Image processing method and system based on Chebyshev polynomial network
By introducing Chebishev polynomial network into neural networks, the nonlinear correlation of the network layer is enhanced, and the problem of traditional neural networks insufficient fitting ability to complex functions is solved, achieving more efficient image processing accuracy and diversity applications.
Patent Information
- Application Number
- CN202411359028.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2024-12-27
AI Technical Summary
The correlation method between different layers in traditional neural networks is relatively simple, which limits the network's ability to fit complex functions and makes it difficult to effectively handle complex image processing tasks.
Using Chebishev polynomial network, the output of the network layer is nonlinearly combined with the Chebishev polynomial terms, and a flexible and efficient network structure is constructed through Chebishev polynomial recursive relationship and Hadamama product operations to enhance the expression and quasi- tion of the network. Compatible with ability.
It significantly improves the modeling performance and processing accuracy of neural networks in the field of image processing, adapts to different network structures, and is widely used in image classification, semantic segmentation and image generation tasks.
Smart Images
Figure CN119204088A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of image processing, and in particular to an image processing method and system based on a Chebyshev polynomial network. Background Art
[0002] With the continuous advancement of computing power and deep learning technology, artificial intelligence has made outstanding achievements in fields such as image processing. From the initial manual feature extraction, to methods based on statistical pattern recognition (such as support vector machines SVM) and signal processing (such as wavelet transform), to the widely used end-to-end neural networks, image processing technology has continued to innovate and break through. However, the connection between different layers in traditional neural networks is relatively simple, usually just linear full connection or convolution operations. This fixed relationship between layers restricts the network's ability to fit complex functions.
[0003] In order to further improve the expressive power of neural networks, researchers have begun to explore new methods to expand the connections between network layers. One idea is to construct a polynomial network, that is, to nonlinearly combine the output of the network layer with a series of polynomial terms, thereby giving the network the ability of polynomial approximation. Since any continuous function can be approximated by a polynomial function with a certain accuracy, the polynomial network has a stronger function approximation ability and is expected to significantly improve the modeling performance of neural networks in fields such as image processing.
[0004] When constructing a polynomial network, the choice of polynomial family is crucial. In theory, we can use different polynomial systems such as power basis polynomials, Legendre polynomials, Chebyshev polynomials, etc. Among them, Chebyshev polynomials have good mathematical properties and are the best consistent approximation polynomials on a continuous interval, which can minimize the maximum deviation of polynomial approximation. Therefore, using Chebyshev polynomials to construct network inter-layer associations can give the network the best polynomial approximation ability, thereby maximizing the network's fit to the target function. Summary of the invention
[0005] In view of the defects in the prior art, the purpose of the present invention is to provide an image processing method and system based on a Chebyshev polynomial network.
[0006] To achieve the above object, according to a first aspect of the present disclosure, there is provided an image processing method based on a Chebyshev polynomial network, comprising:
[0007] Determine a backbone network according to a preset image processing task, wherein the preset image processing task includes image classification, semantic segmentation, and image generation;
[0008] The backbone network and a preset Chebyshev polynomial module are connected in parallel to construct a Chebyshev polynomial network;
[0009] Training the Chebyshev polynomial network to determine a trained Chebyshev polynomial network;
[0010] The trained Chebyshev polynomial network is used to perform the image processing task and determine the image processing task result.
[0011] Optionally, the backbone network includes a backbone network module, and the preset Chebyshev polynomial module includes one or more Chebyshev polynomial modules.
[0012] Optionally, the step of connecting the backbone network and a preset Chebyshev polynomial module in parallel to construct a Chebyshev polynomial network includes:
[0013] The backbone network module and one or more Chebyshev polynomial modules are connected in parallel to determine the Chebyshev polynomial network.
[0014] Optionally, the method further comprises:
[0015] According to the preset image processing task, the image training data and the loss function corresponding to the preset image processing task are determined.
[0016] Optionally, the training the Chebyshev polynomial network to determine a trained Chebyshev polynomial network includes:
[0017] The Chebyshev polynomial network is trained using the image training data and loss function corresponding to the preset image processing task to determine the trained Chebyshev polynomial network.
[0018] Optionally, the adopting the trained Chebyshev polynomial network to perform the image processing task and determining the image processing task result includes:
[0019] Inputting an input signal into the backbone network module to determine a first output result;
[0020] Performing a mapping operation on the first output result to determine a mapping result;
[0021] Inputting the mapping result into the first Chebyshev module of the preset Chebyshev polynomial module, and determining the Chebyshev polynomial of each order according to the preset order;
[0022] The first output result is summed with the product of the Chebyshev polynomial of each order and the first output result to determine the image processing task result.
[0023] According to a second aspect of the present disclosure, there is provided an image processing system based on a Chebyshev polynomial network, comprising:
[0024] A backbone network determination module, used to determine a backbone network according to a preset image processing task, wherein the preset image processing task includes image classification, semantic segmentation, and image generation;
[0025] A network construction module, used to connect the backbone network and a preset Chebyshev polynomial module in parallel to construct a Chebyshev polynomial network;
[0026] A network training module, used for training the Chebyshev polynomial network and determining a trained Chebyshev polynomial network;
[0027] The image processing module is used to use the trained Chebyshev polynomial network to perform the image processing task and determine the image processing task result.
[0028] According to a third aspect of the present disclosure, a non-temporary computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the steps of the method provided in the first aspect of the present disclosure are implemented.
[0029] According to a fourth aspect of the present disclosure, there is provided an electronic device, including:
[0030] a memory having a computer program stored thereon;
[0031] A processor is used to execute the computer program in the memory to implement the steps of the method provided in the first aspect of the present disclosure.
[0032] Compared with the prior art, the embodiments of the present disclosure have at least one of the following beneficial effects:
[0033] Through the above technical scheme, according to the preset image processing tasks, the backbone network is determined, the backbone network and the Chebyshev polynomial module are connected in parallel, and a Chebyshev polynomial network is constructed to enhance the network's expression and fitting capabilities. It can also adapt to different network structures and widely process image processing tasks, such as image classification, image segmentation, and image generation tasks, thereby improving the image processing accuracy of image processing tasks.
[0034] The embodiments of the present disclosure construct a flexible and efficient Chebyshev polynomial network by introducing the Chebyshev polynomial recursive relationship and the Hadamard product operation, giving the network polynomial approximation capability, enhancing the network's expression and fitting capabilities, and enhancing the network's modeling and processing capabilities in the field of image processing, thereby meeting the application requirements of various scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Other features, objects and advantages of the present disclosure will become more apparent by reading the detailed description of non-limiting embodiments with reference to the following drawings:
[0036] Figure 1 The present invention is a flowchart of an image processing method based on a Chebyshev polynomial network according to an exemplary embodiment.
[0037] Figure 2 The figure is a flowchart of another image processing method based on a Chebyshev polynomial network according to an exemplary embodiment.
[0038] Figure 3 The diagram is a schematic diagram of the structure of a Chebyshev polynomial network according to an exemplary embodiment.
[0039] Figure 4 The figure is a block diagram of an image processing system based on a Chebyshev polynomial network according to an exemplary embodiment. DETAILED DESCRIPTION
[0040] The present disclosure is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present disclosure, but are not intended to limit the present disclosure in any form. It should be noted that, for those of ordinary skill in the art, several variations and improvements may be made without departing from the concept of the present disclosure. These all fall within the scope of protection of the present disclosure.
[0041] Figure 1 The present invention is a flowchart of an image processing method based on a Chebyshev polynomial network according to an exemplary embodiment.
[0042] like Figure 1 As shown, the present disclosure provides an image processing method based on a Chebyshev polynomial network, including S11 to S14.
[0043] S11, determining a backbone network according to a preset image processing task.
[0044] Among them, the preset image processing tasks include image classification, semantic segmentation and image generation, and each image task corresponds to a different backbone network.
[0045] The backbone network can include multi-layer perceptron network (MLP), convolutional neural network (CNN), recurrent neural network (RNN) and Transformer.
[0046] S12, connecting the backbone network and the preset Chebyshev polynomial module in parallel to construct a Chebyshev polynomial network.
[0047] S13, training the Chebyshev polynomial network to determine a trained Chebyshev polynomial network.
[0048] S14, using the trained Chebyshev polynomial network to perform the image processing task and determine the image processing task result.
[0049] Through the above technical scheme, according to the preset image processing tasks, the backbone network is determined, the backbone network and the Chebyshev polynomial module are connected in parallel, and a Chebyshev polynomial network is constructed to enhance the network's expression and fitting capabilities. It can also adapt to different network structures and widely process image processing tasks, such as image classification, image segmentation, and image generation tasks, thereby improving the image processing accuracy of image processing tasks.
[0050] Figure 2 The figure is a flow chart of another image processing method based on a Chebyshev polynomial network according to an exemplary embodiment.
[0051] like Figure 2 As shown, in a possible embodiment, S11, according to a preset image processing task, determining a backbone network includes:
[0052] If the preset image processing task is image classification, the backbone network can use any one of the following: multi-layer perceptron MLP, plain network, convolutional neural network CNN, residual network ResNet, and convolutional neural network MobineNetV2.
[0053] If the preset image task is semantic segmentation, the backbone network can use the UNet network.
[0054] If the preset image task is image generation, the backbone network can use the UNet-diffusion network.
[0055] In a possible embodiment, the backbone network includes a backbone network module, and the preset Chebyshev polynomial module includes one or more Chebyshev polynomial modules.
[0056] In a possible embodiment, S12, connecting the backbone network and the preset Chebyshev polynomial module in parallel to construct a Chebyshev polynomial network includes:
[0057] The backbone network module and one or more Chebyshev polynomial modules are connected in parallel to determine a Chebyshev polynomial network.
[0058] Among them, the output of the backbone network module is used as the input of the first Chebyshev polynomial module, and the output of each previous Chebyshev polynomial module is used as the input of the next Chebyshev polynomial module, and so on.
[0059] In a possible embodiment, an image processing method based on a Chebyshev polynomial network further includes S15.
[0060] S15, according to the preset image processing task, determining the image training data and loss function corresponding to the preset image processing task.
[0061] Among them, according to the preset image processing tasks, such as image classification, semantic segmentation and image generation, image training data corresponding to the image processing tasks are selected from the public image datasets.
[0062] As an example, if the preset image processing task is image classification, the image training data can use any one or more of the CIFAR10 and CIFAR100 data sets, and the loss function uses the cross entropy loss function (Cross Entropyloss).
[0063] As another example, if the preset image processing task is semantic segmentation, the image training data can use any one or more of the ACDC and BraTS19 datasets, and the loss function uses the weighted sum of the cross entropy loss function (Cross Entropy loss) and the Dice loss (Dice loss).
[0064] As another example, if the preset image processing task is image generation, the image training data may use the MNIST data set, and the loss function may use the MSE loss function.
[0065] Wherein, step S15 may be executed in parallel with step S11, or executed after step S12.
[0066] In a possible embodiment, S13, training the Chebyshev polynomial network to determine the trained Chebyshev polynomial network may include:
[0067] The Chebyshev polynomial network is trained using image training data and loss functions corresponding to a preset image processing task to determine a trained Chebyshev polynomial network.
[0068] As an example, if the preset image processing task is image classification, any one or more of the CIFAR10 and CIFAR100 data sets are used as image training data to train the Chebyshev polynomial network, and the cross entropy loss function (Cross Entropy loss) is used to optimize the Chebyshev polynomial network to determine the trained Chebyshev polynomial network.
[0069] As another example, if the preset image processing task is semantic segmentation, any one or more of the ACDC and BraTS19 datasets are used as image training data to train the Chebyshev polynomial network, and the weighted sum of the cross entropy loss function (Cross Entropy loss) and the Dice loss (Dice loss) is used as the loss function to optimize the Chebyshev polynomial network to determine the trained Chebyshev polynomial network.
[0070] As another example, if the preset image processing task is image generation, the MNIST data set is used as image training data to train the Chebyshev polynomial network, and the square loss function (MSE loss) is used as the loss function to optimize the Chebyshev polynomial network to determine the trained Chebyshev polynomial network.
[0071] Figure 3 The diagram is a schematic diagram of the structure of a Chebyshev polynomial network according to an exemplary embodiment.
[0072] like Figure 3 As shown, in a possible embodiment, S14, using a trained Chebyshev polynomial network to perform an image processing task and determine an image processing task result, includes S21 to S24.
[0073] S21, inputting an input signal into a backbone network module to determine a first output result.
[0074] The input signal is represented by X, and the first output result is represented by f(X).
[0075] S22: Perform a mapping operation on the first output result to determine a mapping result.
[0076] The mapping operation may include any one or more combinations of linear transformation, up / down sampling, and convolution / deconvolution.
[0077] The mapping result is represented as g(X). The dimension of the input signal X is changed through the mapping operation, and the dimension of the mapping result g(X) is controlled to be the same as the output dimension of the backbone network module.
[0078] S23, inputting the mapping result into the first Chebyshev module of the preset Chebyshev polynomial module, and determining the Chebyshev polynomial of each order according to the preset order.
[0079] The preset order is n, the highest order n is a preset integer greater than 1, and the Chebyshev polynomials T of order 0 to n are determined. i (g(X)), i = 0, 1, ..., n. The details are as follows:
[0080] 0th order Chebyshev polynomial T0(g(X))=1;
[0081] 1st order Chebyshev polynomial T1(g(X))=g(X);
[0082] Starting from order 2, the i-th order Chebyshev polynomial is:
[0083] T i+1 (g(X))=2g(X)T i (g(X))-T i-1 (g(X));
[0084] Among them, g(X) represents the mapping result, and X represents the input signal.
[0085] Starting from the 2nd order, calculate the nth order Chebyshev polynomial T through the recursive relationship n (g(X)).
[0086] S24, summing the first output result and the product of each order of Chebyshev polynomial and the first output result to determine the image processing task result.
[0087] Specifically, the following formula is used to determine the image processing task result:
[0088] y pred =f(X)+f(X)·[T0(g(X))+T1(g(X))+…T n (g(X))];
[0089] Wherein, f(X) represents the first output result, T0(g(X)) represents the 0th order Chebyshev polynomial, T1(g(X)) represents the 1st order Chebyshev polynomial, T n (g(X)) represents an n-th order Chebyshev polynomial.
[0090] By introducing the Chebyshev polynomial recursion relationship and Hadamard product operation, a flexible and efficient Chebyshev polynomial network is constructed, giving the network polynomial approximation capability, enhancing the network's expression and fitting capabilities, and enhancing the network's modeling and processing capabilities in the field of image processing to meet the application needs of various scenarios.
[0091] In a possible embodiment, a trained Chebyshev polynomial network is used to perform an image classification task, the input of the Chebyshev polynomial network is an image matrix, and the output is an image category, and the image processing task result is the output image category.
[0092] In another possible embodiment, a trained Chebyshev polynomial network is used to perform the semantic segmentation task, the input of the Chebyshev polynomial network is an image matrix, and the output is a segmented image, and the image processing task result is the output segmented image.
[0093] In another possible embodiment, a trained Chebyshev polynomial network is used to perform the image generation task, the input of the Chebyshev polynomial network is random noise, and the output is the generated image, then the image processing task result is the output generated image.
[0094] In a possible embodiment, the image classification task is performed using the image processing method based on the Chebyshev polynomial network provided by the present disclosure.
[0095] Among them, image classification refers to the image processing task of inputting the image to be processed and outputting a preset category label, distinguishing objects of different categories based on the features reflected in the image information.
[0096] The sample data used for image classification in the present disclosure may adopt image data to be classified.
[0097] Specifically, the present disclosure uses the training image data in the public datasets CIFAR10 and CIFAR100 as image training data. The specific process of the image classification method is:
[0098] M11, determine the backbone network based on the image classification task.
[0099] Among them, the backbone network can adopt any one of the multi-layer perceptron MLP, convolutional neural network CNN, residual network ResNet, and convolutional neural network MobineNetV2.
[0100] M12, connect the backbone network and the preset Chebyshev polynomial module in parallel to build a Chebyshev polynomial network.
[0101] M13, obtain the training image data in the public datasets CIFAR10 and CIFAR100 as image training data.
[0102] M14, using the training image data in the public data sets CIFAR10 and CIFAR100 and the cross entropy loss function (Cross Entropy loss) to train the Chebyshev polynomial network and determine the trained Chebyshev polynomial network.
[0103] M15, inputs the image matrix of the image sample to be classified into the Chebyshev polynomial network, outputs the image category label, and determines the result of the image classification task.
[0104] The backbone network: multi-layer perceptron MLP, convolutional neural network CNN, residual network ResNet, convolutional neural network MobineNetV2 and Chebyshev polynomial network using the backbone network were trained on the public datasets CIFAR10 and CIFAR100 for 120 epochs, respectively, with an initial learning rate of 0.1, using the SGD optimizer, and a weight decay of 5e-4.
[0105] The test results of the image classification task of multi-layer perceptron MLP, convolutional neural network CNN, residual network ResNet, convolutional neural network MobineNetV2 and Chebyshev polynomial network are shown in the following table:
[0106]
[0107] Table 1
[0108]
[0109] Table 2
[0110] Among them, Table 1 is a comparison table of image classification test accuracy data of the multilayer perceptron MLP and the Chebyshev polynomial network provided by the present invention whose backbone network is the multilayer perceptron MLP, and Table 2 is a comparison table of image classification test accuracy data of the convolutional neural network CNN, the residual network ResNet, the convolutional neural network MobineNetV2, and the Chebyshev polynomial network with the convolutional neural network CNN, the residual network ResNet, and the convolutional neural network MobineNetV2 as the backbone network.
[0111] As shown in Tables 1 and 2, compared with the corresponding backbone network, the Chebyshev polynomial network has a higher test accuracy in the image classification task. Therefore, the image processing method based on the Chebyshev polynomial network provided by the present invention is more suitable for image classification tasks, can enhance the expression and fitting capabilities of the network, and can adapt to different network structures.
[0112] In a possible embodiment, the image processing method based on the Chebyshev polynomial network provided by the present disclosure is used to perform a semantic segmentation task.
[0113] Semantic segmentation means distinguishing pixels belonging to different objects in an image. As a basic task of computer vision, semantic segmentation divides visual input into different interpretable categories. For example, in the field of autonomous driving, target detection requires a large amount of road and target object feature image data. Through semantic segmentation of image data, all pixels belonging to cars, roads, signs, and objects in the image are calibrated to defined colors with RGB values. Semantic segmentation can complete road labeling, pedestrian labeling, face labeling, vehicle labeling, and object labeling, but it is not limited to this.
[0114] The sample data for semantic segmentation disclosed in the present invention may adopt image data to be semantically segmented.
[0115] Specifically, the present disclosure uses the training image data of the public dataset ACDC and BraTS19 dataset as image training data. The specific process of the semantic segmentation method is:
[0116] M21, determine the backbone network based on the semantic segmentation task.
[0117] Among them, the backbone network can adopt the backbone network UNet.
[0118] M22, connects the backbone network and the preset Chebyshev polynomial module in parallel to construct a Chebyshev polynomial network.
[0119] M23, obtains training image data from the public dataset ACDC and BraTS19 dataset as image training data.
[0120] M24, uses the training image data in the public dataset ACDC and BraTS19 dataset and the weighted sum of the cross entropy loss function (Cross Entropy loss) and the Dice loss (Dice loss) as the loss function to train the Chebyshev polynomial network and determine the trained Chebyshev polynomial network.
[0121] M25, inputs the image matrix of the image sample to be semantically segmented into the Chebyshev polynomial network, outputs the segmented image, and determines the result of the image semantic segmentation task.
[0122] The backbone network UNet and the Chebyshev polynomial network using the backbone network UNet were trained for 30,000 iterations on the public datasets ACDC and BraTS19, respectively. The initial learning rate was set to 0.01, the SGD optimizer was used, and the weight decay was 1e-4.
[0123] The dice values of the backbone network UNet and the Chebyshev polynomial network using the backbone network UNet in the semantic segmentation task are shown in the following table:
[0124]
[0125] Table 3
[0126] As shown in Table 3, compared with the backbone network UNet, the Chebyshev polynomial network using the backbone network UNet has a higher dice value in the semantic segmentation test task. Therefore, the image processing method based on the Chebyshev polynomial network proposed in the present disclosure is more suitable for semantic segmentation tasks, can enhance the expression and fitting capabilities of the network, and can adapt to different network structures.
[0127] In a possible embodiment, the image generation task is performed using the image processing method based on the Chebyshev polynomial network provided by the present disclosure.
[0128] Image generation refers to generating new images from random noise through a model. Image generation is an important task in computer vision. In the image generation process, the model learns the mapping relationship between input noise and output image to generate images with specific features. For example, in the field of art creation, image generation can automatically generate images with artistic style, saving a lot of manual creation time.
[0129] There are many methods for image generation, such as Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs). These methods train the generated images so that they can realistically reflect the distribution of the target data. During the image generation process, the model learns the complex structure and pattern of the data, thereby generating high-quality images. Commonly used training datasets in image generation tasks include MNIST and CIFAR-10.
[0130] Image generation enables computers to automatically generate images with specific features, greatly expanding the application scope of computer vision.
[0131] Specifically, the present disclosure uses the training image data in the public data set MNIST as the image training data. The specific process of the image generation method is as follows:
[0132] M31, determine the backbone network based on the image classification task.
[0133] Among them, the backbone network can adopt the UNet-diffusion network.
[0134] M32, connects the backbone network and the preset Chebyshev polynomial module in parallel to construct a Chebyshev polynomial network.
[0135] M33, obtain training image data in the public data set MNIST as image training data.
[0136] M34, uses the training image data in the public data set MNIST and the square loss function (MSE loss) as the loss function to train the Chebyshev polynomial network and determine the trained Chebyshev polynomial network.
[0137] M35, inputs random noise into the Chebyshev polynomial network, outputs the generated image, and determines the result of the image generation task.
[0138] The backbone network UNet-diffusion and the Chebyshev polynomial network using the backbone network UNet-diffusion were trained and sampled for 5 rounds on the public dataset MNIST dataset, with 1000 epochs in each round and 1000 samples in each round, generating 64 images for quality evaluation.
[0139] The fid values of the images generated by the backbone network UNet-diffusion and the Chebyshev polynomial network using the backbone network UNet-diffusion in the image generation task are shown in the following table:
[0140]
[0141]
[0142] Table 4
[0143] As shown in Table 4, compared with the backbone network UNet-diffusion, the Chebyshev polynomial network using the backbone network UNet-diffusion has a higher fid value in the image generation task. Therefore, the image processing method based on the Chebyshev polynomial network proposed in the present invention is more suitable for semantic segmentation tasks, can enhance the expression and fitting capabilities of the network, and can adapt to different network structures.
[0144] Figure 4 The figure is a block diagram of an image processing system based on a Chebyshev polynomial network according to an exemplary embodiment.
[0145] Based on the same concept, the present disclosure also provides an image processing system 100 based on a Chebyshev polynomial network. Figure 4 As shown, it includes a backbone network determination module 110, a network construction module 120, a network training module 130, and an image processing module 140.
[0146] A backbone network determination module 110, used to determine a backbone network according to a preset image processing task, wherein the preset image processing task includes image classification, semantic segmentation, and image generation;
[0147] A network construction module 120, used to connect the backbone network and a preset Chebyshev polynomial module in parallel to construct a Chebyshev polynomial network;
[0148] A network training module 130 is used to train the Chebyshev polynomial network and determine a trained Chebyshev polynomial network;
[0149] The image processing module 140 is used to use the trained Chebyshev polynomial network to perform the image processing task and determine the image processing task result.
[0150] Through the above technical scheme, according to the preset image processing tasks, the backbone network is determined, the backbone network and the Chebyshev polynomial module are connected in parallel, and a Chebyshev polynomial network is constructed to enhance the network's expression and fitting capabilities. It can also adapt to different network structures and widely process image processing tasks, such as image classification, image segmentation, and image generation tasks, thereby improving the image processing accuracy of image processing tasks.
[0151] Regarding the embodiment of the above system, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0152] Based on the same concept as above, in another embodiment of the present disclosure, there is also provided an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor is used to execute an image processing method based on a Chebyshev polynomial network when executing the program.
[0153] Optionally, the memory is used to store programs; the memory may include volatile memory (English: volatile memory), such as random-access memory (English: random-access memory, abbreviated: RAM), such as static random-access memory (English: static random-access memory, abbreviated: SRAM), double data rate synchronous dynamic random access memory (English: Double Data Rate Synchronous Dynamic Random Access Memory, abbreviated: DDR SDRAM), etc.; the memory may also include non-volatile memory (English: non-volatile memory), such as flash memory (English: flash memory). The memory is used to store computer programs (such as applications, functional modules, etc. that implement the above method), computer instructions, etc., and the above computer programs, computer instructions, etc. can be partitioned and stored in one or more memories. And the above computer programs, computer instructions, data, etc. can be called by the processor.
[0154] The above-mentioned computer programs, computer instructions, etc. may be stored in one or more memories in partitions, and the above-mentioned computer programs, computer instructions, data, etc. may be called by a processor.
[0155] The processor is used to execute the computer program stored in the memory to implement the various steps in the method involved in the above embodiment. For details, please refer to the relevant description in the above method embodiment.
[0156] The processor and the memory may be independent structures or integrated structures. When the processor and the memory are independent structures, the memory and the processor may be coupled and connected via a bus.
[0157] In an embodiment of the present disclosure, a non-temporary computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of an image processing method based on a Chebyshev polynomial network in any of the above embodiments are implemented.
[0158] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0159] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0160] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0161] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0162] Although the preferred embodiments of the present disclosure have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present disclosure.
[0163] Obviously, those skilled in the art can make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is also intended to include these modifications and variations.
Claims
1. An image processing method based on a Chebyshev polynomial network, characterized in that: include: Determine a backbone network according to a preset image processing task, wherein the preset image processing task includes image classification, semantic segmentation, and image generation; The backbone network and a preset Chebyshev polynomial module are connected in parallel to construct a Chebyshev polynomial network; Training the Chebyshev polynomial network to determine a trained Chebyshev polynomial network; The trained Chebyshev polynomial network is used to perform the image processing task and determine the image processing task result.
2. The method according to claim 1, characterized in that The backbone network includes a backbone network module, and the preset Chebyshev polynomial module includes one or more Chebyshev polynomial modules; The backbone network and the preset Chebyshev polynomial module are connected in parallel to construct a Chebyshev polynomial network, including: The backbone network module and one or more Chebyshev polynomial modules are connected in parallel to determine the Chebyshev polynomial network.
3. The method according to claim 1, characterized in that The method further comprises: According to the preset image processing task, the image training data and the loss function corresponding to the preset image processing task are determined.
4. The method according to claim 3, characterized in that The step of training the Chebyshev polynomial network to determine a trained Chebyshev polynomial network includes: The Chebyshev polynomial network is trained using the image training data and loss function corresponding to the preset image processing task to determine the trained Chebyshev polynomial network.
5. The method according to claim 2, characterized in that: The adopting the trained Chebyshev polynomial network to perform the image processing task and determining the image processing task result comprises: Inputting an input signal into the backbone network module to determine a first output result; Performing a mapping operation on the first output result to determine a mapping result; Inputting the mapping result into the first Chebyshev module of the preset Chebyshev polynomial module, and determining the Chebyshev polynomial of each order according to the preset order; The first output result is summed with the product of the Chebyshev polynomial of each order and the first output result to determine the image processing task result.
6. The method according to claim 5, characterized in that The step of inputting the mapping result into the first Chebyshev module of the preset Chebyshev polynomial module, and determining the Chebyshev polynomial of each order according to the preset order, comprises: 0th order Chebyshev polynomial T0(g(X))=1; 1st order Chebyshev polynomial T1(g(X))=g(X); Starting from order 2, the i-th order Chebyshev polynomial is: T i+1 (g(X))=2g(X)T i (g(X))-T i-1 (g(X)); Wherein, g(X) represents the mapping result, and X represents the input signal.
7. The method according to claim 6, characterized in that The step of summing the first output result and the product of the Chebyshev polynomial of each order and the first output result to determine the image processing task result includes: y pred =f(X)+f(X)·[T0(g(X))+T1(g(X))+…T n (g(X))] Among them, y pred represents the image processing task result, f(X) represents the first output result, T0(g(X)) represents the 0th order Chebyshev polynomial, T1(g(X)) represents the 1st order Chebyshev polynomial, T n (g(X)) represents an n-th order Chebyshev polynomial.
8. An image processing system based on a Chebyshev polynomial network, characterized in that: include: A backbone network determination module, used to determine a backbone network according to a preset image processing task, wherein the preset image processing task includes image classification, semantic segmentation, and image generation; A network construction module, used to connect the backbone network and a preset Chebyshev polynomial module in parallel to construct a Chebyshev polynomial network; A network training module, used for training the Chebyshev polynomial network and determining a trained Chebyshev polynomial network; The image processing module is used to use the trained Chebyshev polynomial network to perform the image processing task and determine the image processing task result.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.
10. A terminal, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 7.