Dynamic calibration based multi-mode fiber wavelength multiplexed wide field imaging quality improvement method

Abstract

The application discloses a kind of multi-mode optical fiber wavelength multiplexing wide field imaging quality promotion methods based on dynamic calibration.The method comprises the following steps: collecting the output speckle pattern corresponding to each wavelength after being transmitted by multi-mode optical fiber;The speckle patterns of different wavelengths are spatially registered and spliced to form a high-dimensional composite speckle vector;PCA is used to bidirectionally compress the dimension of input image and composite speckle vector group, and a linear wavelength multiplexing inverse transmission matrix (WITM) is dynamically estimated and updated in low-dimensional space, and real-time pre-reconstruction and continuous monitoring are realized;Set up optical fiber state monitoring module to analyze pre-reconstruction frame in real time, when structural disturbance or optical fiber deformation is detected, automatically trigger in situ dynamic calibration, realize the rapid adaptive update of WITM, without retraining network.The updated WITM will guide the diffusion neural network image optimizer to iteratively denoise and detail repair the pre-reconstruction result, significantly improve the image fidelity and structure restoration ability.

Description

Technical Field

[0001] This invention belongs to the field of optical imaging and computational optics technology, and in particular, it is a method for improving the quality of wide-field imaging based on dynamic calibration multimode fiber wavelength multiplexing. Background Technology

[0002] Multimode fiber (MMF), due to its small core diameter, multiple transmission modes, and large spatial bandwidth product, is considered an ideal carrier for next-generation ultra-fine endoscopes and high-speed optical communications. However, intermodal dispersion and mode field coupling occur when light propagates in MMF. Furthermore, the fiber is susceptible to environmental disturbances (such as bending, vibration, and temperature changes), resulting in a random speckle pattern in the output light field, making it unsuitable for direct imaging. To address this issue, current technologies primarily rely on transfer matrix (TM) measurements. This method requires scanning the input modes point-by-point and measuring the output response, which is time-consuming, and the measured transfer matrix is ​​extremely sensitive to the fiber's condition. Any minute disturbance can invalidate the transfer matrix, requiring remeasurement, severely limiting its practical application. Moreover, traditional single-wavelength transfer matrix methods have limited information throughput and poor reconstruction quality in noisy environments.

[0003] In recent years, end-to-end imaging methods based on deep learning have emerged. However, these "black box" models also face problems such as poor generalization and weak perturbation resistance, typically requiring a large amount of new data to retrain the network, resulting in high costs. Due to the mean effect in the computation process, reconstructing complex, high-depth, and multi-class targets is extremely difficult. Therefore, there is an urgent need in this field for a new method for high-fidelity multimode fiber imaging that can balance high speed, high precision, and strong robustness. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of the existing technologies by providing a method for improving the quality of wide-field imaging using multimode fiber wavelength multiplexing based on dynamic calibration. The core of this method lies in utilizing speckle information from multiple wavelength channels to perform bidirectional dimensionality reduction on high-dimensional data through principal component analysis (PCA), transforming an ill-conditioned nonlinear inverse problem into a linear problem in a low-dimensional space. Efficient and high-fidelity image reconstruction is then achieved by solving a linear wavelength multiplexing inverse transfer matrix (WITM).

[0005] The technical solution for achieving the objective of this invention is as follows: On one hand, a method for improving the quality of wide-field imaging using multimode fiber wavelength multiplexing based on dynamic calibration is provided, the method comprising:

[0006] Step 1: Acquire several target images to be transmitted or imaged, and divide these images into a first target image set and a second target image set; let the initial target image set be the first target image set;

[0007] Step 2: For each target image in the initial target image set, acquire multi-wavelength speckle images using multimode fiber.

[0008] Step 3, preprocess the speckle data: For each target image, spatially register and stitch the speckle images of different wavelengths acquired in Step 2, and fuse them into a high-dimensional composite speckle vector.

[0009] Step 4, construct the wavelength multiplexing inverse transfer matrix: perform principal component analysis on each target image and its corresponding composite speckle vector in the initial target image set for bidirectional dimensionality reduction, and calculate a linear, wavelength multiplexing inverse transfer matrix in the low-dimensional latent space.

[0010] Step 5, Image Reconstruction Based on Inverse Transfer Matrix: Let the initial target image set be the second target image set, repeat steps 2 to 3 to obtain the composite speckle vector of each target image in the initial target image set, use the inverse transfer matrix obtained in step 4 to perform a linear transformation in low-dimensional space, and then reconstruct the pre-reconstructed image corresponding to each target image through the inverse transformation of principal component analysis, forming the first image dataset; each sample in the first image dataset is in the form of "fiber optic shape - target image - pre-reconstructed image";

[0011] Step 6: Change the fiber optic shape and repeat steps 2 to 3 to obtain the deformed composite speckle vectors corresponding to each target image in the initial target image set. Then, use the inverse transfer matrix obtained in step 4 to reconstruct each composite speckle vector to obtain the mismatched pre-reconstructed image of each target image under the current fiber optic condition, forming the second image dataset. Each sample in the second image dataset is in the form of "fiber optic shape - target image - mismatched pre-reconstructed image".

[0012] Step 7: Merge the first image dataset and the second image dataset to form a third image dataset;

[0013] Step 8: Based on the third image dataset, construct an optical fiber condition monitoring model to determine whether the optical fiber has undergone deformation.

[0014] Step 9: Based on the first image dataset and the neural network, construct an image enhancement model for denoising and detail restoration of the pre-reconstructed image;

[0015] Step 10: For a new target image to be transmitted or imaged, repeat steps 2 to 5 to obtain the corresponding pre-reconstructed image, which is recorded as the pre-reconstructed image to be processed, and input the pre-reconstructed image to be processed into the fiber optic condition monitoring model.

[0016] Step 11: Determine whether the output result of the fiber optic condition monitoring model is normal. If yes, proceed to step 14. Otherwise, repeat step 10 several times and determine whether multiple consecutive results are abnormal. If yes, trigger the self-calibration mechanism and then proceed to step 12. Otherwise, proceed to step 14.

[0017] Step 12: Let the initial target image set be the first target image set, and repeat steps 2 to 4 to obtain a new inverse transfer matrix;

[0018] Step 13: Following the method in Step 5, use the inverse transfer matrix obtained in Step 12 to process the composite speckle vector corresponding to the target image in Step 10 to obtain the corresponding pre-reconstructed image, update it to the pre-reconstructed image to be processed, and then input these pre-reconstructed images to be processed into the fiber optic condition monitoring model and return to execute Step 11.

[0019] Step 14: Input the pre-reconstructed image to be processed into the image enhancement model and output the high-fidelity reconstruction result.

[0020] Furthermore, in one embodiment, in step 2, at least three laser light sources of different wavelengths are used to sequentially or simultaneously illuminate the spatial light modulator to encode the target image to be transmitted or imaged into the light field; the light modulated by the spatial light modulator is coupled into the multimode fiber for transmission; at the output end of the multimode fiber, an image sensor is used to collect and store the output speckle pattern corresponding to each wavelength; the spatial light modulator projects the light carrying the target image information onto the light collimator and couples it into the multimode fiber.

[0021] Furthermore, in one embodiment, the formula for calculating the inverse transfer matrix in step 4 is:

[0022] ;

[0023] in, It is a linear mapping matrix, i.e., an inverse transfer matrix. This represents minimizing the sum of squares of all pixel errors in the matrix operation. and The low-dimensional embedding representations of the image data and the composite speckle vector after dimensionality reduction by principal component analysis, namely the compressed image features and speckle features, are respectively represented as:

[0024]

[0025]

[0026] In the formula, This is the speckle data matrix corresponding to the initial target image set. This is the image data matrix corresponding to the initial target image set. It is the dimension of the vectorized multi-wavelength speckle data. It is the dimension of the vectorized image. The total number of target images in the initial target image set, denoted as the number of training samples; the projection matrix of the speckle data matrix. and its pseudo-reversal , where m is the number of principal components retained in the speckle data matrix. Projection matrix of image data matrix and its pseudo-reversal k is the number of principal components retained in the image data matrix. .

[0027] Furthermore, in one embodiment, step 5 reconstructs the pre-reconstructed image corresponding to each target image, and the specific reconstruction formula is as follows:

[0028]

[0029] in,

[0030]

[0031] In the formula, This represents the reconstructed input target image. This represents the composite speckle vector corresponding to some new target images to be transmitted or imaged.

[0032] Furthermore, in one embodiment, step 8, which involves constructing a fiber optic condition monitoring model, specifically includes:

[0033] Step 8-1: Construct an optical fiber condition monitoring network, including a discrimination network. The discrimination network is a lightweight convolutional neural network. Its input is a standardized target image, and its output is a binary classification result of normal or abnormal. Normal means that the optical fiber condition has not been deformed, and abnormal means that the optical fiber condition has been deformed.

[0034] The discriminant network comprises three convolutional modules connected in series, a fully connected layer, and a binary classification output layer. The three convolutional modules are all 3×3 kernels, with the number of kernels being 32, 64, and 128 respectively. Each convolutional module is followed by a 2×2 max-pooling layer. The fully connected layer contains 128 neurons and is used to extract global features. The binary classification output layer uses a sigmoid activation function to output the discrimination result.

[0035] Step 8-2: Use the third image dataset to train the fiber optic condition monitoring network to obtain the trained fiber optic condition monitoring model.

[0036] Furthermore, in one embodiment, step 8-2 specifically includes the following process:

[0037] Data augmentation is performed on the images in the third image dataset to form an augmented third image dataset; the data augmentation includes random rotation, translation, cropping, scaling, and random horizontal flipping.

[0038] The fiber optic condition monitoring network is trained using the enhanced third image dataset to obtain the trained fiber optic condition monitoring model; wherein, the Adam optimization algorithm and the binary cross-entropy loss function are used during the training process.

[0039] Furthermore, in one embodiment, the image enhancement model in step 9 employs one or more of a diffusion-based generative model, a convolutional neural network, or other advanced deep learning image restoration architecture.

[0040] On the other hand, a multimode fiber wavelength multiplexing wide-field imaging quality improvement system is provided, the system comprising sequentially executing:

[0041] The first module is used to acquire several target images to be transmitted or to be imaged, and to divide these images into a first target image set and a second target image set; let the initial target image set be the first target image set;

[0042] The second module is used to acquire multi-wavelength speckle images of each target image in the initial target image set through multimode optical fiber.

[0043] The third module is used to preprocess the speckle data: for each target image, the speckle images of different wavelengths acquired by the second module are spatially registered and stitched together to form a high-dimensional composite speckle vector.

[0044] The fourth module is used to construct the wavelength multiplexing inverse transfer matrix: principal component analysis is performed on each target image in the initial target image set and its corresponding composite speckle vector to reduce the dimensionality in both directions, and a linear, wavelength multiplexing inverse transfer matrix is ​​calculated in the low-dimensional latent space.

[0045] The fifth module is used for image reconstruction based on the inverse transfer matrix: Let the initial target image set be the second target image set, and repeat the second to third modules to obtain the composite speckle vector of each target image in the initial target image set. Then, use the inverse transfer matrix obtained in the fourth module to perform a linear transformation in the low-dimensional space, and then reconstruct the pre-reconstructed image corresponding to each target image through the inverse transformation of principal component analysis to form the first image dataset; each sample in the first image dataset is in the form of "fiber optic shape - target image - pre-reconstructed image";

[0046] The sixth module is used to change the fiber optic shape. It repeats the second to third modules to obtain the deformed composite speckle vectors corresponding to each target image in the initial target image set. Then, it uses the inverse transfer matrix obtained from the fourth module to reconstruct each composite speckle vector to obtain the mismatched pre-reconstructed image of each target image under the current fiber optic condition, forming the second image dataset. Each sample in the second image dataset is in the form of "fiber optic shape - target image - mismatched pre-reconstructed image".

[0047] The seventh module is used to merge the first image dataset and the second image dataset to form the third image dataset;

[0048] The eighth module is used to construct an optical fiber condition monitoring model based on the third image dataset to determine whether the optical fiber has undergone deformation.

[0049] The ninth module is used to construct an image enhancement model for denoising and detail restoration of the pre-reconstructed image based on the first image dataset and the neural network;

[0050] The tenth module is used to repeatedly run modules two through five for a new target image to be transmitted or imaged, to obtain the corresponding pre-reconstructed image, which is denoted as the pre-reconstructed image to be processed, and input the pre-reconstructed image to be processed into the fiber optic condition monitoring model.

[0051] The eleventh module is used to determine whether the output result of the fiber optic condition monitoring model is normal. If it is, it will jump to the fourteenth module; otherwise, it will repeat the tenth module several times and determine whether the results are abnormal multiple times in a row. If so, it will trigger the self-calibration mechanism and then run the twelfth module; otherwise, it will jump to the fourteenth module.

[0052] The twelfth module is used to set the initial target image set as the first target image set, and repeat the second to fourth modules to obtain a new inverse transfer matrix.

[0053] The thirteenth module is used to process the composite speckle vector corresponding to the target image in the tenth module using the inverse transfer matrix obtained in the twelfth module, in accordance with the method in the fifth module, to obtain the corresponding pre-reconstructed image, update it to the pre-reconstructed image to be processed, and then input these pre-reconstructed images to be processed into the fiber optic condition monitoring model and return to run the eleventh module.

[0054] The fourteenth module is used to input the pre-reconstructed image to be processed into the image enhancement model and output a high-fidelity reconstruction result.

[0055] On the other hand, a multimode fiber imaging system is provided, including: a multi-wavelength laser source, a spatial light modulator, a multimode fiber, an image sensor, and a processor; the processor is configured to execute the multimode fiber wavelength multiplexing wide-field imaging quality improvement method based on dynamic calibration, wherein the processor further integrates a fiber state monitoring module and a closed-loop control module for dynamically detecting fiber state changes and triggering rapid self-calibration during imaging, thereby achieving stable and high-fidelity wide-field imaging under dynamic disturbances.

[0056] On the other hand, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the dynamic calibration-based multimode fiber wavelength multiplexing wide-field imaging quality improvement method.

[0057] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the method for improving the quality of multimode fiber wavelength multiplexing wide-field imaging based on dynamic calibration.

[0058] Compared with the prior art, the significant advantages of this invention are:

[0059] (1) Wavelength multiplexing enhances information dimension: This invention innovatively splices and fuses multi-wavelength speckle information to construct a high-dimensional composite speckle vector, achieving redundant superposition of cross-wavelength information. This strategy significantly increases the measurement dimension used for image reconstruction, fundamentally improves the well-posedness of the inverse problem of multimode fiber imaging, effectively suppresses reconstruction noise, and enhances the signal-to-noise ratio and image detail restoration capabilities.

[0060] (2) PCA Dimensionality Reduction and Physically Interpretable Model: This invention uses Principal Component Analysis (PCA) to perform bidirectional dimensionality reduction on high-dimensional input images and multi-wavelength speckle data, transforming an originally ill-conditioned and nonlinear high-dimensional inverse transport problem into a low-dimensional linear problem. The resulting Wavelength Multiplexing Inverse Transport Matrix (WITM) model combines physical interpretability with computational efficiency, avoiding the uninterpretable problems caused by the traditional deep learning "black box" structure and significantly reducing model training and computation costs.

[0061] (3) Closed-loop anti-disturbance mechanism and adaptive calibration: This invention introduces a fiber optic state monitoring and closed-loop self-calibration control mechanism for the first time in a multimode fiber wide-field imaging system. The system uses a lightweight discrimination network to detect changes in fiber transmission state (such as bending, vibration, temperature drift, etc.) in real time. When imaging distortion is detected, the system automatically triggers an in-situ rapid calibration process to update the inverse transfer matrix (WITM) parameters. This mechanism achieves dynamic adaptive compensation for environmental disturbances, enabling the system to maintain stable high-fidelity imaging performance under complex dynamic conditions, which is significantly better than traditional static calibration methods.

[0062] (4) Inherent robustness and scalability: The spectral redundancy provided by the multi-wavelength channels not only enhances the dimensionality of the model information, but also improves the resistance to noise, speckle drift and fiber perturbation at the physical level. This method has excellent scalability and can smoothly transition to color wide-field imaging. It can achieve true color reconstruction using only red, green and blue light sources without the need for additional color mapping or correction algorithms.

[0063] (5) Modular integration and hybrid computing framework: The WITM physical reconstruction module and the diffusion-based deep learning enhancement module proposed in this invention can be flexibly integrated to form a hybrid imaging paradigm of "preliminary reconstruction of the physical model + fine optimization of the generative model". This framework balances speed and quality, achieving finer noise suppression and texture restoration while maintaining high real-time performance, providing a general modular design approach for next-generation computational optical imaging.

[0064] (6) When the transmission characteristics of multimode optical fiber change due to external disturbances such as bending, vibration or temperature change, the proposed optical fiber condition monitoring model and image enhancement neural network can realize real-time perception of optical fiber condition changes and adaptive optimization of imaging results without retraining or recalibration.

[0065] (7) By introducing the wavelength reuse inverse transfer matrix (WITM) and closed-loop self-calibration mechanism, this invention effectively eliminates the model failure problem caused by fiber disturbance in traditional methods, avoids the time consumption and computational burden caused by repeated training, and thus realizes stable, efficient and high-fidelity multimode fiber imaging in dynamic environments.

[0066] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of an experimental setup for a method to improve image reconstruction quality using multimode fiber wavelength multiplexing, as described in one embodiment.

[0068] Figure 2 This is a flowchart of a method for improving the quality of wide-field imaging based on dynamic calibration using multimode fiber wavelength multiplexing in one embodiment.

[0069] Figure 3 This is a comparison chart of the results of high-fidelity reconstruction of a complex image using single-wavelength illumination in one embodiment.

[0070] Figure 4 This is a comparison of the results of high-fidelity reconstruction of a complex image in one embodiment.

[0071] Figure 5 This is a diagram illustrating the high-fidelity reconstruction result of an image under perturbed conditions in one embodiment. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0073] It should be noted that if the embodiments of the present invention involve descriptions such as "first" and "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" and "second" may explicitly or implicitly include at least one of those features. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0074] In one embodiment, combined Figure 2 A method for improving the quality of wide-field imaging using multimode fiber wavelength multiplexing based on dynamic calibration is provided, the method comprising:

[0075] Step 1: Acquire several target images to be transmitted or imaged, and divide these images into a first target image set and a second target image set; let the initial target image set be the first target image set;

[0076] Step 2: For each target image in the initial target image set, acquire multi-wavelength speckle images using multimode fiber.

[0077] Step 3, preprocess the speckle data: For each target image, spatially register and stitch the speckle images of different wavelengths acquired in Step 2, and fuse them into a high-dimensional composite speckle vector.

[0078] Step 4, construct the wavelength multiplexing inverse transfer matrix: perform principal component analysis on each target image in the initial target image set and its corresponding composite speckle vector to reduce computational complexity and improve the stability of the inverse transfer matrix. A linear, wavelength multiplexing inverse transfer matrix is ​​then calculated in the low-dimensional latent space.

[0079] Step 5, Image Reconstruction Based on Inverse Transfer Matrix: Let the initial target image set be the second target image set, repeat steps 2 to 3 to obtain the composite speckle vector of each target image in the initial target image set, use the inverse transfer matrix obtained in step 4 to perform a linear transformation in low-dimensional space, and then reconstruct the pre-reconstructed image corresponding to each target image through the inverse transformation of principal component analysis, forming the first image dataset; each sample in the first image dataset is in the form of "fiber optic shape - target image - pre-reconstructed image";

[0080] Step 6: Change the fiber optic morphology and repeat steps 2 to 3 to obtain the deformed composite speckle vectors corresponding to each target image in the initial target image set. Then, use the inverse transfer matrix obtained in step 4 to reconstruct each composite speckle vector to obtain the mismatched pre-reconstructed image (belonging to the cracked pre-reconstructed image) of each target image under the current fiber optic state, forming the second image dataset; each sample in the second image dataset is in the form of "fiber optic morphology - target image - mismatched pre-reconstructed image";

[0081] Step 7: Merge the first image dataset and the second image dataset to form a third image dataset;

[0082] Step 8: Based on the third image dataset, construct an optical fiber condition monitoring model to determine whether the optical fiber has undergone deformation.

[0083] Step 9: Based on the first image dataset and the neural network, construct an image enhancement model for denoising and detail restoration of the pre-reconstructed image;

[0084] Step 10: For a new target image to be transmitted or imaged, repeat steps 2 to 5 to obtain the corresponding pre-reconstructed image, which is recorded as the pre-reconstructed image to be processed, and input the pre-reconstructed image to be processed into the fiber optic condition monitoring model.

[0085] Step 11: Determine if the output of the fiber optic status monitoring model is normal. If yes, proceed to step 14. Otherwise, repeat step 10 several times and determine if multiple consecutive results (the number of times can be set by the user and can be 3 to 5 consecutive frames) are abnormal. If so, trigger the self-calibration mechanism and then proceed to step 12. Otherwise, proceed to step 14.

[0086] Step 12: Let the initial target image set be the first target image set, and repeat steps 2 to 4 to obtain a new inverse transfer matrix;

[0087] Step 13: Following the method in Step 5, use the inverse transfer matrix obtained in Step 12 to process the composite speckle vector corresponding to the target image in Step 10 to obtain the corresponding pre-reconstructed image, update it to the pre-reconstructed image to be processed, and then input these pre-reconstructed images to be processed into the fiber optic condition monitoring model and return to execute Step 11.

[0088] Here, the fiber optic status monitoring and updating process can be completed in about 1 minute, which is much faster than the traditional transmission matrix calibration or neural network retraining process, thus realizing closed-loop adaptive control.

[0089] Step 14: Input the pre-reconstructed image to be processed into the image enhancement model and output the high-fidelity reconstruction result.

[0090] Furthermore, in one embodiment, combined with Figure 1 In step 2, at least three laser light sources of different wavelengths are used to illuminate the spatial light modulator sequentially or simultaneously, encoding the target image to be transmitted or imaged into the light field; the light modulated by the spatial light modulator is coupled into the multimode fiber for transmission; at the output end of the multimode fiber, an image sensor is used to collect and store the output speckle pattern corresponding to each wavelength; the spatial light modulator projects the light carrying the target image information onto the light collimator and couples it into the multimode fiber.

[0091] Preferably, the laser source has three wavelengths, corresponding to red light, green light and blue light respectively, so as to achieve true color wide field imaging.

[0092] Furthermore, in one embodiment, the formula for calculating the inverse transfer matrix in step 4 is:

[0093] ;

[0094] in, It is a linear mapping matrix, i.e., an inverse transfer matrix. This represents minimizing the sum of squares of all pixel errors in the matrix operation. and The low-dimensional embedding representations of the image data and the composite speckle vector after dimensionality reduction by principal component analysis, namely the compressed image features and speckle features, are respectively represented as:

[0095]

[0096]

[0097] In the formula, This is the speckle data matrix corresponding to the initial target image set. This is the image data matrix corresponding to the initial target image set. It is the dimension of the vectorized multi-wavelength speckle data (e.g., for a spatial size of...). RGB speckle, )), It is the dimension of the vectorized image (e.g., for a spatial size of...). The image, ), The total number of target images in the initial target image set, denoted as the number of training samples; the projection matrix of the speckle data matrix. and its pseudo-reversal m is the number of principal components retained in the speckle data matrix (so that it can capture at least 95% of the cumulative energy). Projection matrix of image data matrix and its pseudo-reversal k is the number of principal components retained in the image data matrix. .

[0098] Here, the closed-form solution to the problem is:

[0099]

[0100] in, yes The Moore–Penrose pseudo-inverse approximation.

[0101] Furthermore, in one embodiment, step 5 reconstructs the pre-reconstructed image corresponding to each target image, and the specific reconstruction formula is as follows:

[0102]

[0103] in,

[0104]

[0105] In the formula, This represents the reconstructed input target image. This represents the composite speckle vector corresponding to some new target images to be transmitted or imaged.

[0106] Specifically, this refers to:

[0107] Compressing it into a low-dimensional space using speckle PCA:

[0108]

[0109] Then, the transformation is performed using the obtained low-dimensional inverse transfer matrix:

[0110]

[0111] Finally, the image is reconstructed by performing an inverse transform using the image's PCA basis:

[0112] .

[0113] The entire reconstruction process can be concisely represented as a single linear operator. Function:

[0114]

[0115]

[0116] This dual PCA framework achieves efficient inverse mapping from high-dimensional speckle input to target image reconstruction, while significantly reducing computational costs and improving robustness in noisy scenarios.

[0117] Furthermore, in one embodiment, step 8, which involves constructing a fiber optic condition monitoring model, specifically includes:

[0118] Step 8-1: Construct an optical fiber condition monitoring network, including a discrimination network. The discrimination network is a lightweight convolutional neural network. Its input is a standardized target image, and its output is a normal or abnormal binary classification result. Normal means that the optical fiber condition has not been deformed (indicating that the optical fiber condition is stable and the image structure is clear), and abnormal means that the optical fiber condition has been deformed (indicating obvious structural distortion, noise enhancement, or pattern drift).

[0119] The discriminant network comprises three convolutional modules connected in series, a fully connected layer, and a binary classification output layer. The three convolutional modules are all 3×3 kernels, with the number of kernels being 32, 64, and 128 respectively. Each convolutional module is followed by a 2×2 max-pooling layer. The fully connected layer contains 128 neurons and is used to extract global features. The binary classification output layer uses a sigmoid activation function to output the discrimination result.

[0120] Step 8-2: Use the third image dataset to train the fiber optic condition monitoring network to obtain the trained fiber optic condition monitoring model.

[0121] Preferably, in some embodiments, step 8-2 specifically includes:

[0122] Data augmentation is performed on the images in the third image dataset to form an augmented third image dataset; the data augmentation includes random rotation (range ±20 degrees), translation, cropping, scaling (range ±20%), and random horizontal flipping.

[0123] The fiber optic condition monitoring network is trained using the enhanced third image dataset to obtain the trained fiber optic condition monitoring model; wherein, the Adam optimization algorithm and the binary cross-entropy loss function are used during the training process.

[0124] Furthermore, in one embodiment, the image enhancement model in step 9 employs one or more of a diffusion-based generative model, a convolutional neural network, or other advanced deep learning image restoration architecture.

[0125] In one embodiment, a multimode fiber wavelength multiplexing wide-field imaging quality improvement system is provided, the system comprising sequentially executing:

[0126] The first module is used to acquire several target images to be transmitted or to be imaged, and to divide these images into a first target image set and a second target image set; let the initial target image set be the first target image set;

[0127] The second module is used to acquire multi-wavelength speckle images of each target image in the initial target image set through multimode optical fiber.

[0128] The third module is used to preprocess the speckle data: for each target image, the speckle images of different wavelengths acquired by the second module are spatially registered and stitched together to form a high-dimensional composite speckle vector.

[0129] The fourth module is used to construct the wavelength multiplexing inverse transfer matrix: principal component analysis is performed on each target image in the initial target image set and its corresponding composite speckle vector to reduce the dimensionality in both directions, and a linear, wavelength multiplexing inverse transfer matrix is ​​calculated in the low-dimensional latent space.

[0130] The fifth module is used for image reconstruction based on the inverse transfer matrix: Let the initial target image set be the second target image set, and repeat the second to third modules to obtain the composite speckle vector of each target image in the initial target image set. Then, use the inverse transfer matrix obtained in the fourth module to perform a linear transformation in the low-dimensional space, and then reconstruct the pre-reconstructed image corresponding to each target image through the inverse transformation of principal component analysis to form the first image dataset; each sample in the first image dataset is in the form of "fiber optic shape - target image - pre-reconstructed image";

[0131] The sixth module is used to change the fiber optic shape. It repeats the second to third modules to obtain the deformed composite speckle vectors corresponding to each target image in the initial target image set. Then, it uses the inverse transfer matrix obtained from the fourth module to reconstruct each composite speckle vector to obtain the mismatched pre-reconstructed image of each target image under the current fiber optic condition, forming the second image dataset. Each sample in the second image dataset is in the form of "fiber optic shape - target image - mismatched pre-reconstructed image".

[0132] The seventh module is used to merge the first image dataset and the second image dataset to form the third image dataset;

[0133] The eighth module is used to construct an optical fiber condition monitoring model based on the third image dataset to determine whether the optical fiber has undergone deformation.

[0134] The ninth module is used to construct an image enhancement model for denoising and detail restoration of the pre-reconstructed image based on the first image dataset and the neural network;

[0135] The tenth module is used to repeatedly run modules two through five for a new target image to be transmitted or imaged, to obtain the corresponding pre-reconstructed image, which is denoted as the pre-reconstructed image to be processed, and input the pre-reconstructed image to be processed into the fiber optic condition monitoring model.

[0136] The eleventh module is used to determine whether the output result of the fiber optic condition monitoring model is normal. If it is, it will jump to the fourteenth module; otherwise, it will repeat the tenth module several times and determine whether the results are abnormal multiple times in a row. If so, it will trigger the self-calibration mechanism and then run the twelfth module; otherwise, it will jump to the fourteenth module.

[0137] The twelfth module is used to set the initial target image set as the first target image set, and repeat the second to fourth modules to obtain a new inverse transfer matrix.

[0138] The thirteenth module is used to process the composite speckle vector corresponding to the target image in the tenth module using the inverse transfer matrix obtained in the twelfth module, in accordance with the method in the fifth module, to obtain the corresponding pre-reconstructed image, update it to the pre-reconstructed image to be processed, and then input these pre-reconstructed images to be processed into the fiber optic condition monitoring model and return to run the eleventh module.

[0139] The fourteenth module is used to input the pre-reconstructed image to be processed into the image enhancement model and output a high-fidelity reconstruction result.

[0140] Specific limitations regarding the dynamic calibration-based multimode fiber wavelength multiplexing wide-field imaging quality enhancement system can be found in the limitations of the dynamic calibration-based multimode fiber wavelength multiplexing wide-field imaging quality enhancement method described above, and will not be repeated here. Each module in the aforementioned dynamic calibration-based multimode fiber wavelength multiplexing wide-field imaging quality enhancement system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0141] In one embodiment, a multimode fiber imaging system is provided, comprising: a multi-wavelength laser source, a spatial light modulator, a multimode fiber, an image sensor, and a processor; the processor is configured to execute the multimode fiber wavelength multiplexing wide-field imaging quality improvement method based on dynamic calibration, wherein the processor further integrates a fiber state monitoring module and a closed-loop control module for dynamically detecting fiber state changes and triggering rapid self-calibration during imaging, thereby achieving stable, high-fidelity wide-field imaging under dynamic disturbances.

[0142] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements:

[0143] Step 1: Acquire several target images to be transmitted or imaged, and divide these images into a first target image set and a second target image set; let the initial target image set be the first target image set;

[0144] Step 2: For each target image in the initial target image set, acquire multi-wavelength speckle images using multimode fiber.

[0145] Step 3, preprocess the speckle data: For each target image, spatially register and stitch the speckle images of different wavelengths acquired in Step 2, and fuse them into a high-dimensional composite speckle vector.

[0146] Step 4, construct the wavelength multiplexing inverse transfer matrix: perform principal component analysis on each target image and its corresponding composite speckle vector in the initial target image set for bidirectional dimensionality reduction, and calculate a linear, wavelength multiplexing inverse transfer matrix in the low-dimensional latent space.

[0147] Step 5, Image Reconstruction Based on Inverse Transfer Matrix: Let the initial target image set be the second target image set, repeat steps 2 to 3 to obtain the composite speckle vector of each target image in the initial target image set, use the inverse transfer matrix obtained in step 4 to perform a linear transformation in low-dimensional space, and then reconstruct the pre-reconstructed image corresponding to each target image through the inverse transformation of principal component analysis, forming the first image dataset; each sample in the first image dataset is in the form of "fiber optic shape - target image - pre-reconstructed image";

[0148] Step 6: Change the fiber optic shape and repeat steps 2 to 3 to obtain the deformed composite speckle vectors corresponding to each target image in the initial target image set. Then, use the inverse transfer matrix obtained in step 4 to reconstruct each composite speckle vector to obtain the mismatched pre-reconstructed image of each target image under the current fiber optic condition, forming the second image dataset. Each sample in the second image dataset is in the form of "fiber optic shape - target image - mismatched pre-reconstructed image".

[0149] Step 7: Merge the first image dataset and the second image dataset to form a third image dataset;

[0150] Step 8: Based on the third image dataset, construct an optical fiber condition monitoring model to determine whether the optical fiber has undergone deformation.

[0151] Step 9: Based on the first image dataset and the neural network, construct an image enhancement model for denoising and detail restoration of the pre-reconstructed image;

[0152] Step 10: For a new target image to be transmitted or imaged, repeat steps 2 to 5 to obtain the corresponding pre-reconstructed image, which is recorded as the pre-reconstructed image to be processed, and input the pre-reconstructed image to be processed into the fiber optic condition monitoring model.

[0153] Step 11: Determine whether the output result of the fiber optic condition monitoring model is normal. If yes, proceed to step 14. Otherwise, repeat step 10 several times and determine whether multiple consecutive results are abnormal. If yes, trigger the self-calibration mechanism and then proceed to step 12. Otherwise, proceed to step 14.

[0154] Step 12: Let the initial target image set be the first target image set, and repeat steps 2 to 4 to obtain a new inverse transfer matrix;

[0155] Step 13: Following the method in Step 5, use the inverse transfer matrix obtained in Step 12 to process the composite speckle vector corresponding to the target image in Step 10 to obtain the corresponding pre-reconstructed image, update it to the pre-reconstructed image to be processed, and then input these pre-reconstructed images to be processed into the fiber optic condition monitoring model and return to execute Step 11.

[0156] Step 14: Input the pre-reconstructed image to be processed into the image enhancement model and output the high-fidelity reconstruction result.

[0157] For specific limitations on each step, please refer to the limitations of the multimode fiber wavelength multiplexing wide-field imaging quality improvement method based on dynamic calibration mentioned above, which will not be repeated here.

[0158] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program being implemented when executed by a processor:

[0159] Step 1: Acquire several target images to be transmitted or imaged, and divide these images into a first target image set and a second target image set; let the initial target image set be the first target image set;

[0160] Step 2: For each target image in the initial target image set, acquire multi-wavelength speckle images using multimode fiber.

[0161] Step 3, preprocess the speckle data: For each target image, spatially register and stitch the speckle images of different wavelengths acquired in Step 2, and fuse them into a high-dimensional composite speckle vector.

[0162] Step 4, construct the wavelength multiplexing inverse transfer matrix: perform principal component analysis on each target image and its corresponding composite speckle vector in the initial target image set for bidirectional dimensionality reduction, and calculate a linear, wavelength multiplexing inverse transfer matrix in the low-dimensional latent space.

[0163] Step 5, Image Reconstruction Based on Inverse Transfer Matrix: Let the initial target image set be the second target image set, repeat steps 2 to 3 to obtain the composite speckle vector of each target image in the initial target image set, use the inverse transfer matrix obtained in step 4 to perform a linear transformation in low-dimensional space, and then reconstruct the pre-reconstructed image corresponding to each target image through the inverse transformation of principal component analysis, forming the first image dataset; each sample in the first image dataset is in the form of "fiber optic shape - target image - pre-reconstructed image";

[0164] Step 6: Change the fiber optic shape and repeat steps 2 to 3 to obtain the deformed composite speckle vectors corresponding to each target image in the initial target image set. Then, use the inverse transfer matrix obtained in step 4 to reconstruct each composite speckle vector to obtain the mismatched pre-reconstructed image of each target image under the current fiber optic condition, forming the second image dataset. Each sample in the second image dataset is in the form of "fiber optic shape - target image - mismatched pre-reconstructed image".

[0165] Step 7: Merge the first image dataset and the second image dataset to form a third image dataset;

[0166] Step 8: Based on the third image dataset, construct an optical fiber condition monitoring model to determine whether the optical fiber has undergone deformation.

[0167] Step 9: Based on the first image dataset and the neural network, construct an image enhancement model for denoising and detail restoration of the pre-reconstructed image;

[0168] Step 10: For a new target image to be transmitted or imaged, repeat steps 2 to 5 to obtain the corresponding pre-reconstructed image, which is recorded as the pre-reconstructed image to be processed, and input the pre-reconstructed image to be processed into the fiber optic condition monitoring model.

[0169] Step 11: Determine whether the output result of the fiber optic condition monitoring model is normal. If yes, proceed to step 14. Otherwise, repeat step 10 several times and determine whether multiple consecutive results are abnormal. If yes, trigger the self-calibration mechanism and then proceed to step 12. Otherwise, proceed to step 14.

[0170] Step 12: Let the initial target image set be the first target image set, and repeat steps 2 to 4 to obtain a new inverse transfer matrix;

[0171] Step 13: Following the method in Step 5, use the inverse transfer matrix obtained in Step 12 to process the composite speckle vector corresponding to the target image in Step 10 to obtain the corresponding pre-reconstructed image, update it to the pre-reconstructed image to be processed, and then input these pre-reconstructed images to be processed into the fiber optic condition monitoring model and return to execute Step 11.

[0172] Step 14: Input the pre-reconstructed image to be processed into the image enhancement model and output the high-fidelity reconstruction result.

[0173] For specific limitations on each step, please refer to the limitations of the multimode fiber wavelength multiplexing wide-field imaging quality improvement method based on dynamic calibration mentioned above, which will not be repeated here.

[0174] As a specific example, the invention will be further verified and illustrated in one embodiment.

[0175] Figure 1 This is a schematic diagram of the experimental setup for the multimode fiber photonic computing anti-interference method in this embodiment, including: 1. a multi-wavelength coupled laser source; 2. a spatial light modulator; 3. a fiber collimator; 4. a multimode fiber; 5. a color camera; and 6. a computer. Input speckle is loaded onto the spatial light modulator, and the speckle signals of the corresponding channels under different wavelength illumination are acquired by the color camera.

[0176] Combination Figure 2 The present invention implements a method for improving the quality of wide-field imaging based on dynamic calibration multimode fiber wavelength multiplexing for high-fidelity reconstruction. Figure 3 This paper compares the results of high-fidelity reconstruction of complex images using the present invention with those using single-wavelength illumination. It can be seen that the pre-reconstruction results using single-wavelength illumination have low overall image reconstruction quality, with an average structural similarity (SSIM) of 0.36 and a peak signal-to-noise ratio (PSNR) of 17.34 dB on the test set. After fusing multiple wavelengths, the average structural similarity (SSIM) reaches 0.43, and the PSNR improves to 19.47 dB. This demonstrates that wavelength reuse can improve the overall image reconstruction quality and reduce the reconstruction difficulty for subsequent post-processing networks.

[0177] Figure 4 This chart compares the image reconstruction results of complex datasets before and after applying the dynamic calibration-based multimode fiber wavelength multiplexing wide-field imaging method of this invention. Additionally, comparisons using traditional neural networks and pre-reconstruction results are included as a benchmark. The first row shows the speckle image reconstruction results using a traditional neural network, with an average structural similarity (SSIM) of only 0.34 and a peak signal-to-noise ratio (PSNR) of 15.79 dB, indicating the loss of most detail information in the reconstructed image. The second row shows the pre-reconstruction results. While rapid pre-reconstruction using a physical model achieves significant detail recovery, much of this detail is obscured by noise artifacts, resulting in an average structural similarity of only 0.43 and a PSNR of 19.47 dB. In contrast, the multimode fiber wavelength multiplexing method of this invention significantly improves image reconstruction quality, resulting in well-recovered detail information, an average structural similarity of 0.73, and a PSNR of 24.79 dB.

[0178] Figure 5 The image reconstruction results using the dynamic calibration-based multimode fiber wavelength multiplexing wide-field imaging method of this invention under dynamic perturbation show that, under 11 fiber deformation states, different input images can always maintain high reconstruction quality, which is close to the reconstruction results without deformation.

[0179] As can be seen from the above, this invention has the following characteristics: Information enhancement and dynamic disturbance rejection: Through wavelength reuse and dynamic calibration, the robustness of the system to dynamic environments such as fiber disturbance and temperature changes is significantly improved; High efficiency and physical interpretability: PCA dimensionality reduction and linear WITM model realize the rapid solution of ill-conditioned inverse problems, maintaining physical interpretability and taking into account real-time performance; Real-time closed-loop control: Fiber state monitoring and in-situ dynamic calibration form a closed-loop feedback system to achieve adaptive steady-state imaging; Scalable modular structure: The front-end WITM pre-reconstruction module can be flexibly combined with various deep learning back-ends (such as diffusion models, Transformers, etc.) to further improve imaging quality; Strong color imaging compatibility: Natural true-color imaging is achieved through RGB multi-wavelength multiplexing without the need for additional correction.

[0180] In summary, this invention establishes a novel, interpretable, scalable, and robust multimode fiber wide-field imaging framework by combining physical modeling, dynamic calibration, and data-driven enhancement, which has broad application prospects in fields such as biomedical endoscopy, industrial inspection, and high-speed optical communication.

[0181] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention without departing from its spirit and scope should be included within the protection scope of the present invention.

Claims

1. A method for improving the quality of multi-mode fiber wavelength multiplexed wide-field imaging based on dynamic calibration, characterized in that, The method comprises: Step 1, collecting a plurality of target images to be transmitted or imaged, and dividing the images into a first target image set and a second target image set; let the initial target image set be the first target image set; Step 2, for each target image in the initial target image set, a multi-wavelength speckle image is collected through a multi-mode optical fiber; Step 3, pre-processing the speckle data: for each target image, the speckle images of different wavelengths collected in step 2 are spatially registered and spliced to form a high-dimensional composite speckle vector; Step 4, constructing a wavelength multiplexing inverse transmission matrix: simultaneously performing principal component analysis bidirectional dimension reduction processing on each target image in the initial target image set and its corresponding composite speckle vector, and calculating a linear, wavelength multiplexing inverse transmission matrix in a low-dimensional latent space; Step 5, image reconstruction based on the inverse transmission matrix: let the initial target image set be the second target image set, repeat steps 2 to 3 to obtain the composite speckle vector of each target image in the initial target image set, perform linear transformation in the low-dimensional space using the inverse transmission matrix obtained in step 4, and then perform inverse transformation through principal component analysis to reconstruct the pre-reconstruction image corresponding to each target image, forming a first image data set; each sample in the first image data set is in the form of "fiber morphology-target image-pre-reconstruction image"; Step 6, change the fiber morphology, repeat steps 2 to 3 to obtain the deformed composite speckle vector corresponding to each target image in the initial target image set, then use the inverse transmission matrix obtained in step 4 to reconstruct each composite speckle vector to obtain the mismatched pre-reconstruction image of each target image under the current fiber state, forming a second image data set; each sample in the second image data set is in the form of "fiber morphology-target image-mismatched pre-reconstruction image"; Step 7, fuse the first image data set and the second image data set to form a third image data set; Step 8, according to the third image data set, constructing a fiber state monitoring model for judging whether the fiber state has deformed; Step 9, according to the first image data set and a neural network, constructing an image enhancement model for denoising and detail restoration of the pre-reconstruction image; Step 10, for a new target image to be transmitted or imaged, repeat steps 2 to 5 to obtain the corresponding pre-reconstruction image, denoted as a pre-reconstruction image to be processed, and input the pre-reconstruction image to be processed into the fiber state monitoring model; Step 11, judge whether the output result of the fiber state monitoring model is normal, if yes, jump to step 14; otherwise, repeat step 10 for several times, and judge whether the results are abnormal for continuous multiple times, if yes, trigger a self-calibration mechanism, and then execute step 12, otherwise, jump to step 14; Step 12, let the initial target image set be the first target image set, repeat steps 2 to 4 to obtain a new inverse transmission matrix; Step 13, in the manner of step 5, using the inverse transmission matrix obtained in step 12 to process the composite speckle vector corresponding to the target image in step 10, obtain the corresponding pre-reconstruction image, update it as the pre-reconstruction image to be processed, and then input these pre-reconstruction images to be processed into the optical fiber state monitoring model and return to step 11; Step 14, input the pre-reconstruction image to be processed into the image enhancement model, and output a high-fidelity reconstruction result.

2. The dynamic calibration based multi-mode fiber wavelength-division multiplexed wide-field imaging quality improvement method according to claim 1, wherein, In step 2, at least three different wavelength laser light sources are used to irradiate the spatial light modulator in sequence or simultaneously, and the target image to be transmitted or imaged is encoded to the light field; The light modulated by the spatial light modulator is coupled into the multimode optical fiber for transmission; at the output end of the multimode optical fiber, an image sensor is used to collect and store the output speckle pattern corresponding to each wavelength respectively; The spatial light modulator projects the light carrying the target image information to the light collimator and couples it into the multimode optical fiber.

3. The dynamic calibration based multi-mode fiber wavelength-division multiplexed wide-field imaging quality improvement method of claim 1, wherein, The calculation formula of the inverse transmission matrix in step 4 is: ； wherein, is a linear mapping matrix, i.e. an inverse transmission matrix, denotes the minimization of the sum of squares of all pixel errors in the matrix operation, and are, respectively, low-dimensional embedding representations of the image data and the complex speckle vector after dimensionality reduction by principal component analysis, i.e. compressed image features and speckle features, respectively, and are represented as: ； ； In the formula, is the speckle data matrix corresponding to the initial target image set, is the image data matrix corresponding to the initial target image set, is the dimension of the vectorized multi-wavelength speckle data, is the dimension of the vectorized image, is the total number of target images in the initial target image set, denoted as the number of training samples; the projection matrix of the speckle data matrix and its pseudo-inverse , m is the number of principal components retained by the speckle data matrix, ; the projection matrix of the image data matrix and its pseudo-inverse , k is the number of principal components retained by the image data matrix, .

4. The method of claim 3, wherein the method is based on dynamic calibration of multi-mode fiber wavelength-division multiplexed wide-field imaging quality, further comprising: In step 5, the pre-reconstruction image corresponding to each target image is reconstructed, and the specific reconstruction formula is: ； wherein, ； In the formula, represents the reconstructed input target image, represents the complex speckle vector corresponding to some new target image to be transmitted or imaged.

5. The dynamic calibration based multi-mode fiber wavelength-division multiplexed wide-field imaging quality improvement method of claim 1, wherein, In step 8, the optical fiber state monitoring model is constructed, which specifically includes: Step 8-1, constructing an optical fiber state monitoring network, including a discrimination network, the discrimination network is a lightweight convolutional neural network, the input of which is a standardized size target image, and the output is a binary classification result of normal or abnormal; the normal means that the optical fiber state has not been deformed, and the abnormal means that the optical fiber state has been deformed; The discrimination network includes three convolution modules, a fully connected layer and a two-class output layer connected in sequence; the three convolution modules are connected in sequence, the convolution kernel size is 3x3, and the number of convolution kernels is 32, 64 and 128 respectively; each convolution module is connected with a 2x2 max pooling layer; the fully connected layer contains 128 neurons for extracting global features; the two-class output layer adopts a Sigmoid activation function to output the discrimination result; Step 8-2, training the optical fiber state monitoring network using the third image data set to obtain the trained optical fiber state monitoring model.

6. The dynamic calibration based multi-mode fiber wavelength-division multiplexed wide-field imaging quality improvement method of claim 5, wherein, The specific process of step 8-2 includes: Data augmentation is performed on the images in the third image data set to form an enhanced third image data set; the data augmentation includes random rotation, translation, shear, scaling and random horizontal flip processing; The enhanced third image data set is used to train the optical fiber state monitoring network to obtain the trained optical fiber state monitoring model; wherein, the Adam optimization algorithm and the binary cross-entropy loss function are used in the training process.

7. The dynamic calibration based multi-mode fiber wavelength-division multiplexed wide-field imaging quality improvement method of claim 1, wherein, The image enhancement model in step 9 adopts one or more of the diffusion model-based generative model, convolutional neural network or other advanced deep learning image restoration architecture.

8. A multi-mode fiber wavelength multiplexed wide-field imaging quality enhancement system based on the method of any one of claims 1 to 7, characterized in that, The system includes the following modules executed in sequence: A first module for collecting a plurality of target images to be transmitted or imaged, and dividing these images into a first target image set and a second target image set; Let the initial target image set be the first target image set; The second module is configured to collect multi-wavelength speckle images through the multi-mode optical fiber for each target image in the initial target image set; The third module is configured to pre-process the speckle data: for each target image, the speckle images of different wavelengths collected by the second module are spatially registered and spliced to form a high-dimensional composite speckle vector; The fourth module is configured to construct a wavelength multiplexing inverse transmission matrix: the principal component analysis bidirectional dimension reduction processing is performed on each target image in the initial target image set and the corresponding composite speckle vector, and a linear wavelength multiplexing inverse transmission matrix is calculated in a low-dimensional latent space; The fifth module is configured to reconstruct images based on the inverse transmission matrix: the initial target image set is taken as a second target image set, the second module to the third module are repeatedly run, the composite speckle vectors of the target images in the initial target image set are obtained, the inverse transmission matrix obtained by the fourth module is used for linear transformation in the low-dimensional space, and the pre-reconstruction images corresponding to the target images are reconstructed through the inverse transformation of the principal component analysis, to form a first image data set; each sample in the first image data set is in the form of "fiber morphology-target image-pre-reconstruction image"; The sixth module is configured to change the fiber morphology, repeatedly run the second module to the third module, obtain the deformed composite speckle vectors corresponding to the target images in the initial target image set, and then use the inverse transmission matrix obtained by the fourth module to reconstruct each composite speckle vector, to obtain the mismatched pre-reconstruction images of the target images under the current fiber state, to form a second image data set; each sample in the second image data set is in the form of "fiber morphology-target image-mismatched pre-reconstruction image"; The seventh module is configured to fuse the first image data set and the second image data set to form a third image data set; The eighth module is configured to construct a fiber state monitoring model for judging whether the fiber state is deformed, according to the third image data set; The ninth module is configured to construct an image enhancement model for denoising and detail recovery of the pre-reconstruction images, according to the first image data set and a neural network; The tenth module is configured to repeatedly run the second module to the fifth module for a new target image to be transmitted or imaged, to obtain a corresponding pre-reconstruction image, which is recorded as a to-be-processed pre-reconstruction image, and input the to-be-processed pre-reconstruction image into the fiber state monitoring model; The eleventh module is configured to judge whether the output result of the fiber state monitoring model is normal, if yes, the fourteenth module is run; Otherwise, the tenth module is repeatedly run for several times, and it is judged whether the results are abnormal for continuous multiple times, if yes, a self-calibration mechanism is triggered, and then the twelfth module is run, otherwise, the fourteenth module is run; The twelfth module is configured to take the initial target image set as a first target image set, and repeatedly run the second module to the fourth module to obtain a new inverse transmission matrix; The thirteenth module is configured to process the complex speckle vectors corresponding to the target images in the tenth module by using the inverse transmission matrix obtained by the twelfth module in the manner of the fifth module, to obtain corresponding pre-reconstructed images, and to update the pre-reconstructed images as to-be-processed pre-reconstructed images, and then input the to-be-processed pre-reconstructed images into the fiber state monitoring model and return to run the eleventh module; The fourteenth module is configured to input the to-be-processed pre-reconstructed images into the image enhancement model and output high-fidelity reconstruction results.

9. A multi-mode fiber imaging system comprising: A multi-wavelength laser source, a spatial light modulator, a multi-mode optical fiber, an image sensor, and a processor; characterized in that the processor is configured to execute the method of any one of claims 1 to 7, wherein the processor further integrates a fiber state monitoring module and a closed-loop control module for dynamically detecting fiber state changes and triggering fast self-calibration during imaging, thereby achieving stable and high-fidelity wide-field imaging under dynamic disturbance.

10. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the method of any one of claims 1 to 7. The processor executes the computer program to implement the method of any one of claims 1 to 7.

Images