Wavelet transform-based image semi-supervised learning method and related device thereof

Abstract

This application belongs to the field of image processing technology and relates to a semi-supervised image learning method based on wavelet transform. The method includes: acquiring an initial visual image; decomposing and processing the initial visual image based on wavelet transform to obtain a multi-level image combination; filtering the multi-level image combination; transforming and amplifying the filtered multi-level image combination based on inverse wavelet transform to obtain a target visual image training set; and training the target visual image training set based on semi-supervised learning to obtain a target visual image model. This application also provides a wavelet transform-based semi-supervised image learning system, a computer device, and a storage medium thereof.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a wavelet transform-based image semi-supervised learning method, system, computer device and storage medium. Background Technology

[0002] Deep learning models (DNNs) have demonstrated outstanding performance in various applications of computer vision, but a significant drawback is the need for massive amounts of data to train their parameters. In real-world dynamic applications, especially in industrial AI inspection where products are rapidly replaced, and given the limited manpower and time available for sample labeling, there is an urgent need to train high-performance deep learning models using a limited number of labeled samples and a large number of unlabeled samples. This is the essence of semi-supervised model training methods.

[0003] In semi-supervised model training, a limited number of labeled samples leads to insufficient model training. Traditional solutions include data augmentation, such as adding random noise, image flipping, or color adjustments. However, these data augmentations randomly process the image and distort its content features, thus failing to guarantee improved model performance in practical applications. Summary of the Invention

[0004] The purpose of this application is to propose a wavelet transform-based image semi-supervised learning method, system, computer device, and storage medium to solve the technical problem that data amplification cannot solve the problem of insufficient model training due to a small number of labeled samples during the semi-supervised model training process.

[0005] To address the aforementioned technical problems, this application provides a wavelet transform-based semi-supervised image learning method, employing the following technical solution: The method includes the following steps:

[0006] Obtain the initial visual image;

[0007] The initial visual image is processed by wavelet transform decomposition, and a multi-level image combination is obtained.

[0008] Filter multi-level image combinations and use wavelet inverse transform to amplify the filtered multi-level image combinations to obtain the target visual image training set;

[0009] The target visual image model is obtained by training the target visual image training set using semi-supervised learning.

[0010] Furthermore, the step of processing the initial visual image based on wavelet transform decomposition and obtaining a multi-level image combination includes:

[0011] The initial visual image is decomposed into a combination of wavelet images with different wavelets based on wavelet transform;

[0012] Wavelet images are combined by scaling and shifting their positions to form multi-level image combinations.

[0013] Furthermore, the step of filtering multi-level image combinations and obtaining the target visual image training set by transforming and amplifying the filtered multi-level image combinations based on inverse wavelet transform includes:

[0014] The lowest low-frequency image and the highest high-frequency image are selected from the multi-level image combination, and the lowest low-frequency image and the highest high-frequency image are discarded to obtain the selected multi-level image combination.

[0015] The target visual image training set is obtained by combining multi-level images after wavelet inverse transform amplification and screening.

[0016] Furthermore, the step of obtaining the target visual image training set by combining multi-level images after wavelet inverse transform amplification and filtering includes:

[0017] The filtered multi-level image combinations are divided into low-frequency combinations and high-frequency combinations;

[0018] Based on wavelet inverse transform, the low-frequency combination and the high-frequency combination are converted into low-frequency training set and high-frequency training set, respectively;

[0019] The low-frequency training set is added to the high-frequency training set to obtain the target visual image training set.

[0020] Furthermore, the step of obtaining the target visual image model by training the target visual image training set based on semi-supervised learning includes:

[0021] The learning model is obtained by training a set of target visual images using semi-supervised learning, and then iteratively updated.

[0022] The target visual image model is obtained by weighting and combining the parameters of the iterated student model.

[0023] Furthermore, after the step of obtaining the target visual image model based on the training set of target visual images using semi-supervised learning, the method further includes:

[0024] Based on semi-supervised learning, a deep neural network model is obtained by training a training set of target visual images in the time domain.

[0025] The deep neural network module is subjected to semi-supervised deep neural network learning to obtain threads;

[0026] Generate the target visual image model based on the thread.

[0027] Furthermore, after the step of performing semi-supervised deep neural network learning on the deep neural network module and obtaining the thread, the method further includes:

[0028] Calculate the weighting matrix of the training set of target visual images;

[0029] The target visual image training set is obtained by combining the weighted matrix and thread calculation.

[0030] To address the aforementioned technical problems, this application also provides an image semi-supervised learning system based on wavelet transform, comprising:

[0031] The acquisition module is used to acquire the initial visual image;

[0032] The decomposition module is used to decompose the initial visual image based on wavelet transform and obtain a multi-level image combination.

[0033] The filtering module is used to filter multi-level image combinations and to obtain the target visual image training set by transforming and amplifying the filtered multi-level image combinations based on inverse wavelet transform.

[0034] The generation module is used to obtain a target visual image model by training a set of target visual images based on semi-supervised learning.

[0035] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution: it includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the wavelet transform-based image semi-supervised learning method described above.

[0036] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the following technical solution: the computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the wavelet transform-based image semi-supervised learning method as described above.

[0037] Compared with existing technologies, the embodiments of this application have the following main advantages: By acquiring an initial visual image, processing the initial visual image based on wavelet transform decomposition, and obtaining a multi-level image combination, filtering the multi-level image combination, and amplifying the filtered multi-level image combination based on inverse wavelet transform to obtain a target visual image training set, a target visual image model is obtained by training the target visual image training set based on semi-supervised learning. The multi-level representation based on wavelet transform generates multiple target visual image training sets containing different frequency information and quality, improving the diversity of visual image features. Without destroying the structural integrity of the visual image content, the obtained model has better generalization performance. Furthermore, the optimization and fusion of multiple semi-supervised training processes result in faster convergence and better performance of the obtained model. Attached Figure Description

[0038] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0040] Figure 2 This is a flowchart of an embodiment of the wavelet transform-based image semi-supervised learning method of this application;

[0041] Figure 3 This is a schematic diagram of the structure of an embodiment of the wavelet transform-based image semi-supervised learning system of this application;

[0042] Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0044] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0045] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0046] like Figure 1As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0047] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0048] Terminal devices 101, 102, and 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, USB flash drives, cameras, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers, etc.

[0049] Server 105 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal devices 101, 102, and 103.

[0050] It should be noted that the wavelet transform-based image semi-supervised learning method provided in this application embodiment is generally executed by a server, and correspondingly, the wavelet transform-based image semi-supervised learning system is generally set up in the server.

[0051] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0052] Continue to refer to Figure 2 The flowchart illustrates an embodiment of a wavelet transform-based image semi-supervised learning method according to this application. The wavelet transform-based image semi-supervised learning method includes the following steps:

[0053] Step S201: Obtain the initial visual image;

[0054] The initial visual image primarily originates from any of the following sources: existing images on the internet, images in locally stored computer files, images stored on cloud servers, or images stored on terminal devices. Terminal devices specifically refer to smartphones, tablets, USB flash drives, and cameras. The image interacts with the server via wired or wireless communication or a network from these sources, enabling the server to acquire the initial visual image.

[0055] Step S202: The initial visual image is processed based on wavelet transform decomposition to obtain a multi-level image combination;

[0056] It's important to note that a wavelet refers to a wave whose energy is highly concentrated in the time domain. Its energy is finite, concentrated around a specific point, and its integral value is zero, indicating that it is an orthogonal wave, similar to a Fourier wave. Wavelet transform decomposes an image signal into a set of wavelets resulting from the original wavelet's shift and scaling. Wavelets are called "image microscopes" in image processing because their multi-resolution decomposition capability can peel away layers of image information, using low-pass and high-pass filters. Wavelet transform uses a series of wavelets at different scales to decompose the original function, yielding the coefficients of the original function at different wavelet scales. Different wavelets are decomposed through translation and scaling transformations; translation yields the time characteristics of the original function, while scaling yields its frequency characteristics.

[0057] Specifically, based on wavelet transform, the initial visual image is decomposed into a combination of wavelet images of different wavelets. The wavelet image combination is then scaled and shifted to form a multi-level image combination. The initial part of the wavelet w(t) and the original function f(t) is compared, and the coefficient C is calculated. Coefficient C represents the similarity between this part of the function and the wavelet w(t). The wavelet w(t) is shifted right (left) by k units to obtain the wavelet w(tk). The comparison between the initial part of the wavelet w(tk) and the original function f(t) is repeated until the function f ends. The wavelet w(t) is expanded to obtain the wavelet w(t / 2). The comparison and right (left) shift steps are repeated until the function f ends. The wavelet w(t) is continuously expanded, and the above steps are repeated.

[0058] Step S203: Filter multi-level image combinations and obtain the target visual image training set by transforming and amplifying the filtered multi-level image combinations based on wavelet inverse transform.

[0059] It should be noted that the inverse wavelet transform is mainly based on the principle of wavelet transform. The wavelet decomposition result represents the low-frequency main components of the original visual image, as well as the high-frequency classifications in the horizontal, vertical, and diagonal directions. The original visual image is then recovered without distortion using the inverse wavelet transform based on these four classifications. In the visual image reconstruction process, the low-resolution original visual image, because it retains complete basic image information, can be used as the low-frequency component obtained from the wavelet transform decomposition of the desired visual image, while its corresponding high-frequency components can be obtained by applying a high-pass filter.

[0060] However, in this embodiment, before the inverse wavelet transform, the multi-level image combination needs to be screened. The lowest low-frequency image and the highest high-frequency image in the multi-level image combination are screened and discarded, resulting in a screened multi-level image combination. The screened multi-level image combination is then amplified based on the inverse wavelet transform to obtain the target visual image training set. After multiple wavelet transform decompositions, the visual image yields low-frequency and high-frequency components at different levels. The highest high-frequency component or the lowest low-frequency component is discarded. The inverse wavelet transform then yields an approximate visual image of the original visual image, essentially adding a noise perturbation to the original visual image and obtaining an amplified version of the visual image. For example, for each visual image in the training set D, assuming an L-level wavelet transform decomposition result, 2... L+1 -1 high-frequency component. The strategy for discarding high-frequency components can be as follows: Option 1 - Randomly select the Kth layer, K≤L, and randomly discard S high-frequency components, 0 <S<2 K+1 -1. K and S are adjustable parameters; Scheme 2 - Randomly select K layers, such as layers 1 and 3, and discard some high-frequency components in each layer. After inverse wavelet transform, obtain M training sets Ω. m ,1≤m≤M.

[0061] In this embodiment, the inverse wavelet transform method includes dividing the filtered multi-level image combination into low-frequency and high-frequency combinations; converting the low-frequency and high-frequency combinations into low-frequency and high-frequency training sets respectively based on the inverse wavelet transform; and adding the low-frequency and high-frequency training sets to obtain the target visual image training set. Specifically, each column of the transformation result is subjected to an inverse wavelet transform, and then each row of the transformed data is subjected to a one-dimensional inverse wavelet transform to reconstruct the target visual image. The wavelet transform decomposition of the visual image is a process of separating the signal from low frequency to high frequency. During the decomposition process, further wavelet decomposition can be performed on the obtained low-frequency components as needed until the requirements are met. For example, the multi-level image combination is divided into two sets of column vectors: low-frequency and high-frequency combinations. The column wavelet inverse transforms are performed on the low-frequency and high-frequency combination column vectors respectively to obtain the low-frequency and high-frequency training sets. The low-frequency and high-frequency training sets are then added together, and the visual image is obtained after a forward wavelet transform.

[0062] Step S204: Obtain the target visual image model by training the target visual image training set based on semi-supervised learning.

[0063] It's important to note that semi-supervised learning is a key research area in pattern recognition and machine learning, combining supervised and unsupervised learning methods. Semi-supervised learning utilizes a large amount of unlabeled data as well as labeled data simultaneously for pattern recognition. When using semi-supervised learning, it requires as few personnel as possible while still achieving relatively high accuracy.

[0064] In this embodiment, a learning model is obtained by training the target visual image training set based on semi-supervised learning, and the learning model is iteratively updated; the parameters of the iterated student model are weighted and combined to obtain the target visual image model.

[0065] Specifically, for the target visual image training set Ω m Each iteration updates the student model θ t The parameters of the previously iterated student model are weighted and combined to generate the teacher model: θ′ t =aθ′ t-1 +(1-a)θ t Where 'a' is a manually set weight coefficient and 't' is the number of training epochs. The teacher model is the average of the continuous student models; averaging the model weights across the training steps produces a more accurate teacher model. Therefore, from the perspective of deep neural network training, a consistency loss function is introduced. Where θ t For the student model, θ′ t The model used is the teacher model. The parameters of the student model are iterated based on the loss function. After the iteration is complete, the teacher model is updated. The final model can be either the student model or the teacher model.

[0066] In another embodiment, based on semi-supervised learning, a deep neural network model is obtained by training a training set of target visual images in the time domain; the deep neural network module is then subjected to semi-supervised deep neural network learning to obtain threads; and a target visual image model is generated based on the threads.

[0067] Specifically, for the training set Ω of a single target visual image m Iterative semi-supervised learning in the time domain is performed to train a deep neural network model θ with different iteration cycles. mt Different training sets of basic target visual images Ωm correspond to different semi-supervised deep neural networks that learn and obtain threads. In each thread, a teacher model θ' is generated. m,t =aθ' m,t-1 +(1-a)θm,t In multiple training threads, a consistency loss function is introduced, defined as J(θ) = E. x {||f(x,θ' m,t )-f(x,θ m,t )|| 2}+λ·E x {||E m {f(x,θ' m,t )}-f(x,θ m,t )|| 2}, where E m {f(x,θ′ m,t )} represents the average prediction of all teacher models, and J(θ) is used as a positive term for gradient feedback to the student network, guiding its training process to quickly converge to the performance of the teacher network.

[0068] Of course, to avoid the exponential weighting method failing to adapt well to the training process and dynamic changes of training samples in deep neural networks, in another embodiment, based on the semi-supervised deep neural network learning thread of the deep neural network module, the weighting matrix [w] of the target visual image training set is calculated. mk ] 1≤m≤M,1≤k≤T The target visual image training set is obtained by combining the weighted matrix and thread calculation. All training threads share a single teacher model, and the shared teacher is the average of all teacher models, i.e. Similarly, we need to minimize the difference between the student model and the teacher model, as shown in the following formula. Where W is a hyperparameter.

[0069] Specifically, for the M models obtained in the k-th training epoch, it is necessary to find the optimal M hyperparameters W. k =(W 1k W 2k ,..,W Mk These hyperparameters can be optimized using the particle swarm optimization algorithm. In this algorithm, N particles are initialized, each containing hyperparameter W. k , representing the particle's position in the hyperparameter search space. The particle's fitness function is defined as the position of each of the M models in W. k Predictive performance under weighting. Optimal W. k This will maximize the fitness function. During the iterative optimization process, the historical best position of each particle is recorded using the fitness function. And the best historical position among all particles For each particle, the m-th hyperparameter W mk Iteration as follows

[0070] W mk=W mk +c1×R(W mkb -W mk )+c2×R(W mkg -W mk )

[0071] Where c1 and c2 are fixed constants between 1.0 and 1.8, and R is a random number between 0 and 1.

[0072] In this embodiment, an initial visual image is acquired, decomposed using wavelet transform, and multi-level image combinations are obtained. These multi-level image combinations are then filtered, and the filtered multi-level image combinations are amplified using inverse wavelet transform to obtain a target visual image training set. A target visual image model is then obtained by training the target visual image training set using semi-supervised learning. The multi-level representation based on wavelet transform generates multiple target visual image training sets containing different frequency information and qualities, enhancing the diversity of visual image features. Without destroying the structural integrity of the visual image content, the resulting model exhibits better generalization performance. Furthermore, the optimization and fusion of multiple semi-supervised training processes result in faster convergence and better performance.

[0073] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0074] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0075] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0076] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0077] Further reference Figure 3 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of an image semi-supervised learning system based on wavelet transform. This system embodiment is similar to... Figure 2 The method embodiments shown correspond to those described.

[0078] like Figure 3 As shown, the wavelet transform-based image semi-supervised learning system 300 described in this embodiment includes: an acquisition module 301, a decomposition module 302, a filtering module 303, and a generation module 304. Wherein:

[0079] The acquisition module 301 is used to acquire the initial visual image;

[0080] The decomposition module 302 is used to decompose the initial visual image based on wavelet transform and obtain a multi-level image combination.

[0081] The filtering module 303 is used to filter multi-level image combinations and to obtain a target visual image training set by transforming and amplifying the filtered multi-level image combinations based on wavelet inverse transform.

[0082] The generation module 304 is used to obtain the target visual image model by training the target visual image training set based on semi-supervised learning.

[0083] This embodiment has the following main advantages: the acquisition module acquires the initial visual image, the wavelet transform decomposition module decomposes and processes the initial visual image to obtain a multi-level image combination, the filtering module filters the multi-level image combination, and the multi-level image combination after filtering is amplified by inverse wavelet transform to obtain the target visual image training set, and the target visual image model is obtained by training the target visual image training set based on the semi-supervised learning generation module. The multi-level representation based on wavelet transform generates multiple target visual image training sets containing different frequency information and quality, improving the diversity of visual image features. Without destroying the structural integrity of the visual image content, the obtained model has better generalization performance. Furthermore, the optimization and fusion of multiple semi-supervised training processes result in faster convergence and better performance of the obtained model.

[0084] In some optional implementations of this embodiment, the decomposition module 302 includes:

[0085] Decomposition unit, used to decompose an initial visual image into a wavelet image combination of different wavelets based on wavelet transform;

[0086] The extension unit is used to scale and move the wavelet image combination to form a multi-level image combination.

[0087] In some optional implementations of this embodiment, the filtering module 303 includes:

[0088] The filtering unit is used to filter the lowest low-frequency image and the highest high-frequency image in the multi-level image combination, and discard the lowest low-frequency image and the highest high-frequency image to obtain the filtered multi-level image combination.

[0089] The amplification unit is used to obtain a training set of target visual images by combining multi-level images after amplification and screening based on inverse wavelet transform.

[0090] In some optional implementations of this embodiment, the generation module 304 includes:

[0091] The classification unit is used to divide the filtered multi-level image combinations into low-frequency combinations and high-frequency combinations.

[0092] The conversion unit is used to convert the low-frequency combination and the high-frequency combination into low-frequency training set and high-frequency training set respectively based on wavelet inverse transform;

[0093] The generation unit is used to add the low-frequency training set to the high-frequency training set to obtain the target visual image training set.

[0094] In some optional implementations of this embodiment, the system 300 further includes:

[0095] The training module is used to train a deep neural network model on a training set of target visual images in the time domain based on semi-supervised learning.

[0096] The module is used to perform semi-supervised deep neural network learning on the deep neural network module and obtain threads;

[0097] The optimization module is used to generate the target visual image model based on the thread.

[0098] In some optional implementations of this embodiment, the obtaining module includes:

[0099] The computational unit is used to calculate the weighting matrix of the target visual image training set;

[0100] The combination unit is used to combine the weighting matrix with the thread calculation to obtain the target visual image training set.

[0101] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0102] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that... Figure 4 Only a computer device 4 with components 41-43 is shown in the illustration. However, it should be understood that it is not required to implement all of the shown components, and more or fewer components may be implemented instead. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0103] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0104] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions based on wavelet transform-based image semi-supervised learning methods. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0105] In some embodiments, the processor 42 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions for the wavelet transform-based image semi-supervised learning method.

[0106] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0107] The embodiments of this application have the following main advantages: An initial visual image is acquired, decomposed based on wavelet transform, and multi-level image combinations are obtained. These multi-level image combinations are then filtered, and a target visual image training set is obtained by amplifying the filtered multi-level image combinations using inverse wavelet transform. A target visual image model is then obtained by training the target visual image training set using semi-supervised learning. The multi-level representation based on wavelet transform generates multiple target visual image training sets containing different frequency information and qualities, enhancing the diversity of visual image features. Without destroying the structural integrity of the visual image content, the obtained model exhibits better generalization performance. Furthermore, the optimization and fusion of multiple semi-supervised training processes result in faster convergence and better performance of the obtained model.

[0108] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the wavelet transform-based image semi-supervised learning method described above.

[0109] The embodiments of this application have the following main advantages: An initial visual image is acquired, decomposed based on wavelet transform, and multi-level image combinations are obtained. These multi-level image combinations are then filtered, and a target visual image training set is obtained by amplifying the filtered multi-level image combinations using inverse wavelet transform. A target visual image model is then obtained by training the target visual image training set using semi-supervised learning. The multi-level representation based on wavelet transform generates multiple target visual image training sets containing different frequency information and qualities, enhancing the diversity of visual image features. Without destroying the structural integrity of the visual image content, the obtained model exhibits better generalization performance. Furthermore, the optimization and fusion of multiple semi-supervised training processes result in faster convergence and better performance of the obtained model.

[0110] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0111] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A semi-supervised image learning method based on wavelet transform, characterized in that, The method includes the following steps: Obtain the initial visual image; The initial visual image is processed by wavelet transform decomposition, and a multi-level image combination is obtained. Filter multi-level image combinations and use wavelet inverse transform to amplify the filtered multi-level image combinations to obtain the target visual image training set; A target visual image model is obtained by training a target visual image training set using semi-supervised learning. The filtered multi-level image combinations are divided into low-frequency combinations and high-frequency combinations; Based on wavelet inverse transform, the low-frequency combination and the high-frequency combination are converted into low-frequency training set and high-frequency training set, respectively; The low-frequency training set is added to the high-frequency training set to obtain the target visual image training set.

2. The image semi-supervised learning method based on wavelet transform according to claim 1, characterized in that, The steps of processing the initial visual image based on wavelet transform decomposition and obtaining multi-level image combinations include: The initial visual image is decomposed into a combination of wavelet images with different wavelets based on wavelet transform; Wavelet images are combined by scaling and shifting their positions to form multi-level image combinations.

3. The image semi-supervised learning method based on wavelet transform according to claim 1, characterized in that, The steps of filtering multi-level image combinations and obtaining the target visual image training set by transforming and amplifying the filtered multi-level image combinations based on inverse wavelet transform include: The lowest low-frequency image and the highest high-frequency image are selected from the multi-level image combination, and the lowest low-frequency image and the highest high-frequency image are discarded to obtain the selected multi-level image combination. The target visual image training set is obtained by combining multi-level images after wavelet inverse transform amplification and screening.

4. The image semi-supervised learning method based on wavelet transform according to claim 1, characterized in that, The steps for obtaining the target visual image model by training the target visual image training set based on semi-supervised learning include: The learning model is obtained by training a set of target visual images using semi-supervised learning, and then iteratively updated. The target visual image model is obtained by weighting and combining the parameters of the iterated student model.

5. The image semi-supervised learning method based on wavelet transform according to claim 1, characterized in that, After the step of obtaining the target visual image model by training the target visual image training set based on semi-supervised learning, the method further includes: Based on semi-supervised learning, a deep neural network model is obtained by training a training set of target visual images in the time domain. The deep neural network module is subjected to semi-supervised deep neural network learning to obtain threads; Generate the target visual image model based on the thread.

6. The image semi-supervised learning method based on wavelet transform according to claim 5, characterized in that, After the step of performing semi-supervised deep neural network learning on the deep neural network module and obtaining the thread, the method further includes: Calculate the weighting matrix of the training set of target visual images; The target visual image training set is obtained by combining the weighted matrix and thread calculation.

7. A semi-supervised image learning system based on wavelet transform, characterized in that, include: The acquisition module is used to acquire the initial visual image; The decomposition module is used to decompose the initial visual image based on wavelet transform and obtain a multi-level image combination. The filtering module is used to filter multi-level image combinations and to obtain the target visual image training set by transforming and amplifying the filtered multi-level image combinations based on inverse wavelet transform. The filtered multi-level image combinations are divided into low-frequency combinations and high-frequency combinations; Based on wavelet inverse transform, the low-frequency combination and the high-frequency combination are converted into low-frequency training set and high-frequency training set, respectively; The low-frequency training set is added to the high-frequency training set to obtain the target visual image training set; The generation module is used to obtain a target visual image model by training a set of target visual images based on semi-supervised learning.

8. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the wavelet transform-based image semi-supervised learning method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the wavelet transform-based image semi-supervised learning method as described in any one of claims 1 to 6.

Images