating state of the solid-state mode-locked laser is the stable mode-locked state. 3. The solid-state mode-locked laser control method based on twin convolutional neural network according to claim 1, characterized in that: The first sub-network and the second sub-network both include a first convolutional layer, a first pooling layer, a second convolutional layer, and a second pooling layer connected in sequence; wherein the first convolutional layer and the second convolutional layer both use a Relu activation function. 4. A solid-state mode-locked laser control system based on a twin convolutional neural network, characterized in that the solid-state mode-locked laser control method based on a twin convolutional neural network according to any one of claims 1 to 3 is adopted, and the solid-state mode-locked laser control system based on a twin convolutional neural network comprises: A pump source is configured to generate light; wherein the light carries energy; An X-cavity mirror assembly is arranged on the optical path of the light; A laser crystal, disposed in the X-cavity mirror assembly, is configured to absorb the energy and generate laser light; wherein the X-cavity mirror assembly is configured to reflect the light to the laser crystal, receive and reflect the laser light converted by the laser crystal; A detector, arranged on an output optical path of the X-ray cavity mirror assembly, configured to receive the laser reflected by the X-ray cavity mirror assembly; A data acquisition card, connected to the detector and configured to acquire data from the laser; A processor, connected to the data acquisition card, configured to process the data, determine and output the real-time working state of the solid-state mode-locked laser; A state controller is connected to the processor and is configured to adjust the position of the X-cavity mirror assembly according to the real-time working state in response to the real-time working state not including the stable mode-locked state, so that the solid-state mode-locked laser is in the stable mode-locked state. 5. The solid-state mode-locked laser control system based on twin convolutional neural network according to claim 4, characterized in that: The X-cavity mirror assembly comprises a first cavity mirror, a second cavity mirror, a third cavity mirror, a semiconductor saturated absorption mirror and an output mirror; wherein the laser crystal is arranged between the first cavity mirror and the second cavity mirror. 6. The solid-state mode-locked laser control system based on twin convolutional neural network according to claim 5, characterized in that: Adjusting the position of the X-cavity mirror assembly according to the real-time working state includes: The setting positions of the second cavity mirror and the output mirror are adjusted according to the real-time working status. 7. The solid-state mode-locked laser control system based on twin convolutional neural network according to claim 5, characterized in that: The solid-state mode-locked laser control system based on a twin convolutional neural network also includes: a first lens and a second lens, wherein the first lens and the second lens are sequentially arranged between the pump source and the first cavity mirror. 8. The solid-state mode-locked laser control system based on twin convolutional neural network according to claim 4, characterized in that: The data acquisition card includes a first acquisition module and a second acquisition module, wherein the first acquisition module is configured to acquire a first timing signal, a first radio frequency spectrum signal or a first spectrum signal of the solid-state mode-locked laser in different working states, wherein the working state includes one of no light output, continuous light output state, Q-switched state, incomplete mode-locked state and stable mode-locked state; the second acquisition module is configured to acquire one or more of a second timing signal, a second radio frequency spectrum signal or a second spectrum signal of the solid-state mode-locked laser; the second timing signal, the second radio frequency spectrum signal or the second spectrum signal are respectively a real-time timing signal, a real-time radio frequency spectrum signal or a real-time spectrum signal of the solid-state mode-locked laser; The processor comprises: An establishment module is configured to establish a database and a twin convolutional neural network; wherein the twin convolutional neural network includes a first subnetwork and a second subnetwork; the database includes a first transient characteristic parameter, and the first transient characteristic parameter is obtained after feature extraction of one or more of the first time series signal, the first radio frequency spectrum signal, or the first spectrum signal; A training module is configured to train the twin convolutional neural network using the database; wherein the first subnetwork and the second subnetwork are trained in the same manner; the input of the first subnetwork is the first transient characteristic parameter, and the output of the first subnetwork is the first characteristic parameter; the input of the second subnetwork is the second transient characteristic parameter; the second transient characteristic parameter is obtained after feature extraction of one or more of the second time series signal, the second radio frequency spectrum signal, or the second spectrum signal, and the output of the second subnetwork is the second characteristic parameter; A comparison module, configured to compare the Euclidean distance between the second feature parameter and the first feature parameter; The determination module is configured to determine, in response to the Euclidean distance not being within a preset range, that the real-time working state of the solid-state mode-locked laser does not include the stable mode-locked state. 9. The solid-state mode-locked laser control system based on twin convolutional neural network according to claim 8, characterized in that: The determination module is further configured to determine, in response to the Euclidean distance being within the preset range, that the real-time operating state of the solid-state mode-locked laser is the stable mode-locked state. 10. The solid-state mode-locked laser control system based on twin convolutional neural network according to claim 8, characterized in that: The first sub-network and the second sub-network both include a first convolutional layer, a first pooling layer, a second convolutional layer, and a second pooling layer connected in sequence; wherein the first convolutional layer and the second convolutional layer both use a Relu activation function.