A method, system and electronic device for managing a pool of image denoising models

Abstract

Some embodiments of the present application provide a method, system and electronic device for managing an image denoising model pool. The method can collect sample data containing different noises, and extract a first feature vector of the sample data by an image denoising model. The first feature vector and model information of the image denoising model are saved, the model information including model hyperparameters and model weights. The image denoising model is associated with the first feature vector and the model information to generate model association information, and a model index is generated according to the model association information. A to-be-processed image is obtained, and a second feature vector of the to-be-processed image is extracted. The similarity between the first feature vector and the second feature vector is calculated, and the target denoising model with the highest similarity is queried according to the model index, so as to perform image denoising on the to-be-processed image by the target denoising model. The method can automatically select the most suitable denoising model to perform denoising processing on the to-be-processed image, thereby improving the denoising efficiency of the model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a management method and system of an image denoising model pool and an electronic device. BACKGROUND

[0002] During the digitization and transmission of images, various forms of noise, such as additive noise, multiplicative noise, and quantization noise, may be introduced due to factors such as imaging devices and external environments. Noise can interfere with the original image signal and affect the imaging effect of the image signal. Therefore, a specific image denoising model can be used to perform denoising on the image to improve the image effect.

[0003] An image denoising model is generated by training specific sample data based on a deep learning algorithm. For images taken under different hardware devices and different lighting conditions, different training targets and different sample data need to be set during training. The denoising model trained by different training targets and sample data can be applicable to different hardware devices and lighting conditions. The hardware devices and lighting conditions applicable to a trained denoising model are limited. If the image to be processed does not meet the use range of the model, the denoising model needs to be retrained by collecting data.

[0004] However, since the amount of sample data is relatively large, the process of collecting data consumes a long time, which prolongs the preparation time of the image denoising model and reduces the denoising efficiency of the image denoising model. SUMMARY

[0005] The present application provides a management method and system of an image denoising model pool and an electronic device to solve the problem of low denoising efficiency of the image denoising model.

[0006] In a first aspect, some embodiments of the present application provide a management method of an image denoising model pool, comprising:

[0007] Collecting sample data, the sample data being image data containing noise;

[0008] Extracting a first feature vector of the sample data by an image denoising model, the first feature vector being a high-dimensional feature vector of the sample data;

[0009] Saving the first feature vector and model information of the image denoising model, the model information including model hyperparameters and model weights;

[0010] Associating the image denoising model with the first feature vector and the model information to generate model association information;

[0011] Generating a model index according to the model association information;

[0012] obtaining a to-be-processed image, and extracting a second feature vector of the to-be-processed image, the second feature vector being a high-dimensional feature vector of the to-be-processed image;

[0013] calculating a similarity between the first feature vector and the second feature vector;

[0014] querying a target denoising model according to the model index, the target denoising model being an image denoising model corresponding to the first feature vector with the highest similarity;

[0015] performing image denoising on the to-be-processed image by using the target denoising model.

[0016] With reference to the first aspect, in an implementable mode, after the sample data is collected, the method further includes: performing preprocessing on the sample data, the preprocessing including data classification, data missing value processing, and data smoothing processing; detecting image source information of the sample data, the image source information including an illumination condition and an image signal processor of the sample data; and labeling a feature tag of the sample data according to the image source information, the feature tag including an illumination condition tag and an image signal processor tag.

[0017] With reference to the first aspect, in an implementable mode, the model hyperparameters include noise variance and brightness value; and the model information further includes a model identification code, a model performance index, a model version, and a model update record, the model performance index including a model calculation speed and a model denoising amount.

[0018] With reference to the first aspect, in an implementable mode, associating the image denoising model with the first feature vector and the model information includes: detecting model hyperparameters corresponding to the first feature vector to generate a model hyperparameter set; querying a model identification code corresponding to the first feature vector; matching the first feature vector with the model identification code, and matching the model hyperparameter set with the model identification code.

[0019] With reference to the first aspect, in an implementable mode, the method further includes: setting an alternative value of the model hyperparameters, the alternative value being different from a numerical value of the model hyperparameters; performing permutation and combination on the model hyperparameters and the alternative value to serve as the model hyperparameter set; and matching the model hyperparameter set with the first feature vector.

[0020] In combination with the first aspect, in an implementable manner, the similarity between the first feature vector and the second feature vector is calculated, including: detecting an image hyperparameter corresponding to the second feature vector, the image hyperparameter being a hyperparameter of the image to be processed; querying a model hyperparameter of the first feature vector; and solving an edit distance between the image hyperparameter and the model hyperparameter based on a fuzzy matching algorithm.

[0021] In combination with the first aspect, in an implementable manner, the method further includes: detecting a use record of the image denoising model; generating a model evaluation score based on the use record, the model evaluation score being used to represent an applicable range of the image denoising model; and outputting the model evaluation score.

[0022] In the second aspect, some embodiments of the present application further provide a management system of an image denoising model pool, including a data processing module, a model pool database, a model management module, and a denoising module, wherein:

[0023] The data processing module is configured to collect sample data, the sample data being image data containing noise; and extract a first feature vector of the sample data by using an image denoising model.

[0024] The model pool database is configured to save the first feature vector and model information of the image denoising model, the model information including a model hyperparameter and a model weight.

[0025] The model management module is configured to associate the image denoising model with the first feature vector and the model information to generate model association information; generate a model index based on the model association information; obtain an image to be processed and extract a second feature vector of the image to be processed; calculate a similarity between the first feature vector and the second feature vector; and query a target denoising model according to the model index, the target denoising model being an image denoising model corresponding to the first feature vector with the highest similarity.

[0026] The denoising module is configured to perform image denoising on the image to be processed by using the target denoising model.

[0027] In combination with the second aspect, in an implementable manner, after collecting the sample data, the data processing module is further configured to perform preprocessing on the sample data, the preprocessing including data classification, data missing value processing, and data smoothing processing; detect image source information of the sample data, the image source information including an illumination condition and an image signal processor of the sample data; and label a feature tag of the sample data according to the image source information, the feature tag including an illumination condition tag and an image signal processor tag.

[0028] In a third aspect, some embodiments of the present application further provide an electronic device, comprising a processor, a memory and a bus;

[0029] The processor and the memory complete mutual communication through the bus;

[0030] The memory stores computer program instructions executable by the processor, and the processor is configured to:

[0031] Collect sample data, the sample data being image data containing noise;

[0032] Extract a first feature vector of the sample data through an image denoising model;

[0033] Save the first feature vector and model information of the image denoising model, the model information including model hyperparameters and model weights;

[0034] Correlate the image denoising model with the first feature vector and the model information to generate model correlation information;

[0035] Generate a model index according to the model correlation information;

[0036] Obtain a to-be-processed image and extract a second feature vector of the to-be-processed image;

[0037] Calculate a similarity between the first feature vector and the second feature vector;

[0038] Query a target denoising model according to the model index, the target denoising model being an image denoising model corresponding to the first feature vector with the highest similarity;

[0039] Perform image denoising on the to-be-processed image through the target denoising model.

[0040] According to the above technical solution, the method provided by some embodiments of the present application can collect sample data containing different noise, and extract a first feature vector of the sample data through an image denoising model. Then, the first feature vector and model information of the image denoising model are saved, the model information including model hyperparameters and model weights. The image denoising model is correlated with the first feature vector and the model information to generate model correlation information, and a model index is generated according to the model correlation information. Then, a to-be-processed image is obtained, and a second feature vector of the to-be-processed image is extracted. A similarity between the first feature vector and the second feature vector is calculated, and a target denoising model with the highest similarity is queried according to the model index, so as to perform image denoising on the to-be-processed image through the target denoising model. The method can automatically select the most suitable denoising model to perform denoising processing on the to-be-processed image, thereby improving the denoising efficiency of the model. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0042] Figure 1 The flowchart of the management method of the image denoising model pool provided by some embodiments of the present application is shown in the figure.

[0043] Figure 2 The flowchart of the method for saving model information provided by some embodiments of the present application is shown in the figure.

[0044] Figure 3 The flowchart of the model calibration organization and provided by some embodiments of the present application is shown in the figure.

[0045] Figure 4 The flowchart of the model evaluation and update provided by some embodiments of the present application is shown in the figure.

[0046] Figure 5 The framework diagram of the management system of the image denoising model pool provided by some embodiments of the present application is shown in the figure.

[0047] Figure 6 The structural diagram of the electronic device provided by some embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0048] In order to make the purposes, technical solutions and advantages of the exemplary embodiments of the present application clearer, the technical solutions in the exemplary embodiments of the present application will be described clearly and completely below with reference to the drawings in the exemplary embodiments of the present application. Obviously, the described exemplary embodiments are only some of the embodiments of the present application, not all the embodiments.

[0049] Image noise refers to unnecessary or redundant interference information existing in image data, such as additive noise, multiplicative noise and quantization noise, etc. Image noise will interfere with the original image signal in the image data, thereby reducing the imaging quality of the image signal. Therefore, in order to improve the imaging quality, image denoising needs to be performed on the image data. Image denoising refers to the process of reducing noise in an image, also known as image de-noising.

[0050] In some embodiments, the image can be denoised by a specific image denoising model, which is based on a deep learning algorithm and generated by training with specific sample data. For images taken under different hardware devices and different lighting conditions, different training targets need to be set and different sample data need to be collected during training.

[0051] However, the denoising model trained by different training targets and sample data is also applicable to different hardware devices and lighting conditions. The hardware devices and lighting conditions applicable to a trained denoising model are limited. If the image to be processed does not meet the use range of the model, the denoising model needs to be retrained by collecting data. Therefore, the image denoising model has certain limitations and cannot cope with images of multiple device types and lighting conditions. Moreover, since the sample data for training the image denoising model has a large amount of data, the process of collecting data consumes a long time, which prolongs the preparation time of the image denoising model and reduces the denoising efficiency of the image denoising model.

[0052] Based on the above application scenarios, in order to improve the low denoising efficiency of the image denoising model, some embodiments of the present application provide a management method of an image denoising model pool, as shown in Figure 1 The method comprises the following program steps:

[0053] S1: Collect sample data.

[0054] The sample data is image data containing noise, and the sample data is image samples with different noise and lighting conditions. By collecting sample images under multiple conditions, the application range of the image denoising model pool can be expanded.

[0055] In order to ensure the quality and accuracy of the data, in some embodiments, after collecting the sample data, the sample data is also preprocessed. The preprocessing includes data classification, data missing value processing, and smoothing processing and other data cleaning operations. The image source information of the sample data is detected. The image source information includes the lighting conditions and the image signal processor (Image Signal Processor, ISP) of the sample data. Then, the feature labels of the sample data are labeled according to the image source information, and the feature labels include the lighting condition labels and the image signal processor labels, so as to improve the accuracy of the sample data.

[0056] S2: Extract a first feature vector of the sample data by the image denoising model.

[0057] After the sample data is collected, the sample data is subjected to feature extraction by the trained image denoising model to extract a first feature vector of the sample data. The first feature vector is a high-dimensional feature vector of the sample data, that is, a high-dimensional feature vector representing the image features of the sample data.

[0058] S3: Save the first feature vector and the model information of the image denoising model.

[0059] After extracting the first feature vector from the sample data, it is necessary to store the model information of the image denoising model and the first feature vector. The first feature vector is used for matching and similarity calculation with the feature vector of the image to be processed; the model information includes model hyperparameters and model weights, where the hyperparameters are the values ​​for each model; and the model weights are loaded and applied when using the image denoising model.

[0060] To facilitate the storage of the first feature vector and the model information of the image denoising model, in some embodiments, a model pool database is also constructed, and the first feature vector and model information are stored in the model pool database. Before adding a new image denoising model, the model pool database needs to be initialized first, and the existing image denoising model data in the model pool needs to be verified in order to prepare for the new model to be added to the pool.

[0061] In some embodiments, model hyperparameters include noise variance and luminance values, such as noise variance Var based on imaging time and luminance L based on the lighting scene. Model information also includes a model identifier (ID), model performance metrics, model version, and model update history. The model performance metrics include model computation speed and model denoising efficiency. Model denoising efficiency is a quantitative measure of the model's denoising effect and can be used to evaluate the model's denoising performance.

[0062] For example, such as Figure 2 As shown, image data from various ISPs under different lighting conditions are collected, and a trained image denoising model is used to extract features from the image data to extract high-dimensional feature vectors representing image features. The model information of the image denoising model and the extracted high-dimensional feature vectors are stored in a model pool database. The model pool database stores the model ID, file name, and storage path; the model's feature vectors, used for matching and similarity calculation with the feature vectors of the image to be processed; model hyperparameters, recording the hyperparameter values ​​of each model, including noise variance Var based on imaging time and brightness L based on the lighting scene; model weights, storing the model's weight parameters for loading and application when using the model for image denoising; model performance metrics, recording the model's performance metrics during training and evaluation, such as denoising effect and computation speed; and model version and update records, recording the model's version information and update history, facilitating the management and tracking of model evolution.

[0063] S4: Associate the image denoising model with the first feature vector and model information to generate model association information.

[0064] To facilitate querying models stored in the model pool database, after saving the first feature vector and model information, the first feature vector, model information, and corresponding image denoising models are associated to generate model association information. This model association information serves as an index for searching and matching image denoising models. Therefore, for each image denoising model's corresponding first feature vector, its corresponding hyperparameter value combination is labeled and associated with the image denoising model. Then, an index for the model pool is established to organize and classify the first feature vector and hyperparameters for rapid retrieval and selection of suitable models.

[0065] To facilitate the association of the above information, the first feature vector, model information, and corresponding image denoising model can be associated by associating model IDs. Therefore, in some embodiments, when associating the image denoising model with the first feature vector and model information, the model hyperparameters corresponding to the first feature vector are detected to generate a set of model hyperparameters. The model identifier code corresponding to the first feature vector is queried, and the first feature vector is matched with the model identifier code, as is the set of model hyperparameters.

[0066] S5: Generate a model index based on model association information.

[0067] After generating model association information, an index of the image denoising model is generated using this information, so that the image denoising model can be queried and matched in the model pool database.

[0068] For example, such as Figure 3 As shown, after saving the high-dimensional feature vectors of the sample data and the model data, the models are calibrated and organized. That is, for each model's corresponding feature vector, its corresponding hyperparameter value combination is calibrated and associated with the model. Then, an index and database for the model pool are established to organize and classify the model feature vectors and hyperparameter combinations for quick retrieval and selection of suitable models.

[0069] S6: Obtain the image to be processed and extract the second feature vector of the image to be processed.

[0070] After completing the above steps, noise reduction processing can be performed on the image using the model pool of this application embodiment. An image to be processed is obtained, and feature extraction is performed on the image to obtain a second feature vector. The second feature vector is a high-dimensional feature vector of the image to be processed, representing the image features of the image.

[0071] S7: Calculate the similarity between the first feature vector and the second feature vector.

[0072] After the second feature vector of the to-be-processed image is extracted, the most suitable denoising model is selected from the model pool database through a preset model selection strategy. Similarity matching can be performed between the feature vector of the to-be-processed image and the model feature vector in the database, and the model most similar to the to-be-processed image is selected for denoising processing.

[0073] In some embodiments, when the similarity between the first feature vector and the second feature vector is calculated, the image hyperparameters corresponding to the second feature vector are detected. The image hyperparameters are the hyperparameters of the to-be-processed image. The model hyperparameters of the first feature vector are queried, and the edit distance between the image hyperparameters and the model hyperparameters is solved based on a fuzzy matching algorithm. The smaller the edit distance is, the higher the similarity between them is.

[0074] It should be noted that the above-mentioned similarity calculation method is only an exemplary description, and the management method of the image denoising model pool provided by the embodiments of the present application can also use other similarity matching algorithms. In this regard, the present application does not make any limitation.

[0075] S8: Query the target denoising model according to the model index.

[0076] The target denoising model is the image denoising model corresponding to the first feature vector with the highest similarity. After the similarity between the first feature vector and the second feature vector is calculated, the most similar image denoising model is selected according to the numerical value of the similarity. The higher the similarity is, the more suitable the image denoising model is for the to-be-processed image, and the better the corresponding denoising effect is.

[0077] S9: Perform image denoising on the to-be-processed image through the target denoising model.

[0078] After the target denoising model with the highest similarity is determined, the to-be-processed image is denoised through the model.

[0079] In order to increase the scene coverage of the image denoising model, in some embodiments, the alternative values of the model hyperparameters are also set. The alternative values are different from the numerical values of the model hyperparameters. The model hyperparameters and the alternative values are arranged and combined to form a model hyperparameter set. The model hyperparameter set is then matched with the first feature vector.

[0080] For example, the hyperparameters related to image denoising, such as noise variance Var during imaging and brightness L based on the light scene, are determined. A set of alternative values is set for each hyperparameter to cover the needs of different scenes. The combination of different hyperparameter values is labeled and matched with the first feature vector of the corresponding model, so that the most matched model can be quickly retrieved and selected according to the hyperparameter values of the to-be-processed image.

[0081] Since the ISP model of the image and the illumination condition are constantly updated, in order to ensure the scope of application of the model pool, in some embodiments, the use record of the image denoising model is also detected, and a model evaluation score is generated according to the use record. The model evaluation score is used to represent the scope of application of the image denoising model. The model evaluation score is output as a basis for the user to decide whether to update the model pool.

[0082] As shown in the example, Figure 4 The image denoising model is periodically evaluated, and according to the performance indicators and user feedback and other information, it is decided whether to update the image denoising model or whether to add a new image denoising model. This embodiment can ensure the quality and diversity of the model pool through continuous model management and updating.

[0083] Based on the above-mentioned image denoising model pool management method, some embodiments of the present application also provide an image denoising model pool management system, as shown in Figure 5 The system includes a data processing module 100, a model pool database 200, a model management module 300, and a denoising module 400, wherein:

[0084] The data processing module 100 is configured to collect sample data, the sample data being image data containing noise; and extract a first feature vector of the sample data through an image denoising model;

[0085] The model pool database 200 is configured to save the first feature vector and model information of the image denoising model, the model information including model hyperparameters and model weights;

[0086] The model management module 300 is configured to associate the image denoising model with the first feature vector and the model information to generate model association information; generate a model index according to the model association information; obtain a to-be-processed image and extract a second feature vector of the to-be-processed image; calculate the similarity between the first feature vector and the second feature vector; query a target denoising model according to the model index, the target denoising model being the image denoising model corresponding to the first feature vector with the highest similarity;

[0087] The denoising module 400 is configured to perform image denoising on the to-be-processed image through the target denoising model.

[0088] In some embodiments, the data processing module 100 is further configured to, after performing the sample data collection, perform pre-processing on the sample data, the pre-processing including data classification, data missing value processing, and data smoothing processing; detect image source information of the sample data, the image source information including illumination conditions and image signal processors of the sample data; and label feature tags of the sample data according to the image source information, the feature tags including illumination condition tags and image signal processor tags.

[0089] In some embodiments, the management system further includes a model evaluation and update module 500, the model update and evaluation module 500 being configured to detect usage records of the image denoising model; generate a model evaluation score according to the usage records, the model evaluation score being used to represent an applicable range of the image denoising model; output the model evaluation score, and perform model updating or adding through user decision.

[0090] In some embodiments, the management system further includes a hyperparameter control module 600, the hyperparameter control module 600 being configured to perform calibration and matching on model hyperparameters and first feature vectors; and obtain image hyperparameters of the image to be processed, and retrieve a target denoising model according to the image hyperparameters.

[0091] Based on the above embodiments, the following describes the management method and system of the image denoising model pool provided by the embodiments of the present application through four use cases.

[0092] Use case one, model pool construction and model entry: model pool database initialization, verification of existing model data in the model pool, and preparation for new model entry. Load sample data from multiple ISPs and illumination scenes after pre-processing and labeling. Use the trained image denoising model to extract features from the sample data to obtain high-dimensional feature vectors representing image features. Build a model database to store necessary information such as feature vectors, hyperparameter values, weights, and parameters of each model.

[0093] Use case two, model organization and management: calibrate the feature vectors of each model and associate them with corresponding hyperparameter values to establish an index and database of the model pool. The user inputs an image to be processed or selects an image to be processed from the image library provided by the system. Refer to the ISP information and illumination scene of the image to be processed to extract image features and start the model recommendation algorithm. Retrieve the feature vectors corresponding to the existing models in the model pool. Use the similarity matching method to recommend and select models from the model pool database.

[0094] Case three, hyperparameter control and organization: define the hyperparameters related to image denoising, such as variance K based on Gaussian noise during imaging, variance B based on Poisson noise during imaging, and brightness L based on the illumination scene. According to the user-provided ISP information and the illumination scene, recommend the K / B / L hyperparameter combination. The user selects the recommended hyperparameter combination while allowing the user to manually adjust the hyperparameter values. Different hyperparameter value combinations are calibrated and matched with the feature vectors of the corresponding models to quickly retrieve and select the most suitable model.

[0095] Case four, model evaluation and update: based on the usage records of the models in the model pool, quantitatively evaluate the models using standard performance indicators. The user decides whether to update the models or add new models.

[0096] Based on the above image denoising model pool management system, some embodiments of the present application also provide an electronic device 700 comprising the image denoising model pool management system described in the above embodiments.

[0097] In some embodiments, as shown in Figure 6 The electronic device 700 includes at least one processor 701, at least one communication interface 702, at least one memory 703, and at least one bus 704. The bus 704 is used to realize the direct connection communication of these components, the communication interface 702 is used for signaling or data communication with other node devices, and the memory 703 stores computer program instructions executable by the processor 701. When the electronic device 700 is running, the processor 701 communicates with the memory 703 through the bus 704, the processor 701 can call the computer program stored in the memory 703, and execute the computer program to realize the image denoising model pool management method provided by the embodiments of the present application.

[0098] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

[0099] For the sake of explanation, the foregoing descriptions have been presented in terms of specific embodiments. However, it is to be appreciated that specific embodiments described herein are not intended to limit the scope of the present application, which is defined with reference to the following claims. Various modifications and changes can be made thereto by those skilled in the art which fall within the scope of the present application as defined by the following claims. The embodiments were chosen and described in order to explain the principles of the application and the practical application and to enable others skilled in the art to understand for implementing various embodiments and with various modifications as are suited to the particular use contemplated.

Claims

1. A method for managing an image denoising model pool, characterized in that, The method comprises the following steps: collecting sample data, the sample data being image data containing noise; extracting a first feature vector of the sample data by an image denoising model, the first feature vector being a high-dimensional feature vector of the sample data; saving the first feature vector and model information of the image denoising model, the model information including model hyperparameters and model weights; associating the image denoising model with the first feature vector and the model information to generate model association information; generating a model index according to the model association information; obtaining a to-be-processed image and extracting a second feature vector of the to-be-processed image, the second feature vector being a high-dimensional feature vector of the to-be-processed image; calculating the similarity between the first feature vector and the second feature vector; querying a target denoising model according to the model index, the target denoising model being the image denoising model corresponding to the first feature vector with the highest similarity; performing image denoising on the to-be-processed image by the target denoising model.

2. The method of claim 1, wherein, After collecting the sample data, the method further comprises the following steps: performing preprocessing on the sample data, the preprocessing including data classification, data missing value processing, and data smoothing processing; detecting image source information of the sample data, the image source information including the lighting condition and image signal processor of the sample data; labeling feature labels of the sample data according to the image source information, the feature labels including lighting condition labels and image signal processor labels.

3. The method of claim 1, wherein, The model hyperparameters include noise variance and brightness value; the model information further includes model identification code, model performance index, model version, and model update record, and the model performance index includes model calculation speed and model denoising amount.

4. The method of claim 3, wherein, Associating the image denoising model with the first feature vector and the model information comprises the following steps: detecting model hyperparameters corresponding to the first feature vector to generate a model hyperparameter set; querying the model identification code corresponding to the first feature vector; matching the first feature vector with the model identification code and matching the model hyperparameter set with the model identification code.

5. The method of claim 4, wherein, The method further comprises the following steps: setting alternative values of the model hyperparameters, the alternative values being different from the values of the model hyperparameters; performing permutation and combination on the model hyperparameters and the alternative values to obtain the model hyperparameter set; matching the model hyperparameter set with the first feature vector.

6. The method of claim 1, wherein, Calculating the similarity between the first feature vector and the second feature vector comprises the following steps: detecting image hyperparameters corresponding to the second feature vector, the image hyperparameters being hyperparameters of the to-be-processed image; querying model hyperparameters of the first feature vector; solving the edit distance between the image hyperparameters and the model hyperparameters based on a fuzzy matching algorithm.

7. The method of claim 1, wherein, The method further comprises the following steps: detecting the usage record of the image denoising model; generating a model evaluation score according to the usage record, the model evaluation score being used to represent the applicable range of the image denoising model; outputting the model evaluation score.

8. A management system of a pool of image denoising models, characterized in that, The method comprises the following steps: a data processing module configured to collect sample data, the sample data being image data containing noise; extract a first feature vector of the sample data through an image denoising model; a model pool database configured to save the first feature vector and model information of the image denoising model, the model information including model hyperparameters and model weights; a model management module configured to associate the image denoising model with the first feature vector and the model information to generate model association information, generate a model index according to the model association information, obtain a to-be-processed image, extract a second feature vector of the to-be-processed image, calculate a similarity between the first feature vector and the second feature vector, and query a target denoising model according to the model index, the target denoising model being an image denoising model corresponding to the first feature vector with the highest similarity; a denoising module configured to perform image denoising on the to-be-processed image through the target denoising model.

9. The management system of a pool of image denoising models according to claim 8, characterized in that, The data processing module is further configured to, after collecting sample data: perform preprocessing on the sample data, the preprocessing including data classification, data missing value processing, and data smoothing processing; detect image source information of the sample data, the image source information including illumination conditions and image signal processors of the sample data; label feature tags of the sample data according to the image source information, the feature tags including illumination condition tags and image signal processor tags.

10. An electronic device, comprising: comprise: a processor, a memory, and a bus; the processor and the memory complete mutual communication through the bus; the memory stores computer program instructions executable by the processor, and the processor is configured to: collect sample data, the sample data being image data containing noise; extract a first feature vector of the sample data through an image denoising model; save the first feature vector and model information of the image denoising model, the model information including model hyperparameters and model weights; associate the image denoising model with the first feature vector and the model information to generate model association information; generate a model index according to the model association information; obtain a to-be-processed image, and extract a second feature vector of the to-be-processed image; calculate a similarity between the first feature vector and the second feature vector; query a target denoising model according to the model index, the target denoising model being an image denoising model corresponding to the first feature vector with the highest similarity; perform image denoising on the to-be-processed image through the target denoising model.

Images