A simulation system and method for pattern recognition in underwater vortex optical communication

Abstract

This invention discloses a simulation system and method for pattern recognition in underwater vortex optical communication, belonging to the field of underwater optical communication technology. The system includes: a beam generation module for generating conjugate superimposed vortex beams with different topological charges and illuminating them onto an underwater disturbance module; an underwater disturbance module for setting up an underwater disturbance environment; a data acquisition module for capturing beam intensity images of the conjugate superimposed vortex beams after transmission through the underwater disturbance module; and a data processing module including a deep learning recognition model and a display control device. The deep learning recognition model is used to recognize the beam intensity images obtained by the data acquisition module; the display control device is used to display the recognition results, the parameters for training the deep learning recognition model, and to adjust the topological charge of the conjugate superimposed vortex beams in the beam generation module. The method involves using the above system for recognition. This invention is adaptable to complex underwater environments and exhibits high robustness in the recognition simulation system.

Description

Technical Field

[0001] This invention belongs to the field of underwater optical communication technology, and more specifically, relates to a simulation system and method for underwater vortex optical communication pattern recognition. Background Technology

[0002] Oceans cover approximately 70% of the Earth's surface and play a crucial role in climate regulation, biodiversity, and resource reserves. With increasingly scarce land resources, the urgency of marine exploration and development is constantly rising, necessitating high-speed, reliable underwater wireless communication technologies. Traditional underwater communication technologies, such as acoustic and radio frequency (RF) communication, have significant limitations: acoustic communication, while having a long transmission distance, suffers from limited bandwidth and high latency; RF communication experiences severe attenuation in water, typically limiting communication distances to only a few meters.

[0003] Underwater optical communication, with its high bandwidth, low latency, and strong anti-interference capabilities, has become a highly promising solution, with blue-green light in the 450-550nm band providing its physical basis. Vortex beams carrying orbital angular momentum, as novel optical carriers, can construct high-dimensional multiplexing spaces through the orthogonality of different topological charges, thereby improving channel capacity. However, single vortex beams are prone to mode dispersion in complex underwater channels, limiting their practical application. While conjugate superimposed vortex beams possess multidimensional structures and unique intensity distributions, they still face challenges in underwater transmission, such as absorption and scattering by suspended particles, wavefront distortion caused by water turbulence, and beam splitting due to biological obstruction. Traditional vortex mode detection methods, such as diffraction gratings and spiral phase plates, are complex to deploy and perform poorly in complex underwater environments.

[0004] A search revealed Chinese patent application number 202510316100.3, published on June 24, 2025, which discloses an underwater vortex optical communication anti-turbulence method based on a photoelectric hybrid deep neural network. This patent integrates a diffractive deep neural network and a convolutional neural network. At the front end, the diffractive deep neural network performs all-optical information compression, and the processed information is transmitted to the back end of the convolutional neural network. Through training, the model completes its task, outputting the coefficients of the predicted Zernike polynomials to predict and compensate for ocean turbulence, thereby improving the mode purity of the vortex beam. However, this scheme involves a single vortex beam, and its turbulence modeling relies on Zernike polynomials, which cannot cover all turbulence scenarios.

[0005] Deep learning has been gradually applied to optical vortex pattern recognition to mitigate the effects of turbulence or fiber optic disturbances, but its application in underwater optical communication, especially in the recognition of conjugate superimposed vortex beams, is still in its early stages. Existing research is mostly limited to single beam types and simplified scenarios, and its simulation of real underwater complex disturbances is insufficient, making it difficult to meet the needs of practical applications.

[0006] Therefore, in order to adapt to complex underwater environments and ensure the high robustness of the identification system, it is of great significance to develop a simulation system and method for underwater vortex optical communication pattern recognition. Summary of the Invention

[0007] 1. The problem to be solved

[0008] The purpose of this invention is to provide a simulation system for pattern recognition in underwater vortex optical communication, aiming to achieve highly robust pattern recognition in complex underwater environments. Furthermore, a method for pattern recognition in underwater vortex optical communication is also provided.

[0009] 2. Technical Solution To solve the above problems, the technical solution adopted by the present invention is as follows: A first aspect of the present invention provides a simulation system for underwater vortex optical communication pattern recognition, comprising a beam generation module, an underwater disturbance module, a data acquisition module, and a data processing module, wherein: The beam generation module is used to generate conjugate superimposed vortex beams with different topological charge numbers and illuminate the underwater disturbance module. Underwater disturbance module, setting up an underwater disturbance environment; The data acquisition module is used to capture the beam intensity image of the conjugate superimposed vortex beam after it has been transmitted through the underwater disturbance module; The data processing module includes a deep learning recognition model and a display control device. The deep learning recognition model is a ResNet-50 network, used to recognize the beam intensity image obtained by the data acquisition module. The display control device is used to display the recognition results, the parameters for training the deep learning recognition model, and to adjust the topological charge of the conjugate superimposed vortex beam in the beam generation module, and supports system parameter adjustment.

[0010] As one possible implementation, the beam generation module is arranged sequentially along the optical path direction, comprising a laser, an attenuator, a polarizer, a beam expander and collimator, a spatial light modulator, and a 4f spatial filter. The laser output beam is attenuated by the attenuator and then adjusted to linearly polarized light by the polarizer. The beam expander and collimator consists of two convex lenses, a first lens and a second lens, which expand and collimate the linearly polarized light into a parallel Gaussian beam. The parallel Gaussian beam is incident on the pure phase spatial light modulator, which performs phase modulation on the parallel Gaussian beam to generate a conjugate superimposed vortex beam containing the target topological charge. The modulated beam is incident on the 4f spatial filter to finally output a high-quality conjugate superimposed vortex beam. The 4f spatial filter is arranged sequentially along the optical path direction, comprising a third lens, an aperture stop, and a fourth lens.

[0011] When using the above technical solution, the generation principle of the conjugate superimposed vortex beam is as follows: the topological charge is... l The complex amplitude of the vortex beam is Its conjugate beam (topological charge) Complex amplitude is The intensity distribution after the interference of the two beams is as follows The number of petals formed is 2. | Petal-like intensity distribution.

[0012] As one possible implementation, the laser output beam is green light, and the laser is preferably a 535 nm semiconductor laser; the spatial light modulator is a pure phase type, and the 4f spatial filter component is set with an adjustable aperture in the Fourier plane to isolate the +1st diffraction order, with a diffraction efficiency of about 15%.

[0013] As one possible implementation, in the underwater disturbance module, the underwater disturbance environment is composed of one or more factors such as turbulence intensity, turbidity, obstruction and density non-uniformity. The turbulence intensity includes weak, medium and strong turbulence, and the turbidity includes low, medium and high turbidity, forming 9 combined disturbance environments. As one possible implementation, the method for setting the turbulence intensity in the underwater disturbance environment is as follows: weak, medium, and strong turbulence are simulated by setting water pumps with power of P1, P2, and P3 respectively in the water, where P1 < P2 < P3.

[0014] As one possible implementation, the method for setting the turbidity in the underwater disturbance environment is as follows: low, medium and high turbidity are simulated by adding kaolin with masses of m1, m2 and m3 to the water, respectively, where m1 < m2 < m3.

[0015] As one possible implementation, the method for setting the occlusion in the underwater disturbance environment is as follows: a black occlusion is generated in the beam intensity image obtained by the data acquisition module to simulate the occlusion situation in the water, wherein the area of ​​the black occlusion covers 0.25% to 76.56% of the area of ​​the beam intensity image, preferably 25% to 56.25%.

[0016] As one possible implementation, the data acquisition module consists of a focusing lens and a CCD camera. After being transmitted through the underwater disturbance module, the conjugate superimposed vortex beam is detected by the CCD camera on the focusing plane under the action of the focusing lens, and a beam intensity image is formed.

[0017] As one possible implementation, the CCD camera sets a 400×400 pixel area as the region of interest for image acquisition, wherein a black occlusion smaller than 350×350 pixels is generated in the region of interest, preferably smaller than 300×300 pixels, and even more preferably smaller than 200×200 pixels.

[0018] As one possible implementation, the deep learning recognition model uses the ResNet-50 network as the core classification model. This network includes a 7×7 convolutional layer (stride 2), a 3×3 max pooling layer (stride 2), and four stages with bottleneck residual blocks. Finally, it outputs 16-class pattern recognition results through global average pooling and fully connected layers.

[0019] As one possible implementation, the display control device is used to display the recognition results, the loss value and accuracy during the training process, and the system operating status, and to set the laser power, spatial light modulator phase parameters, water pump power, CCD camera acquisition parameters, and network training parameters.

[0020] A second aspect of the present invention provides a method for underwater vortex optical communication pattern recognition, employing any of the above-mentioned systems for recognition, comprising the following steps: S1. A conjugate superimposed vortex beam is generated using a beam generation module; S2. Construct an underwater disturbance module; S3. Acquire intensity images and construct dataset: Transmit the conjugate superimposed vortex beam generated in step S1 through the underwater disturbance module constructed in step S2, use the data acquisition module to capture beam intensity distribution images in the set region of interest, and construct a dataset, dividing the dataset into training set, validation set and test set. S4. Training and recognition of the dataset: Input the training set and validation set from step S3 into the deep learning recognition model for training, use the test set to test the trained deep learning recognition model, and input the recognition results into the display control device.

[0021] As one possible implementation, the display control device can view the pattern recognition results in step S4, the loss value and accuracy during network training, and the operating parameters of each module; based on the recognition effect, adjust the spatial light modulator phase parameters, water pump power, CCD camera acquisition parameters, or network training parameters to optimize system performance.

[0022] As one possible implementation, in step S3, the dataset constructed by the data acquisition module contains 16 conjugate superimposed vortex modes. Each mode collects 5,000 image samples under 9 disturbance conditions (3 intensities of turbulence and 3 intensities of turbulence can be freely combined to form 9), for a total of 720,000 images. The dataset is divided into training set, validation set and test set according to a preset ratio, preferably in an 8:1:1 ratio. The optimal training configuration for the ResNet-50 network is 300 image samples per class and 25 epochs of training.

[0023] 3. Beneficial effects Compared with the prior art, the beneficial effects of the present invention are as follows: The system and method of this invention first employ a conjugate superimposed vortex beam as a transmission carrier. Its petal-shaped intensity distribution has strong information redundancy, and it can still retain modal information under local occlusion or non-uniform scattering, thus improving the anti-interference capability of the physical layer. Then, by combining the residual structure and hierarchical convolution of the ResNet-50 network, distributed local features are effectively integrated to adapt to the spatial information characteristics of the conjugate superimposed vortex beam, thereby achieving high-precision recognition of degraded images. Finally, image preprocessing further improves the recognition stability under extremely strong interference. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the structure of the underwater vortex optical communication pattern recognition simulation system of the present invention.

[0025] Figure 2 The image shows the phase hologram (left) of the conjugate superimposed vortex beam l=±4 in an embodiment of the present invention, and the petal-shaped intensity distribution (right) generated by loading it onto a spatial light modulator.

[0026] Figure 3 This is a schematic diagram of the underwater disturbance module in an embodiment of the present invention.

[0027] Figure 4 These are some of the experimental samples collected under nine different disturbance environments in this embodiment of the invention.

[0028] Figure 5 This is a schematic diagram of the ResNet-50 model structure in an embodiment of the present invention.

[0029] Figure 6 This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 2g of kaolin (weak scattering) and 1W of stirring power (weak perturbation) in an embodiment of the present invention.

[0030] Figure 7 This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 2g of kaolin (weak scattering) and 1.5W of stirring power (medium perturbation) in an embodiment of the present invention.

[0031] Figure 8 This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 2g of kaolin (weak scattering) and 2W of stirring power (strong perturbation) in an embodiment of the present invention.

[0032] Figure 9 This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 4g of kaolin (medium scattering) and 1W of stirring power (weak perturbation) in an embodiment of the present invention.

[0033] Figure 10This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 4g of kaolin (medium scattering) and 1.5W of stirring power (medium perturbation) in an embodiment of the present invention.

[0034] Figure 11 This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 4g of kaolin (medium scattering) and 2W of stirring power (strong perturbation) in an embodiment of the present invention.

[0035] Figure 12 This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 6g of kaolin (strong scattering) and 1W of stirring power (weak perturbation) in an embodiment of the present invention.

[0036] Figure 13 This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 6g of kaolin (strong scattering) and 1.5W of stirring power (medium perturbation) in an embodiment of the present invention.

[0037] Figure 14 This is a graph showing the loss and accuracy of ResNet-50 under the perturbation conditions of 6g of kaolin (strong scattering) and 2W of stirring power (strong perturbation) in an embodiment of the present invention.

[0038] Figure 15 This is a schematic diagram showing different shading areas in an embodiment of the present invention.

[0039] Figure 16 This describes the recognition performance of ResNet-50 under a 20×20 pixel black occlusion in an embodiment of the present invention.

[0040] Figure 17 This is the recognition performance of ResNet-50 under a 40×40 pixel black occlusion in an embodiment of the present invention.

[0041] Figure 18 This describes the recognition performance of ResNet-50 under a 60×60 pixel black occlusion in an embodiment of the present invention.

[0042] Figure 19 This describes the recognition performance of ResNet-50 under a black occlusion of 80×80 pixels in an embodiment of the present invention.

[0043] Figure 20 This describes the recognition performance of ResNet-50 under a 100×100 pixel black occlusion in an embodiment of the present invention.

[0044] Figure 21 This describes the recognition performance of ResNet-50 under a 120×120 pixel black occlusion in an embodiment of the present invention.

[0045] Figure 22This describes the recognition performance of ResNet-50 under a 140×140 pixel black occlusion in an embodiment of the present invention.

[0046] Figure 23 This refers to the recognition performance of ResNet-50 under a black occlusion of 160×160 pixels in an embodiment of the present invention.

[0047] Figure 24 This describes the recognition performance of ResNet-50 under a 200×200 pixel black occlusion in an embodiment of the present invention.

[0048] Figure 25 This refers to the recognition performance of ResNet-50 under a black occlusion of 250×250 pixels in an embodiment of the present invention.

[0049] Figure 26 This describes the recognition performance of ResNet-50 under a black occlusion of 300×300 pixels in an embodiment of the present invention.

[0050] Figure 27 This refers to the recognition performance of ResNet-50 under a black occlusion of 350×350 pixels in an embodiment of the present invention.

[0051] In the picture: 1. Beam generation module; 11. Laser; 12. Attenuator; 13. Polarizer; 14. Beam expander and collimator assembly; 141. First lens; 142. Second lens; 15. Spatial light modulator; 16. 4f spatial filter assembly; 161. Third lens; 162. Aperture stop; 163. Fourth lens; 2. Underwater disturbance module; 3. Data acquisition module; 31. Focusing lens; 32. CCD camera; 4. Data processing module. Detailed Implementation

[0052] The present invention will be further described below with reference to specific embodiments.

[0053] Example like Figure 1 The diagram illustrates a simulation system for underwater vortex optical communication pattern recognition, comprising a beam generation module 1, an underwater disturbance module 2, a data acquisition module 3, and a data processing module 4. These modules work collaboratively to achieve highly robust pattern recognition in complex underwater environments. The structural composition and operational process of each module are as follows: The beam generation module 1 is arranged along the optical path direction as follows: a laser 11, an attenuator 12, a polarizer 13, a beam expander and collimator 14, a spatial light modulator 15, and a 4f spatial filter 16. The output beam from the laser 11 is attenuated by the attenuator 12 and then adjusted to linear polarization by the polarizer 13. The beam expander and collimator 14 consists of two convex lenses, a first lens 141 and a second lens 142, which expand and collimate the linearly polarized light into a parallel Gaussian beam. This parallel Gaussian beam is incident on the pure phase spatial light modulator 15, where phase modulation is performed to generate a conjugate superimposed vortex beam containing the target topological charge. The modulated beam is then incident on the 4f spatial filter 16, ultimately outputting a high-quality conjugate superimposed vortex beam. The 4f spatial filter 16 is arranged along the optical path direction as follows: a third lens 161, an aperture 162, and a fourth lens 163. Laser 11 is a 535 nm semiconductor laser (5 mW power). Its output beam is attenuated by a neutral density attenuator 12 and then adjusted to linear polarization by a polarizer 13. A beam expander and collimator 14 expands and collimates the beam into a parallel Gaussian beam. The parallel Gaussian beam is incident on a pure phase spatial light modulator 15 (BNT SLM-PAA-380, 8 μm pixel size). The spatial light modulator 15 loads a hologram generated by the data processing module 4 to perform phase modulation on the beam, generating a vortex beam containing the target topological charge and its conjugate topological charge. The modulated beam is then incident on a 4f spatial filter 16, which has an adjustable aperture 162 in the Fourier plane, allowing only the +1st order diffraction order to pass through, ultimately obtaining a high-quality conjugate superimposed vortex beam with 2|l| (where l is the topological charge number). For example... Figure 2 The image shows a phase hologram (left) of a conjugate superimposed vortex beam l=±4, and a petal-shaped intensity distribution (right) generated when it is loaded onto a spatial light modulator.

[0054] In underwater disturbance module 2, the underwater disturbance environment is composed of factors such as turbulence intensity, turbidity, obstruction, and density non-uniformity. The specific composition and working process are as follows: Figure 3As shown, a transparent glass water tank measuring 20 cm × 20 cm × 25 cm was used, and 6 liters of clean water were injected into the tank. Analytical pure kaolin (Al₂O₃·2SiO₂·2H₂O) was used as the kaolin addition unit. Low, medium, and high turbidity environments were constructed by adding 2 g, 4 g, and 6 g of kaolin, respectively. An adjustable power water pump (JINGNUO: model JN-200) was placed at the bottom of the tank. By adjusting the pump power to 1 W, 1.5 W, and 2 W, weak, medium, and strong average flow rates were generated for stirring, respectively. By combining different turbidities with different stirring intensities, nine controllable underwater disturbance environments were constructed to simulate scattering and turbulence interference in the real ocean. The controllable underwater disturbance simulation environment constructed in this embodiment can accurately simulate various real underwater interferences. The system has strong versatility and provides reliable technical support for the actual deployment of underwater optical communication systems.

[0055] The data acquisition module 3 consists of a focusing lens 31 and a CCD camera 32. After being transmitted through the underwater disturbance module 2, the conjugate superimposed vortex beam is detected by the CCD camera 32 on the focusing plane under the action of the focusing lens 31, forming a beam intensity image. Specific structure and operation: A CCD camera (MV-CS016-10UM, pixel pitch 6.9 μm) is used as the image acquisition device. The conjugate superimposed vortex beam, transmitted through the underwater disturbance environment, is incident on the target surface of the CCD camera. A region of interest of 400×400 pixels is set, and only the beam intensity distribution image within this region is acquired to reduce redundant information. For 16 different topological charge conjugate superimposed vortex modes, 5000 image samples are acquired under the above 9 disturbance environments. Some experimental samples are shown below. Figure 4 As shown. At this point, 12 sets of black rectangular occluders of preset sizes can be added. Occlusions are generated by randomly selecting positions within a 400×400 pixel ROI to simulate local occlusion caused by underwater organisms or suspended particles. This ensures that the occlusion scene is consistent with the randomness and locality of occlusion in the real underwater environment. A total of 720,000 image samples were obtained to construct a large-scale dataset. The dataset was then randomly divided into training, validation, and test sets in an 8:1:1 ratio for subsequent training, validation, and testing of deep learning models.

[0056] Data processing module 4 includes a deep learning recognition model and a display control device. The deep learning recognition model is a ResNet-50 network used to recognize the beam intensity image obtained by the data acquisition module. The display control device is used to display the recognition results, the parameters for training the deep learning recognition model, and to adjust the topological charge of the conjugate superimposed vortex beam in beam generation module 1, and supports system parameter adjustment. Wherein: The deep learning recognition model uses the ResNet-50 network as the core classification model. This network includes 7×7 convolutional layers (stride 2), 3×3 max pooling layers (stride 2), and four stages containing bottleneck residual blocks. Finally, it outputs 16 pattern recognition results through global average pooling and fully connected layers. Its specific structure and working process are as follows: The ResNet-50 network is selected as the core classification model, and its structure is as follows... Figure 5 As shown, it includes Stage 0 (7×7 convolutional layer + 3×3 max pooling layer) and four Stages (Stage 1-4) with bottleneck residual blocks; Stage 0 converts the 224×224 input image into a 56×56 feature map with 64 channels; Stages 1-4 progressively downsample the spatial dimension and increase the number of channels through bottleneck residual blocks, and finally convert the feature map into a 2048-dimensional vector through global average pooling, and output the recognition results of 16 patterns through fully connected layers and the Softmax function.

[0057] The display and control equipment uses an industrial control computer to display the following information in real time through dedicated software: 1) Pattern recognition results of the ResNet-50 network (including recognition category and confidence level); 2) Loss curve and accuracy curve during model training; 3) Operating parameters of each module (laser power, spatial light modulator phase parameters, water pump power, CCD acquisition parameters, etc.). Simultaneously, operators can manually adjust these parameters through this module to optimize the system's recognition performance in different underwater environments.

[0058] The present embodiment of a simulation method for pattern recognition in underwater vortex optical communication includes the following steps: S1. Generate a conjugate superimposed vortex beam using beam generation module 1: Turn on the 535 nm semiconductor laser, adjust the neutral density attenuator 12 to make the beam power reach the target value; adjust the polarizer 13 to make the beam linearly polarized; after the beam is expanded and collimated by the beam expansion and collimation component 14, it is incident on the spatial light modulator 15; generate a hologram of the corresponding topological charge on the computer of the data processing module, load it onto the spatial light modulator 15 to perform phase modulation on the beam; adjust the adjustable aperture 162 of the 4f spatial filter component 16 to allow only the +1st order diffraction order to pass through, and obtain the target conjugate superimposed vortex beam.

[0059] S2. Construct underwater disturbance module 2: Pour 6 liters of clean water into the transparent glass tank, add kaolin of a preset mass according to experimental requirements, and stir evenly with a stirring rod; place the adjustable power water pump at the bottom of the tank, connect the power supply and adjust it to the target power to form stable turbulence in the tank; let it stand for a period of time until the scattering medium in the tank is evenly distributed and the turbulence is stable, and the underwater disturbance environment construction is completed.

[0060] S3. Acquire intensity images and construct dataset: The conjugate superimposed vortex beam is accurately transmitted through the underwater disturbance environment and incident on the CCD camera target surface; set the acquisition parameters of the CCD camera (exposure time, gain, etc.), and set the region of interest for a 400×400 pixel area; for 16 conjugate superimposed vortex modes, acquire 5000 image samples under the current disturbance environment; change the disturbance environment (adjust the kaolin concentration or water pump power), and repeat the above acquisition process until image acquisition for all 9 disturbance environments is completed; divide the 720,000 acquired images into training set, validation set and test set in an 8:1:1 ratio.

[0061] S4. Training and Recognition of the Dataset: Input the training and validation sets into the ResNet-50 network, set the Adam optimizer, initial learning rate of 0.01, and other training parameters, and train for 25 epochs; during training, monitor the model performance through the validation set, and adjust the regularization parameters if overfitting occurs; for test samples with occlusion, generate random occlusions in step S3; input the test set (including occluded and unoccluded samples) into the trained model, obtain the pattern recognition results, and input them into the display control device.

[0062] S5. Result Viewing and Parameter Adjustment: View the recognition accuracy, loss curve, and working parameters of each module on the test set through the display control device; if the recognition accuracy does not meet expectations, adjust the hologram parameters of the spatial light modulator (optimize beam quality), water pump power (adjust turbulence intensity), CCD acquisition parameters (improve image quality), or network training parameters (optimize model performance), and re-perform beam generation, image acquisition, and model training until a satisfactory recognition effect is obtained.

[0063] Figures 6-14 The diagram shows the loss and accuracy curves of ResNet-50 under nine perturbation conditions in this embodiment of the invention. During model training, the Adam optimizer was used, and the initial learning rate was set to 0.001. Experiments determined the optimal training configuration to be 300 image samples per class and 25 epochs of training. Under these conditions, the model could achieve 100% recognition accuracy under all perturbation environments.

[0064] Figure 15 Schematic diagrams of different occlusion areas in embodiments of the present invention. Figures 16-27The recognition performance of ResNet-50 under different occlusion areas was demonstrated in the embodiments of the present invention, with 4g of kaolin (strong scattering) and a stirring power of 1.5W (medium turbulence). For scenarios with occlusion (simulating instantaneous local occlusion of underwater organisms or suspended particles), when the occlusion area is ≤200×200 pixels (accounting for 25% of the ROI), the model can converge stably and achieve 100% recognition accuracy; when the occlusion area is 250×250 pixels (accounting for 39.06% of the ROI) to 300×300 pixels (accounting for 56.25% of the ROI), the model can still achieve 100% maximum recognition accuracy, but the loss curve oscillates violently, and the convergence stability decreases; when the occlusion area reaches 350×350 pixels (accounting for 76.56% of the ROI), the model cannot converge stably, the core modal features are severely damaged, and the recognition performance degrades significantly.

[0065] The above experimental results show that the system proposed in this invention achieves a recognition accuracy close to 100% under nine different perturbation environments, demonstrating excellent generalization ability (e.g., Figures 6-14 As shown). When the occlusion area is ≤50%, the model achieves a stable recognition rate of 100%, significantly outperforming the traditional diffraction grating method (accuracy is only 75% and 40% under medium to high interference, and <10% under 30% occlusion); when the occlusion area exceeds 76.56% (350×350 pixels), the model fails to converge stably (e.g. Figure 27 (As shown). This invention effectively solves the robustness problem of optical communication pattern recognition in complex underwater environments by combining the physical properties of conjugate superimposed vortex beams with the algorithmic advantages of deep learning models, providing important technical support for the practical application of underwater optical communication systems.

[0066] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the scope of protection of the present invention.

Claims

1. A simulation system for underwater vortex optical communication pattern recognition, characterized in that: It includes a beam generation module (1), an underwater disturbance module (2), a data acquisition module (3), and a data processing module (4), wherein: The beam generation module (1) is used to generate conjugate superimposed vortex beams with different topological charge numbers and irradiate the underwater disturbance module (2). Underwater disturbance module (2) sets up the underwater disturbance environment; The data acquisition module (3) is used to capture the beam intensity image of the conjugate superimposed vortex beam after it has been transmitted through the underwater disturbance module (2); The data processing module (4) includes a deep learning recognition model and a display control device. The deep learning recognition model is used to recognize the beam intensity image obtained by the data acquisition module. The display control device is used to display the recognition results, the parameters for training the deep learning recognition model, and the topological charge of the conjugate superimposed vortex beam in the beam generation module (1).

2. The underwater vortex optical communication pattern recognition simulation system according to claim 1, characterized in that: The beam generation module is arranged along the optical path direction as follows: laser (11), attenuator (12), polarizer (13), beam expander and collimator (14), spatial light modulator (15) and 4f spatial filter (16). The laser beam (11) is attenuated by the attenuator (12) and then adjusted to linearly polarized light by the polarizer (13). The beam expansion and collimation assembly (14) consists of two convex lenses, the first lens (141) and the second lens (142), which expand and collimate the linearly polarized light into a parallel Gaussian beam. The parallel Gaussian beam is incident on the spatial light modulator (15) and the phase of the parallel Gaussian beam is modulated to generate a conjugate superimposed vortex beam containing the target topological charge. The modulated beam is incident on the 4f spatial filter assembly (16) and finally outputs a high-quality conjugate superimposed vortex beam. The 4f spatial filter assembly (16) is arranged with the third lens (161), the aperture (162) and the fourth lens (163) in sequence along the optical path.

3. A simulation system for underwater vortex optical communication pattern recognition according to claim 2, characterized in that: The laser (11) outputs a green light beam, and the laser (11) is preferably a 535 nm semiconductor laser; the spatial light modulator (15) is a pure phase type, and the 4f spatial filter component (16) is set with an adjustable aperture in the Fourier plane to isolate the +1 order diffraction order, with a diffraction efficiency of about 15%.

4. The underwater vortex optical communication pattern recognition simulation system according to claim 3, characterized in that: In the underwater disturbance module (2), the underwater disturbance environment is composed of one or more factors, such as turbulence intensity, turbidity, obstruction and density inhomogeneity.

5. A simulation system for underwater vortex optical communication pattern recognition according to claim 4, characterized in that: The method for setting the turbulence intensity in the underwater disturbance environment is as follows: weak, medium and strong turbulence are simulated by setting water pumps with power of P1, P2 and P3 respectively in the water, where P1 < P2 < P3; The method for setting the turbidity in the underwater disturbance environment is as follows: low, medium and high turbidity are simulated by adding kaolin with masses of m1, m2 and m3 to the water, respectively, where m1 < m2 < m3; The method for setting the occlusion in the underwater disturbance environment is as follows: a black occlusion is generated in the beam intensity image obtained by the data acquisition module (3) to simulate the occlusion situation in the water, wherein the area of ​​the black occlusion covers 0.25% to 76.56% of the area of ​​the beam intensity image, preferably 25% to 56.25%.

6. A simulation system for underwater vortex optical communication pattern recognition according to claim 5, characterized in that: The data acquisition module (3) consists of a focusing lens (31) and a CCD camera (32). After the conjugate superimposed vortex beam is transmitted through the underwater disturbance module (2), it is detected on the focusing plane by the CCD camera (32) under the action of the focusing lens (31) and a beam intensity image is formed.

7. A simulation system for underwater vortex optical communication pattern recognition according to claim 6, characterized in that: The CCD camera (32) sets a 400×400 pixel area as the region of interest for image acquisition, wherein a black occlusion of less than 350×350 pixels is generated in the region of interest, preferably less than 300×300 pixels, and preferably less than 200×200 pixels.

8. A simulation system for underwater vortex optical communication pattern recognition according to claim 7, characterized in that: The deep learning recognition model uses the ResNet-50 network.

9. A simulation system for underwater vortex optical communication pattern recognition according to claim 8, characterized in that: The display control device is used to display the recognition results, the loss value and accuracy during the training process, and the system working status, and to set the laser power, spatial light modulator phase parameters, water pump power, CCD camera acquisition parameters, and network training parameters.

10. A method for pattern recognition in underwater vortex optical communication, characterized in that, The identification is performed using any one of the systems described in claims 1 to 9, and the steps are as follows: S1. Use beam generation module (1) to generate conjugate superimposed vortex beams; S2. Construct an underwater disturbance module (2); S3. Acquire intensity images and construct dataset: Transmit the conjugate superimposed vortex beam generated in step S1 through the underwater disturbance module (2) constructed in step S2, and use the data acquisition module (3) to capture beam intensity distribution images in the set region of interest, and construct a dataset, dividing the dataset into training set, validation set and test set; S4. Training and recognition of the dataset: Input the training set and validation set from step S3 into the deep learning recognition model for training, use the test set to test the trained deep learning recognition model, and input the recognition results into the display control device.

Images