Integrated learning method in image restoration

Abstract

The invention relates to the technical field of image processing model training, in particular to an integrated learning method in image restoration, which comprises the following steps: acquiring a real-time image and predicting a restored image; dividing barrel sets; determining a model weight; integrating the weights to form a Gaussian mixture model; and adjusting the prediction duration. According to the method, the multi-model integration and the EM algorithm are introduced, so that the image restoration process is more robust and accurate, full utilization of data on different scales is ensured by performing long vector conversion and bucket set division on the restored image, the weight distribution of each model is effectively optimized by means of the Gaussian mixture model, the detail restoration degree of the restored image is improved, and the image restoration efficiency is improved. A real-time brightness fluctuation detection and adjustment mechanism can adaptively adjust the prediction duration, effectively deals with a complex illumination change scene, and effectively solves the problems of poor recovery result generalization and large computing resource consumption caused by lack of a post-training integration method with flexibility and plug-and-play convenience.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing model training, and in particular to an integrated learning method in image restoration. Background Art

[0002] Image restoration has made remarkable progress in the past few decades, especially with the rise of deep learning. Researchers have developed many architectures to solve the image restoration problem, including convolutional neural networks (CNNs), visual transformers (ViTs), and more recently, visual state-space models (Mamba). However, a single model with different architectures or randomly initialized states will deviate from the ground truth when predicting, leading to suboptimal restoration results. To address this issue, ensemble learning, a traditional but influential machine learning technique, has been applied to image restoration, which aims to improve the generalization and robustness of the ensemble results by combining multiple base models. However, most ensemble methods in image restoration focus on ensemble learning in the training phase, which requires the ensemble strategy to be determined when training multiple models, sacrificing the flexibility of replacing models and the convenience of plug-and-play. The image restoration industry has an urgent need for advanced post-training ensemble methods, however, due to the lack of such a method, researchers still tend to use averaging as the main choice of post-training ensemble methods.

[0003] Despite such an industrial demand, post-trained ensembles in image restoration remain challenging for traditional ensemble algorithms originally designed for classification or regression problems. Unlike classification and regression, the prediction results of image restoration are matrices, where each pixel is correlated with other pixels and ranges from 0 to 255. Therefore, traditional methods such as bagging and boosting (e.g., AdaBoost) either require huge computational resources for the restoration task or fail to generalize well due to the imbalance between the number of candidate results and feature dimensions. As an alternative, Jiang et al. proposed a post-trained ensemble algorithm for super-resolution by optimizing the maximum a posteriori problem with reconstruction constraints, named RefESR. However, this constraint requires an explicit expression of the degradation process, which is unknown for other restoration tasks besides super-resolution; it also requires prior knowledge of the performance of the base models, which further limits its practicality. Due to these issues of traditional and recent ensemble methods, researchers in the field of image restoration prefer weighted averaging as the main ensemble method. Summary of the invention

[0004] To this end, the present invention provides an integrated learning method in image restoration to overcome the problems of poor generalization of restoration results and high consumption of computing resources in the prior art due to the lack of flexibility and plug-and-play convenience of post-training integrated methods.

[0005] To achieve the above object, the present invention provides an integrated learning method for image restoration, comprising:

[0006] Collect real-time images at the monitoring points of the target mission;

[0007] Predicting all the real-time images within a preset prediction time period using a plurality of preset restoration models to form a plurality of restored images;

[0008] Determine the plurality of long vectors from all restored images corresponding to each preset restoration model;

[0009] Dividing the long vector into a plurality of buckets of the same width according to the preset bucket width, and forming a plurality of mutually exclusive bucket sets according to the number of buckets and the preset number;

[0010] Determine a number of weights of the model to be constructed according to the bucket set and a preset EM algorithm;

[0011] Integrating the weights into integrated weights and storing them in a preset lookup table to form a plurality of Gaussian mixture models;

[0012] The brightness fluctuation value is determined according to the real-time brightness of all the real-time images within the preset adjustment time after forming a plurality of Gaussian mixture models, and the preset prediction time is adjusted according to the brightness fluctuation value until the brightness fluctuation value is less than the preset fluctuation value.

[0013] Further, the brightness fluctuation value is determined based on the real-time brightness of all the real-time images within the preset adjustment time after forming a plurality of Gaussian mixture models, and the brightness fluctuation value is determined based on the real-time brightness of all the real-time images within the preset adjustment time, including:

[0014] The real-time brightness of the real-time image is extracted, and the standard deviation of the real-time brightness is calculated to form a brightness fluctuation value.

[0015] Further, the adjusting the preset prediction duration according to the brightness fluctuation value until the brightness fluctuation value is less than the preset fluctuation value includes:

[0016] When the brightness fluctuation value is greater than a preset standard brightness fluctuation value, the preset prediction duration is adjusted according to the brightness fluctuation value, the preset standard brightness fluctuation value and the preset adjustment coefficient until the brightness fluctuation value is less than the preset fluctuation value.

[0017] Furthermore, the determining of several weights of the model to be constructed according to the bucket set and the preset EM algorithm includes:

[0018] Mask mapping is performed on each of the bucket sets to obtain a number of screened pixel points to form the several short vectors, and the weight of the model to be constructed is estimated according to the short vectors and a preset EM algorithm to form a number of weights.

[0019] Furthermore, estimating the weight of the model to be constructed according to the short vector and the preset EM algorithm includes:

[0020] The reference real image pixel points corresponding to the short vectors are used as sample points of the model to be constructed and the preset EM algorithm, the mean and variance of each short vector are calculated to obtain the short vector mean and short vector variance, and the weight of the model to be constructed is estimated according to the sample points, the short vector mean, the short vector variance and the preset EM algorithm.

[0021] Furthermore, estimating the weight of the model to be constructed according to the sample points, the short vector mean, the short vector variance and the preset EM algorithm includes:

[0022] Assigning the short vector mean to the Gaussian distribution mean, and assigning the short vector variance to the Gaussian distribution variance;

[0023] In step E, the posterior probability that each sample point belongs to each Gaussian component is calculated, and the weight of the Gaussian component is updated based on all the posterior probabilities to form an updated weight;

[0024] In step M, the mean and variance of each Gaussian component are recalculated based on the updated weight to form an updated mean and an updated variance;

[0025] The weight of each Gaussian component is updated according to the updated mean and the updated variance until convergence or a maximum number of steps is reached.

[0026] Further, determining the plurality of long vectors from all restored images corresponding to each preset restoration model includes:

[0027] All restored images corresponding to each preset restoration model are flattened and concatenated into a long vector to form the plurality of long vectors.

[0028] Furthermore, dividing the long vector into a plurality of buckets of the same width according to the preset bucket width includes:

[0029] The ratio of the pixel value range of the long vector to the preset bucket width is calculated to obtain the number of buckets.

[0030] Further, forming a plurality of mutually exclusive bucket sets according to the number of buckets and the preset number includes:

[0031] The number of the buckets is multiplied by the preset number to obtain the number of bucket sets, and the buckets are divided according to the number of bucket sets to form bucket sets with mutually exclusive ranges of the number of bucket sets.

[0032] Furthermore, the step of integrating the weights into integrated weights and storing them in a preset lookup table to form a plurality of Gaussian mixture models includes:

[0033] Integrating all of the weights into an integrated weight vector by weighted summation;

[0034] Normalizing the integrated weight vector to form a normalized integrated weight vector;

[0035] The normalized integrated weight vector, the Gaussian distribution mean, and the Gaussian distribution variance are stored in a preset lookup table to form a plurality of Gaussian mixture models.

[0036] Compared with the prior art, the beneficial effect of the present invention lies in that, by introducing multi-model integration and EM algorithm, the image restoration process is made more robust and accurate, and by performing long vector conversion and bucket set division on the restored image, the full utilization of data at different scales is ensured, and the weight distribution of each model is effectively optimized with the help of Gaussian mixture model, so as to improve the detail restoration of the restored image. At the same time, the detection and adjustment mechanism of real-time brightness fluctuation enables the system to have the ability to adaptively adjust the prediction duration, effectively cope with complex lighting change scenes, reduce the brightness distortion of the restored image, and improve the overall image quality. By integrating and storing the weights in a lookup table, the method also has good scalability and convenient subsequent optimization potential, and effectively solves the problems of poor generalization of restoration results and high consumption of computing resources due to the post-training integration method that lacks flexibility and plug-and-play convenience.

[0037] Furthermore, by performing mask mapping on the bucket set and obtaining effective screening pixels, it can be ensured that the formed short vector is representative, thereby improving the accuracy of the model's weight estimation.

[0038] Furthermore, by calculating the mean and variance of the short vector, the statistical characteristics of the image data can be better captured, thereby improving the accuracy of image restoration.

[0039] Furthermore, through this iterative process, the model to be constructed can gradually approach the optimal weight distribution, making the model's restoration of the image more accurate and stable.

[0040] Furthermore, by calculating the brightness fluctuation value of the real-time image, the system can dynamically monitor the impact of lighting changes, reduce the fluctuations caused by lighting changes, and make the restored image quality more stable and consistent.

[0041] Furthermore, by dynamically adjusting the prediction duration, this method can adapt to changes in lighting conditions and ensure that the image restoration model can maintain a high restoration accuracy even when the lighting fluctuates greatly.

[0042] Furthermore, by concatenating them into a long vector, the information of multiple restored images can be effectively integrated, thereby improving data availability and processing efficiency in subsequent processing.

[0043] Furthermore, this method can effectively divide the data of a long vector into multiple buckets of the same width, ensuring that each bucket can contain a similar range of pixel values.

[0044] Furthermore, the mutually exclusive design improves the accuracy of data classification, allowing subsequent model processing to more effectively identify and utilize the features of different data segments.

[0045] Furthermore, the Gaussian mixture model finally generated by integrating the integrated weight vector can effectively represent different features in the image restoration process. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of the integrated learning method in image restoration in this embodiment;

[0047] Figure 2 A flowchart of the preset EM algorithm for this embodiment;

[0048] Figure 3 A decision logic diagram for adjusting the preset prediction duration for this embodiment;

[0049] Figure 4 Flowchart for forming several Gaussian mixture models for this embodiment. DETAILED DESCRIPTION

[0050] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0052] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0053] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0054] See also Figure 1 As shown, it is a flow chart of the integrated learning method in image restoration in this embodiment;

[0055] This embodiment provides an integrated learning method for image restoration, including:

[0056] Collect real-time images at the monitoring points of the target mission;

[0057] Predicting all the real-time images within a preset prediction time period using a plurality of preset restoration models to form a plurality of restored images;

[0058] Determine the plurality of long vectors from all restored images corresponding to each preset restoration model;

[0059] Dividing the long vector into a plurality of buckets of the same width according to the preset bucket width, and forming a plurality of mutually exclusive bucket sets according to the number of buckets and the preset number;

[0060] Determine a number of weights of the model to be constructed according to the bucket set and a preset EM algorithm;

[0061] Integrating the weights into integrated weights and storing them in a preset lookup table to form a plurality of Gaussian mixture models;

[0062] The brightness fluctuation value is determined according to the real-time brightness of all the real-time images within the preset adjustment time after forming a plurality of Gaussian mixture models, and the preset prediction time is adjusted according to the brightness fluctuation value until the brightness fluctuation value is less than the preset fluctuation value.

[0063] The preset prediction time refers to the time range for image restoration or prediction. It depends on the image acquisition frequency, processing speed and real-time requirements of the target application. It is usually set to 1 second to 10 seconds, and is set to 5 seconds in this embodiment. A moderate prediction time helps to obtain sufficient image data while ensuring the real-time performance of the system, which helps to better capture dynamic changes.

[0064] The target task includes but is not limited to image restoration tasks such as deblurring and super-resolution.

[0065] The preset restoration model refers to a specific algorithm or technology used in the image restoration process. These models are designed to process input real-time image data, extract features, remove noise, or reconstruct missing image information. Common image restoration models include convolutional neural network (CNN), sparse coding model, non-local mean (NLM) method and wavelet transform. The choice of model usually depends on the characteristics of the image (such as resolution, noise type), the restoration target (such as denoising, super resolution, completion), and the available computing resources and real-time requirements. The selected model should match the application scenario to effectively improve the image quality. In practice, the number of preset restoration models is usually 3 to 5 to ensure diversity and effectiveness. In this embodiment, the preset restoration model is set to 4 types, namely convolutional neural network (CNN), sparse coding model, non-local mean (NLM) and wavelet transform model. The advantage of selecting multiple image restoration models is that the advantages of different models in specific situations can be fully utilized.

[0066] The preset bucket width is the width of each bucket when the bucket is divided in the long vector. It depends on the distribution range and feature scale of the data. It is usually set to 1 / 10 to 1 / 20 of the data range. In this embodiment, it is set to 1 / 15 of the data range. Reasonable bucket width helps to ensure uniform distribution of data and avoid sparse or overly dense data, thereby improving the training effect of the model.

[0067] The preset lookup table is a table for storing and quickly searching for calculated weights or parameters. It depends on the model complexity and computational efficiency requirements. The size of the lookup table is usually determined by the number of weights and the complexity of the model. In this embodiment, it is set to store the weight information of 10 models, which can significantly improve the efficiency of subsequent calculations, avoid repeated calculations, and improve real-time performance.

[0068] The preset fluctuation value is a threshold for judging brightness fluctuation. It depends on the characteristics of the image data and the degree of brightness change in the application scenario. It is usually set to a standard deviation of 0.05 to 0.1. In this embodiment, it is set to a standard deviation of 0.08. A reasonable fluctuation value can effectively monitor brightness changes and ensure image restoration quality.

[0069] First, real-time images of monitoring points are collected, and multiple preset restoration models are used to predict images within a specified time to generate several restored images. These restored images are then converted into long vectors and divided into buckets according to the preset width. The weights of the model are estimated based on these buckets and the EM algorithm. The weights are stored in the lookup table after weighted integration to form multiple Gaussian mixture models. Finally, the prediction duration is adjusted by detecting the brightness fluctuation value of the real-time image until the brightness fluctuation value is less than the preset value.

[0070] By introducing multi-model integration and EM algorithm, the image restoration process is made more robust and accurate. By performing long vector conversion and bucket set division on the restored image, the full utilization of data at different scales is ensured. The weight distribution of each model is effectively optimized with the help of Gaussian mixture model, which improves the detail restoration of the restored image. At the same time, the real-time brightness fluctuation detection and adjustment mechanism enables the system to have the ability to adaptively adjust the prediction duration, effectively cope with complex lighting change scenes, reduce the brightness distortion of the restored image, and improve the overall image quality. By integrating and storing the weights in a lookup table, this method also has good scalability and convenient subsequent optimization potential, which effectively solves the problems of poor generalization of restoration results and high consumption of computing resources due to the lack of flexibility and plug-and-play convenience of post-training integration methods.

[0071] Specifically, the weights of the model to be constructed are determined according to the bucket set and the preset EM algorithm, including:

[0072] Mask mapping is performed on each of the bucket sets to obtain a number of screened pixel points to form the several short vectors, and the weight of the model to be constructed is estimated according to the short vectors and a preset EM algorithm to form a number of weights.

[0073] Mask mapping is an image processing technique that selectively operates on certain pixels by applying a mask (usually a binary value or a Boolean matrix) to each pixel, while ignoring other parts that do not need to be processed. Through mask mapping, complex images can be divided into multiple regions of interest or specific pixels can be screened out for further processing. In this embodiment, the role of mask mapping is to screen the pixels in the bucket set, obtain several pixels of interest, and form a preset number of short vectors. The mask mapping process selects pixels based on certain rules or conditions (such as pixel brightness, color, position, etc.), which can reduce the computational burden and perform subsequent model estimation and weight calculation only on key pixels.

[0074] In the image restoration method, the bucket set is first masked to screen out valid pixels corresponding to each bucket set. By analyzing these pixels, short vectors matching the preset number are formed. Subsequently, these short vectors are combined with the preset EM algorithm to estimate the weights of the model to be constructed, and finally a number of weights are obtained. These weights will be used for the subsequent construction and optimization of the image restoration model to ensure that the model can effectively process image data.

[0075] By performing mask mapping on the bucket set and obtaining effective screening pixels, it is possible to ensure that the formed short vector is representative, thereby improving the accuracy of the model's weight estimation.

[0076] Specifically, estimating the weight of the model to be constructed according to the short vector and the preset EM algorithm includes:

[0077] The reference real image pixel points corresponding to the short vectors are used as sample points of the model to be constructed and the preset EM algorithm, the mean and variance of each short vector are calculated to obtain the short vector mean and short vector variance, and the weight of the model to be constructed is estimated according to the sample points, the short vector mean, the short vector variance and the preset EM algorithm.

[0078] The reference real image pixel point refers to the real color value or intensity value of each pixel in the standard image used as a reference or comparison during the image restoration or reconstruction process. These pixel points usually represent the image features under ideal conditions and are used to evaluate and guide the performance of the restoration algorithm. Generally, in image restoration tasks, the reference real image pixel points are used to evaluate the quality of the restoration results. In this embodiment, the reference real image pixel points are determined by manually annotated or automatically generated high-quality image data.

[0079] The weight of the model to be constructed is estimated based on the short vector and the preset EM algorithm. First, the pixel points of the benchmark real image corresponding to the short vector are used as the sample points of the model to be constructed, and the mean and variance of each short vector are calculated. Then, the weight is iteratively estimated using the EM algorithm through these sample points, the mean and variance of the short vector, and finally the weight of the model to be constructed is obtained. This process ensures that the model weight can be dynamically adjusted and adapted to different image features.

[0080] By calculating the mean and variance of short vectors, the statistical characteristics of image data can be better captured, thereby improving the accuracy of image restoration.

[0081] Please continue reading Figure 2 As shown, it is a flow chart of the preset EM algorithm in this embodiment;

[0082] Specifically, estimating the weight of the model to be constructed according to the sample points, the short vector mean, the short vector variance and the preset EM algorithm includes:

[0083] Assigning the short vector mean to the Gaussian distribution mean, and assigning the short vector variance to the Gaussian distribution variance;

[0084] In step E, the posterior probability that each sample point belongs to each Gaussian component is calculated, and the weight of the Gaussian component is updated based on all the posterior probabilities to form an updated weight;

[0085] In step M, the mean and variance of each Gaussian component are recalculated based on the updated weight to form an updated mean and an updated variance;

[0086] The weight of each Gaussian component is updated according to the updated mean and the updated variance until convergence or a maximum number of steps is reached.

[0087] The weights of the model to be constructed are estimated by the EM algorithm. First, the mean of the short vector is assigned to the mean of the Gaussian distribution, and the variance is assigned to the variance of the Gaussian distribution. In the E step, the posterior probability that each sample point belongs to each Gaussian component is calculated, and the weights of each Gaussian component are updated based on these posterior probabilities to form updated weights. In the M step, the mean and variance of each Gaussian component are recalculated according to the updated weights, and the process is continued to iterate until the mean, variance and weight of the Gaussian component converge or the maximum number of iterations is reached.

[0088] Through this iterative process, the model to be constructed can gradually approach the optimal weight distribution, making the model's image restoration more accurate and stable.

[0089] Specifically, the brightness fluctuation value is determined based on the real-time brightness of all the real-time images within the preset adjustment time after forming a plurality of Gaussian mixture models, and the brightness fluctuation value is determined based on the real-time brightness of all the real-time images within the preset adjustment time, including:

[0090] The real-time brightness of the real-time image is extracted, and the standard deviation of the real-time brightness is calculated to form a brightness fluctuation value.

[0091] Within the preset adjustment time, all the real-time images collected are analyzed for brightness. First, the brightness information of each frame is extracted from these real-time images, and the brightness value of each frame is calculated. Then, based on these brightness values, the standard deviation of the brightness is calculated, which measures the degree of fluctuation of the image brightness change. Finally, the standard deviation is used as the brightness fluctuation value to reflect the stability or fluctuation of the image brightness.

[0092] By calculating the brightness fluctuation value of the real-time image, the system can dynamically monitor the impact of lighting changes, reduce the fluctuations caused by lighting changes, and make the restored image quality more stable and consistent.

[0093] Please continue reading Figure 3 As shown, it is a decision logic diagram for adjusting the preset prediction time in this embodiment;

[0094] Specifically, the adjusting the preset prediction duration according to the brightness fluctuation value until the brightness fluctuation value is less than the preset fluctuation value includes:

[0095] When the brightness fluctuation value is greater than the preset standard brightness fluctuation value, the preset prediction duration is adjusted according to the brightness fluctuation value, the preset standard brightness fluctuation value and the preset adjustment coefficient until the brightness fluctuation value is less than the preset fluctuation value, wherein the adjusted preset prediction duration is calculated by multiplying the current prediction duration by an adjustment factor. This adjustment factor is based on the difference between the current brightness fluctuation value and the preset standard brightness fluctuation value and the preset adjustment coefficient. If the current brightness fluctuation value exceeds the standard value, the prediction duration will be increased accordingly so that the system can better adapt to environmental changes, thereby improving the effect of image restoration.

[0096] The preset adjustment coefficient is a parameter used to adjust the relationship between the prediction duration and the brightness fluctuation value. Its function is to control the adjustment range of the prediction duration when the brightness fluctuation value exceeds the preset standard. Generally, the selection of the preset adjustment coefficient depends on the needs of the specific application and the system's ability to respond to brightness changes. It is usually set between 0.1 and 1, and is set to 0.5 in this embodiment to adapt to most common scenarios and provide good responsiveness and stability.

[0097] When the brightness fluctuation value is greater than the preset standard brightness fluctuation value, the system dynamically adjusts the prediction duration based on the current brightness fluctuation value, the preset standard brightness fluctuation value, and a preset adjustment coefficient. This adjustment process will continue until the brightness fluctuation value is less than the preset fluctuation value, indicating that the lighting change tends to be stable and the adjustment process ends.

[0098] By dynamically adjusting the prediction duration, this method can adapt to changes in lighting conditions and ensure that the image restoration model can maintain high restoration accuracy even when the lighting fluctuates greatly.

[0099] Specifically, determining the plurality of long vectors from all restored images corresponding to each preset restoration model includes:

[0100] All restored images corresponding to each preset restoration model are flattened and concatenated into a long vector to form the plurality of long vectors.

[0101] First, all restored images corresponding to each preset restoration model are flattened, that is, the two-dimensional image data is converted into a one-dimensional vector for subsequent operations. Then, these flattened image vectors are spliced ​​according to the preset number to form several long vectors. These long vectors serve as the basic data for subsequent processing, which helps the system better analyze and process the restoration effect.

[0102] By concatenating them into long vectors, the information of multiple restored images can be effectively integrated, thereby improving data availability and processing efficiency in subsequent processing.

[0103] Specifically, dividing the long vector into a plurality of buckets of the same width according to the preset bucket width includes:

[0104] The ratio of the pixel value range of the long vector to the preset bucket width is calculated to obtain the number of buckets.

[0105] First, we need to calculate the range of pixel values ​​in the long vector, that is, the difference between the maximum and minimum values. Then, we calculate the ratio of the range to the preset bucket width to determine the number of buckets that can be divided. This calculation process ensures that each bucket can evenly cover the pixel values ​​in the long vector, so that the data can be classified and analyzed more accurately in subsequent processing.

[0106] In this way, the data of a long vector can be effectively divided into multiple buckets of the same width, ensuring that each bucket can contain a similar range of pixel values.

[0107] Specifically, forming a plurality of mutually exclusive bucket sets according to the number of buckets and the preset number includes:

[0108] The number of the buckets is multiplied by the preset number to obtain the number of bucket sets, and the buckets are divided according to the number of bucket sets to form bucket sets with mutually exclusive ranges of the number of bucket sets.

[0109] First, the number of buckets is calculated to the power of the preset number to obtain the number of bucket sets that need to be generated. This calculation ensures that each bucket set has sufficient diversity and richness to cover different data features. Next, based on the calculated number of bucket sets, the previously divided buckets are further subdivided to ensure that the ranges between each bucket set are mutually exclusive, that is, there is no overlap.

[0110] The mutually exclusive design improves the accuracy of data classification, allowing subsequent model processing to more effectively identify and utilize the features of different data segments.

[0111] Please continue reading Figure 4 As shown, it is a flow chart of forming several Gaussian mixture models in this embodiment;

[0112] Specifically, the weights are integrated into integrated weights and stored in a preset lookup table to form a plurality of Gaussian mixture models, including:

[0113] Integrating all of the weights into an integrated weight vector by weighted summation;

[0114] Normalizing the integrated weight vector to form a normalized integrated weight vector;

[0115] The normalized integrated weight vector, the Gaussian distribution mean, and the Gaussian distribution variance are stored in a preset lookup table to form a plurality of Gaussian mixture models.

[0116] First, an integrated weight vector is integrated by weighted summing of all weights. This step ensures that the contribution of each weight can be combined into a whole, so that the model can take into account the influence of each part when processing. Next, the integrated weight vector is normalized to ensure that the sum of all weights is 1, thus forming a normalized integrated weight vector. This step is crucial for subsequent probability models because it ensures that the weight of each Gaussian component is reasonable in the entire model. Finally, the normalized integrated weight vector, the mean and variance of the Gaussian distribution are stored in a preset lookup table to form several Gaussian mixture models.

[0117] The Gaussian mixture model finally generated by integrating the integrated weight vector can effectively represent different features in the image restoration process.

[0118] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0119] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An ensemble learning method for image restoration, characterized in that: include: Collect real-time images at the monitoring points of the target mission; Predicting all the real-time images within a preset prediction time period using a plurality of preset restoration models to form a plurality of restored images; Determine the plurality of long vectors from all restored images corresponding to each preset restoration model; Dividing the long vector into a plurality of buckets of the same width according to the preset bucket width, and forming a plurality of mutually exclusive bucket sets according to the number of buckets and the preset number; Determine a number of weights of the model to be constructed according to the bucket set and a preset EM algorithm; Integrating the weights into integrated weights and storing them in a preset lookup table to form a plurality of Gaussian mixture models; The brightness fluctuation value is determined according to the real-time brightness of all the real-time images within the preset adjustment time after forming a plurality of Gaussian mixture models, and the preset prediction time is adjusted according to the brightness fluctuation value until the brightness fluctuation value is less than the preset fluctuation value.

2. The ensemble learning method for image restoration according to claim 1, characterized in that: The determining of the brightness fluctuation value based on the real-time brightness of all the real-time images within the preset adjustment time after forming a plurality of Gaussian mixture models, wherein the real-time brightness of all the real-time images within the preset adjustment time is determined to be the brightness fluctuation value comprises: The real-time brightness of the real-time image is extracted, and the standard deviation of the real-time brightness is calculated to form a brightness fluctuation value.

3. The ensemble learning method for image restoration according to claim 2, characterized in that: The step of adjusting the preset prediction duration according to the brightness fluctuation value until the brightness fluctuation value is less than the preset fluctuation value includes: When the brightness fluctuation value is greater than a preset standard brightness fluctuation value, the preset prediction duration is adjusted according to the brightness fluctuation value, the preset standard brightness fluctuation value and the preset adjustment coefficient until the brightness fluctuation value is less than the preset fluctuation value.

4. The ensemble learning method for image restoration according to claim 3, characterized in that: The step of determining the weights of the model to be constructed according to the bucket set and the preset EM algorithm includes: Mask mapping is performed on each of the bucket sets to obtain a number of screened pixel points to form the several short vectors, and the weights of the model to be constructed are estimated according to the short vectors and a preset EM algorithm to form a number of weights.

5. The ensemble learning method for image restoration according to claim 4, characterized in that: The estimating the weight of the model to be constructed according to the short vector and the preset EM algorithm includes: The reference real image pixel points corresponding to the short vectors are used as sample points of the model to be constructed and the preset EM algorithm, the mean and variance of each short vector are calculated to obtain the short vector mean and short vector variance, and the weight of the model to be constructed is estimated according to the sample points, the short vector mean, the short vector variance and the preset EM algorithm.

6. The ensemble learning method for image restoration according to claim 5, characterized in that: The estimating the weight of the model to be constructed according to the sample points, the short vector mean, the short vector variance and the preset EM algorithm includes: Assigning the short vector mean to the Gaussian distribution mean, and assigning the short vector variance to the Gaussian distribution variance; In step E, the posterior probability that each sample point belongs to each Gaussian component is calculated, and the weight of the Gaussian component is updated based on all the posterior probabilities to form an updated weight; In step M, the mean and variance of each Gaussian component are recalculated based on the updated weight to form an updated mean and an updated variance; The weight of each Gaussian component is updated according to the updated mean and the updated variance until convergence or a maximum number of steps is reached.

7. The ensemble learning method for image restoration according to claim 6, characterized in that: The step of determining the plurality of long vectors from all restored images corresponding to each preset restoration model comprises: All restored images corresponding to each preset restoration model are flattened and concatenated into a long vector to form the plurality of long vectors.

8. The ensemble learning method for image restoration according to claim 7, characterized in that: The step of dividing the long vector into a plurality of buckets of the same width according to the preset bucket width includes: The ratio of the pixel value range of the long vector to the preset bucket width is calculated to obtain the number of buckets.

9. The ensemble learning method for image restoration according to claim 8, characterized in that: The forming of a plurality of mutually exclusive bucket sets according to the number of buckets and the preset number includes: The number of the buckets is multiplied by the preset number to obtain the number of bucket sets, and the buckets are divided according to the number of bucket sets to form bucket sets with mutually exclusive ranges of the number of bucket sets.

10. The ensemble learning method in image restoration according to claim 9, characterized in that: The step of integrating the weights into integrated weights and storing them in a preset lookup table to form a plurality of Gaussian mixture models comprises: Integrating all of the weights into an integrated weight vector by weighted summation; Normalizing the integrated weight vector to form a normalized integrated weight vector; The normalized integrated weight vector, the Gaussian distribution mean, and the Gaussian distribution variance are stored in a preset lookup table to form a plurality of Gaussian mixture models.

Images