Dual identification system based on fractional order vortex beam and improved non-Kolmogorov turbulence model and use method

Abstract

According to the dual recognition system based on the fractional order vortex light beam and the improved non-Kolmogorov turbulence model and the use method, classification and recognition of the fractional order topological charge and the propagation distance can be achieved in the improved non-Kolmogorov turbulence model. And compared with a traditional integer order scheme, the recognition precision is improved by 6.65% to 15%. According to the method, the precision and robustness of optical ranging and optical communication can be remarkably improved, and theoretical support and application prospects are provided for a future optical system.

Description

Technical Field

[0001] The present invention relates to the fields of optical signal processing and deep learning, and in particular to a dual recognition system and a usage method based on fractional-order vortex beams and an improved non-Kolmogorov turbulence model. Background Art

[0002] In optical systems, vortex beams have been widely used in fields such as optical ranging, communication, and imaging due to their carrying orbital angular momentum (OAM). However, traditional integer-order topological charge vortex beams are vulnerable to noise and turbulence in a turbulent environment, resulting in difficult signal recognition. In recent years, fractional-order topological charge beams, due to their radial notch characteristics related to fractions appearing in the transverse diffraction field, have provided a new way to improve the recognition accuracy. Summary of the Invention

[0003] The purpose of the present invention is to propose an identification system that combines the ECKNet (Efficient Convolutional KANNetwork) deep learning network and an improved non-Kolmogorov turbulence model. This system simulates the propagation behavior of vortex beams under different turbulence conditions and uses a deep learning model for data classification and feature extraction. Experiments show that this system is significantly superior to traditional solutions in terms of recognition accuracy.

[0004] To achieve the above object, the present invention proposes a dual recognition system based on fractional-order vortex beams and an improved non-Kolmogorov turbulence model, including a deep learning network module, a non-Kolmogorov turbulence simulation module, a data acquisition and preprocessing module, and a display and control module;

[0005] The deep learning network module is ECKNet for feature extraction and classification of optical signals;

[0006] The non-Kolmogorov turbulence simulation module realizes the propagation simulation of vortex beams in a turbulent environment;

[0007] The data acquisition and preprocessing module is used to acquire and process optical field data with different topological charges and propagation distances;

[0008] The display and control module displays the classification results and sets the system parameters.

[0009] Furthermore, the deep learning network module embeds the Kolmogorov–Arnold network into EfficientNet V2 and adds the CBAM attention mechanism to improve the classification ability and computational efficiency of the model.

[0010] Further, the non-Kolmogorov turbulence simulation module uses a Gaussian beam to load a helical phase wavefront through a non-Kolmogorov atmospheric turbulence model to obtain the intensity distribution map of fractional-order vortex light;

[0011] The split-step phase screen propagation method is used to simulate the influence of different-scale turbulence on the vortex beam, including the adjustment of outer scale and inner scale parameters.

[0012] Further, the data acquisition and preprocessing module is used to acquire and process the light field data of different topological charges and propagation distances. There are 10 types of topological charges and 10 types of propagation distances, with a total of 100 categories. All images are divided into a training set, a validation set, and a test set according to a ratio of 3:1:1.

[0013] Further, the display and control module displays the classification results and error rate of the system and obtains a confusion matrix diagram.

[0014] The present invention also proposes a usage method of a dual recognition system based on fractional-order vortex beams and an improved non-Kolmogorov turbulence model, including the following steps:

[0015] Step 1: Use a Gaussian beam, load a helical phase wavefront on the Gaussian beam, and then transmit the beam through atmospheric turbulence to obtain a fractional-order vortex beam;

[0016] Step 2: Classify the intensity distribution map obtained in Step 1 according to different topological charges and different propagation distances, and divide the classified images into a training set, a validation set, and a test set according to a ratio of 3:1:1 for convenient subsequent network model training;

[0017] Step 3: Feed the training set and the validation set into the ECKNet network model for training and validation. ECKNet is a combination of EfficientNet V2 and the Kolmogorov–Arnold network. A two-layer Kolmogorov–Arnold network is used to replace the MLP fully connected layer in the EfficientNetV2 network to improve the classification ability of the network, and a CBAM attention mechanism module is added to the MBconv module of the EfficientNetV2 network model to improve the feature extraction ability of the network; obtain the classification results through the training and validation of the ECKNet network model;

[0018] Step 4: Use the test set to test the trained model in Step 3 and display the test results in the display module.

[0019] Compared with the prior art, the advantages of the present invention are as follows: The technical solution of the present invention can classify and identify fractional topological charges and propagation distances in an improved non-Kolmogorov turbulence model. The recognition accuracy of the solution has been improved by 6.65% to 15% compared with the traditional integer-order solution. This method can significantly improve the accuracy and robustness of optical ranging and optical communication, providing theoretical support and application prospects for future optical systems. Description of the Drawings

[0020] Figure 1 It is the overall architecture diagram of the dual recognition system in the embodiment of the present invention

[0021] Figure 2 It is the structural diagram of the non-Kolmogorov turbulence model in the embodiment of the present invention;

[0022] Figure 3 It is the ECKNet network structure diagram in the embodiment of the present invention;

[0023] Among them, (a) is the overall structure diagram of the ECKNet network, and (b) and (c) are the MBconv modules with and without using the SE attention module respectively. Whether the SE module is used or not, the CBAM attention mechanism module will be used at the end.

[0024] Figure 4 It is the beam intensity distribution comparison diagram in the embodiment of the present invention;

[0025] Figure 5 It is the schematic diagram of the classification accuracy results of the model in the embodiment of the present invention under different parameters. Detailed Embodiments

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described below.

[0027] As Figure 1 shown, the present invention proposes a dual recognition system based on fractional-order vortex beams and an improved non-Kolmogorov turbulence model, including a deep learning network module, a non-Kolmogorov turbulence simulation module, a data acquisition and preprocessing module, and a display and control module;

[0028] The deep learning network module is ECKNet, (Efficient Convolutional KAN Network), which is a combination of EfficientNet V2 and the Kolmogorov–Arnold network, as Figure 3As shown in the figure, a two-layer Kolmogorov–Arnold network is used to replace the MLP fully connected layer in the EfficientNetV2 network to improve the classification ability of the network, and a CBAM attention mechanism module is added to the MBconv (mobile bottleneck convolution) module of the EfficientNetV2 network model to improve the feature extraction ability of the network. The classification results are obtained through the training and validation of the ECKNet network model.

[0029] The non-Kolmogorov turbulence simulation module is as Figure 2 shown. A Gaussian beam is used to load a helical phase wavefront through a non-Kolmogorov atmospheric turbulence model to obtain the intensity distribution diagram of fractional-order vortex light. According to the literature "Evolution of an optical vortex with an initial fractional topological charge", the Gaussian beam loaded with a helical phase wavefront can be expressed as:

[0030]

[0031] In the formula: l represents the topological charge of the fractional-order vortex light, and ω represents the beam waist. According to the literature "Optical scintillations and fade statistics for FSO communications through moderate-to-strong non-Kolmogorov turbulence", the improved non-Kolmogorov atmospheric turbulence power spectral density (PSD) can be expressed as:

[0032]

[0033] In the formula: Δz represents the interval between two adjacent phase screens, k is the spectral parameter, α = 22 / 6, β = 7 / 6, a 1 = 1.802, b 1 = 0.254, k 0 = 2π / L 0 k l is expressed as formula 3, l 0 and L 0 are the inner scale and the outer scale.

[0034] Data acquisition and preprocessing module:

[0035] The intensity distribution of the light beam passing through the turbulence is recorded using a turbulent phase screen system, and the collected data is normalized and denoised. Each set of data is divided into a training set, a validation set, and a test set in a ratio of 3:1:1. The training set and the validation set are used to train the network model, and the test set is used to test the effect of the trained model. The model is trained using an Nvidia A16 GPU, the Adam optimizer is used, the initial learning rate is 0.001, and the StepLR strategy is adopted to gradually reduce the learning rate. The learning rate is set to be reduced to 0.1 times the original every 10 epochs, and each dataset is trained for 50 epochs. The model performs feature mapping through a two-layer KAN network, and the final classification result is presented in the display module. As Figure 4 shown.

[0036] Display and control module, which displays the classification result and sets the system parameters.

[0037] Experimental results and analysis

[0038] In the experiment, as Figure 5 shown, there are 10 different topological charges ranging from 5.1 to 6.0 with an interval of 0.1, and 10 different transmission distances ranging from 600 meters to 1500 meters with an interval of 100 meters for the topological charge types, and the total classification is 100 classes. When the atmospheric turbulence intensity SNR is 30, 20, and 10 dB respectively, the classification accuracies of the system for different topological charges and propagation distances are 98.61%, 98.24%, and 97.27% respectively. When the atmospheric turbulence intensity SNR is 30, 20, and 10 dB respectively, the classification accuracies of the system for different topological charges and propagation distances are 97.43%, 96.78%, and 90.79% respectively. When the atmospheric turbulence intensity SNR is 30, 20, and 10 dB respectively, the classification accuracies of the system for different topological charges and propagation distances are 95.38%, 94.59%, and 81.48% respectively. The comparative experiment shows that when SNR = 10 dB and the atmospheric turbulence intensity is 1×10 -16 m -2 / 3 and 2×10 -16 m -2 / 3 respectively, the fractional-order scheme improves the recognition accuracy by 6.65% and 15% compared with the integer-order scheme.

[0039] System application examples

[0040] This system can be widely applied in the following fields:

[0041] Optical ranging: Under atmospheric turbulence conditions, use fractional-order vortex beams to achieve higher-precision distance measurement;

[0042] Optical communication: By identifying optical beams with different topological charges, multi-channel information transmission is achieved, improving communication efficiency;

[0043] Optical imaging: In biological and medical imaging, finer image processing is achieved by adjusting the topological charge.

[0044] The present invention proposes a novel optical signal recognition system. By combining fractional-order vortex beams with a deep learning model, efficient signal classification and recognition in a complex atmospheric environment are achieved. This system not only improves the recognition accuracy but also provides important technical support for future optical ranging and communication systems. The above is only the preferred embodiment of the present invention and does not impose any limitation on the present invention. Any person skilled in the art, within the scope of the technical solution of the present invention, making any form of equivalent substitution or modification and other changes to the technical solution and technical content disclosed by the present invention, all belong to the content within the scope of the technical solution of the present invention and still fall within the protection scope of the present invention.

Claims

1. A dual identification system based on fractional vortex beam and improved non-Kolmogorov turbulence model, characterized in that: It includes deep learning network module, non-Kolmogorov turbulence simulation module, data acquisition and preprocessing module and display and control module; The deep learning network module is ECKNet for feature extraction and classification of optical signals; The non-Kolmogorov turbulence simulation module realizes the propagation simulation of the vortex beam in a turbulent environment; The data acquisition and preprocessing module is used to acquire and process light field data with different topological charges and propagation distances; The display and control module displays the classification results and performs system parameter settings.

2. The dual identification system based on fractional-order vortex beam and improved non-Kolmogorov turbulence model according to claim 1 is characterized in that: The deep learning network module embeds the Kolmogorov–Arnold network into EfficientNetV2 and adds the CBAM attention mechanism to improve the classification ability and computational efficiency of the model.

3. The dual identification system based on fractional-order vortex beam and improved non-Kolmogorov turbulence model according to claim 1 is characterized in that: The non-Kolmogorov turbulence simulation module uses a Gaussian beam to load a spiral phase wavefront through a non-Kolmogorov atmospheric turbulence model to obtain a fractional-order vortex light intensity distribution diagram; A step-by-step phase screen transmission method is used to simulate the effects of turbulence at different scales on the vortex beam, including the adjustment of outer and inner scale parameters.

4. The dual identification system based on fractional-order vortex beam and improved non-Kolmogorov turbulence model according to claim 1, characterized in that: The data acquisition and preprocessing module is used to acquire and process light field data of different topological charges and propagation distances. There are 10 types of topological charges, 10 types of transmission distances, and a total of 100 types. All images are divided into training set, verification set and test set in a ratio of 3:1:

1.

5. The dual identification system based on fractional-order vortex beam and improved non-Kolmogorov turbulence model according to claim 1, characterized in that: The display and control module displays the classification results and error rate of the system and obtains a confusion matrix diagram.

6. A method for using a dual identification system based on a fractional-order vortex beam and an improved non-Kolmogorov turbulence model, using the dual identification system based on a fractional-order vortex beam and an improved non-Kolmogorov turbulence model as claimed in any one of claims 1 to 5, comprising the following steps: Step 1: Use a Gaussian beam and load a spiral phase wavefront on the Gaussian beam, then transmit the beam through atmospheric turbulence to obtain a fractional-order vortex beam; Step 2: Classify the light intensity distribution map obtained in step 1 according to different topological charges and different transmission distances, and divide the classified images into training set, verification set and test set in a ratio of 3:1:1 to facilitate subsequent network model training; Step 3: Send the training set and the validation set to the ECKNet network model for training and validation. The ECKNet is a combination of EfficientNet V2 and Kolmogorov–Arnold network. A two-layer Kolmogorov–Arnold network is used to replace the MLP fully connected layer in the EfficientNetV2 network to improve the classification ability of the network. A CBAM attention mechanism module is added to the MBconv module of the EfficientNetV2 network model to improve the feature extraction ability of the network. The classification results are obtained through the training and validation of the ECKNet network model. Step 4: Use the test set to test the trained model in step 3, and display the test results in the real module.

Images