Jingchu style digital artwork generation method and system based on artificial intelligence

Abstract

The invention discloses a Jingchu style digital artwork generation method and system based on artificial intelligence. The method comprises the steps of Jingchu artwork collection, data preprocessing, Jingchu style feature combination optimization, digital artwork generation model construction and Jingchu style digital artwork generation. The invention belongs to the technical field of digital artwork generation, and particularly relates to a Jingchu style digital artwork generation method and system based on artificial intelligence. According to the scheme, a Jingchu style feature combination is creatively obtained and optimized in advance, and the style consistency of generated works is ensured; a particle swarm optimization algorithm is improved by combining particle velocity weight value adjustment and learning factor dynamic adjustment, and it is ensured that the generated artwork better meets the expectation; by introducing a diversity loss item and a discriminator penalty item to carry out optimization and generation of the model, it is ensured that the artwork has enough variability, the model stability is improved, and therefore the Jingchu style digital artwork is remarkably improved in multiple aspects.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital artwork generation, and specifically relates to a method and system for generating digital artwork in Jingchu style based on artificial intelligence. Background Art

[0002] Jingchu culture is a regional culture with a long history and rich characteristics in China, covering the historical heritage, artistic style and symbolic symbols of the Chu State. However, the creation of traditional artworks faces problems such as high cost, long time and difficulty in ensuring style consistency. By using artificial intelligence technology, especially deep learning and generative adversarial networks, we can simulate and learn the Jingchu style and quickly generate digital artworks that conform to the cultural background. This method not only improves the efficiency of artistic creation, but also promotes the digital preservation and innovative reproduction of traditional culture. However, in the process of generating traditional style digital artworks, there is a problem that the extraction of special cultural style features usually needs to be performed dynamically each time it is generated, resulting in low generation efficiency, poor style consistency and excessive computational burden; in the process of existing feature combination optimization, there is an improper setting of feature weights, which leads to the fact that the artwork generated by the model does not meet the expected technical problem; in the existing models suitable for digital artwork generation, there are technical problems such as poor style consistency of generated artworks, lack of diversity in generated results and unstable generation process. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a Jingchu style digital artwork generation method and system based on artificial intelligence. In view of the problem that in the traditional style digital artwork generation process, the extraction of special cultural style features usually needs to be performed dynamically at each generation, resulting in low generation efficiency, poor style consistency and excessive computational burden, this scheme creatively proposes to obtain and optimize the Jingchu style feature combination in advance, and use the optimized features as input to avoid repeated calculations in each generation process, thereby improving the generation efficiency, ensuring the style consistency of the generated works, and reducing the consumption of computing resources. The Jingchu style features optimized in advance can also enhance the control of the generation model over the style details, so that the generated Jingchu style digital artwork is more in line with expectations and has higher artistic and cultural accuracy. In the process of optimizing the existing feature combination, there is an improper setting of feature weights, which leads to the artwork generated by the model not meeting the expected technical problem This scheme proposes to optimize the selection of feature weights by improving the particle swarm optimization algorithm, combining the method of adjusting the particle velocity weight value and the dynamic adjustment of the learning factor, and can effectively obtain the optimal feature weight combination, thereby ensuring that the generated artwork is more in line with expectations, achieving consistency in style and improved artistic effects; in view of the technical problems of poor consistency in the style of generated artworks, lack of diversity in generated results, and unstable generation process in the existing digital artwork generation model, this scheme innovatively optimizes by introducing diversity loss terms and discriminator penalty terms. Through the diversity loss term, it ensures that the generated image has sufficient variability in style and details, and effectively avoids the repetition of similar images. Through the introduction of the discriminator penalty term, the stability of the discriminator is enhanced, and the generator is promoted to learn richer and more realistic artistic features, so that the generated Jingchu style digital artworks have been significantly improved in terms of style consistency, artistic value and creative diversity.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a method for generating digital artworks in the Jingchu style based on artificial intelligence, and the method comprises the following steps:

[0005] Step S1: Collection of Jingchu artworks;

[0006] Step S2: data preprocessing;

[0007] Step S3: Jingchu style feature combination optimization;

[0008] Step S4: constructing a digital artwork generation model;

[0009] Step S5: Generate Jingchu style digital artwork.

[0010] Furthermore, in step S1, the Jingchu artwork collection is specifically to obtain the original data of Jingchu artworks through data collection from museums and art galleries; the original data of Jingchu artworks include image data of Jingchu artworks from different historical periods and historical data of Jingchu culture; the image data of Jingchu artworks include image data of bronze ware of Jingchu artworks, image data of pottery of Jingchu artworks, painting data of Jingchu culture, image data of sculpture art of Jingchu culture and image data of Jingchu folk artworks.

[0011] Furthermore, in step S2, the data preprocessing is specifically used to perform image data preprocessing, image data annotation and image feature extraction on the original data of Jingchu artworks to obtain standardized artwork generation data;

[0012] The image data preprocessing includes image denoising, image enhancement, image expansion, image alignment and image standardization;

[0013] The image denoising specifically comprises removing noise from the image by using a median filtering algorithm and enhancing image details by sharpening processing;

[0014] The image enhancement is specifically performed by normalizing and unifying the contrast, color saturation and brightness of the image;

[0015] The image expansion specifically includes performing random geometric transformation of the image to balance the image data of each category;

[0016] The image alignment specifically includes removing the blank and irrelevant areas of the image to ensure that the main part of the artwork is centered;

[0017] The image standardization specifically refers to adjusting the size of the image to a uniform specification;

[0018] The image data annotation specifically includes annotating each artwork in detail according to the Jingchu cultural history data, indicating the artwork type, historical period, artistic style characteristics and cultural symbols of each image in the Jingchu artwork image data;

[0019] The image feature extraction specifically uses a convolutional neural network to extract the features of Jingchu cultural artworks in the image, and the features of Jingchu cultural artworks include color, texture, form, pattern and cultural symbol features.

[0020] Further, in step S3, the Jingchu style feature combination optimization includes designing an objective function, obtaining an optimal feature weight and a Jingchu style feature combination; specifically, the following steps are included:

[0021] Step S31: Designing an objective function, specifically combining the characteristics of Jingchu culture to construct an objective function that comprehensively considers style conformity and cultural fit; including the following steps:

[0022] Step S311: Jingchu style conformity, specifically measuring the feature difference between the generated artwork and the Jingchu style image, the formula used is as follows:

[0023] ;

[0024] In the formula, represents the style conformity calculation function, represents the number of features, It means that under weight w, the feature In the performance of existing artwork images, Indicates the image of Jingchu style artwork in performance on a characteristic;

[0025] Step S312: Compatibility of Jingchu culture, the formula used is as follows:

[0026] ;

[0027] In the formula, represents the cultural fit calculation function, Indicates that culture meets quantity, It indicates the performance of the qth cultural symbol in the existing artwork image under the weight w. Indicates the expression of the qth cultural symbol in the Jingchu style image;

[0028] Step S313: Obtain the comprehensive objective function, the formula used is as follows:

[0029] ;

[0030] In the formula, represents the comprehensive objective function, The weight coefficient representing the conformity of the Jingchu style, The weight coefficient indicating the compatibility of Jingchu culture;

[0031] Step S32: obtaining the optimal feature weight, specifically obtaining the optimal weight of the Jingchu style feature by improving the optimization algorithm, specifically including the following steps:

[0032] Step S321: Initialize parameters, specifically by constructing initial algorithm parameters; the initial algorithm parameters include the number of particles N and the maximum number of iterations ;

[0033] Step S322: Initialize the particle swarm, specifically randomly generate particle positions, and the particle positions represent a set of Jingchu style feature weights; the formula used is as follows:

[0034] ;

[0035] In the formula, represents the position of the ith particle, represents the color feature weight parameter, represents the texture feature weight parameter, represents the shape feature weight parameter, represents the pattern feature weight parameter, represents the weight parameter of cultural symbol feature;

[0036] Step S323: Calculate the fitness value, specifically, calculate the fitness value of the particles in the particle swarm through the comprehensive objective function , evaluates the quality of the current particle position;

[0037] Step S324: Obtain the particle velocity weight value, using the following formula:

[0038] ;

[0039] In the formula, represents the particle velocity weight value of the tth iteration, Indicates the maximum value of the particle velocity weight. indicates the minimum value of the particle velocity weight, t indicates the current number of iterations, Represents the adjustment factor, which is used to control the rate at which the particle velocity weight decreases. Indicates the adjustment amplitude used to enhance the particle velocity weight value;

[0040] Step S325: Obtain learning factors, specifically, obtain individual learning factors and group learning factors, and the formula used is as follows:

[0041] ;

[0042] ;

[0043] In the formula, represents the individual learning factor of the tth iteration, represents the population learning factor of the tth iteration, and represents the minimum and maximum values ​​of the individual learning factor, and Indicates the minimum and maximum values ​​of the group learning factor; and Used to control and The decay rate of and Respectively represent control and The decay rate of and They are used to adjust the phase of the sine function and the cosine function respectively;

[0044] Step S326: Update the particle velocity using the following formula:

[0045] ;

[0046] In the formula, represents the velocity of the i-th particle in the t+1th iteration, represents the velocity of the ith particle in the tth iteration, represents the position of the i-th particle in the t-th iteration, represents the local optimal position of individual particles, represents the global optimal position of the particle, and Represents a random number in the range [0,1], represents the noise term parameter;

[0047] Step S327: Update the particle position. The formula used is as follows:

[0048] ;

[0049] In the formula, represents the position of the i-th particle in the t+1-th iteration;

[0050] Step S328: Update the optimal position of the particle. The formula used is as follows:

[0051] ;

[0052] ;

[0053] In the formula, represents the updated local optimal position of the individual particle, represents the updated global optimal position of the particle;

[0054] Step S329: search and determination, specifically, by constructing a search termination condition, searching and determining the global optimal position of the particle, and obtaining the data setting of the global optimal position of the particle;

[0055] The search termination conditions include threshold termination and iteration termination;

[0056] The threshold termination is specifically to set the fitness threshold, when the particle fitness value f i When it is above the fitness threshold, the search is completed;

[0057] The iteration termination specifically refers to terminating the iteration and obtaining the global optimal position of the particle when the maximum number of iterations is reached;

[0058] The global optimal position of the particle specifically refers to the optimal weight of the Jingchu style feature;

[0059] Step S33: Obtaining the comprehensive features of the Jingchu style, specifically adjusting the features of the Jingchu cultural artworks according to the optimal weights of the Jingchu style features to obtain the comprehensive features of the Jingchu style, the formula used is as follows:

[0060] ;

[0061] In the formula, It represents the comprehensive characteristics of Jingchu style. Indicates color characteristics, Represents texture features, Represents shape characteristics, Represents pattern features, Represents cultural symbolic characteristics.

[0062] Further, in step S4, the digital artwork generation model is constructed based on a conditional generative adversarial network, including the following steps:

[0063] Step S41: define the conditional input of the generator, specifically including conditional features and noise , the conditional feature is the comprehensive feature of Jingchu style, and the noise input is a 128-dimensional random noise vector that conforms to uniform distribution;

[0064] Step S42: Construct a generator for converting the conditional features and the noise vector into a digital artwork image that conforms to the Jingchu style. Specifically, the generator receives the conditional features and noise The input is mapped to a feature space of size 1024 through a fully connected layer, and then the image is gradually generated through four convolutional layers and upsampling operations. Each convolution operation can be nonlinearly transformed through the ReLU activation function. The output of the generator is a size of RGB image , the output image represents a Jingchu style digital artwork that matches the given conditional features;

[0065] Step S43: Construct a discriminator, specifically the discriminator accepts the output image of the generator and conditional features As input data, the input image is gradually feature extracted through four convolutional layers, downsampled after the convolution operation, and finally converted into a single output through a fully connected layer. The output is a probability value representing the authenticity of the image, which indicates whether the image conforms to the given Jingchu style. The discriminator feeds back the authenticity of the image to the generator;

[0066] Step S44: Designing a loss function, specifically including the following steps:

[0067] Step S441: introduce a diversity loss term to measure the diversity between generated images. The formula used is as follows:

[0068] ;

[0069] In the formula, represents the diversity loss function, Indicates that the generator is The generated image, Indicates that the generator is The generated image, represents the i-th noise, represents the jth noise;

[0070] Step S442: Design the generator loss function, the formula used is as follows:

[0071] ;

[0072] In the formula, represents the generator loss function, D represents the discriminator function, Represents the weight parameter of the diversity loss term;

[0073] Step S443: Design the discriminator penalty term, the formula used is as follows:

[0074] ;

[0075] ;

[0076] ;

[0077] In the formula, Represents the penalty term of the discriminator in processing real images, represents the penalty term of the discriminator when processing the generated image, represents the total penalty term of the discriminator, represents the control penalty intensity factor, Represents the real image and the corresponding conditional features The gradient of represents the L2 norm, represents a hyperparameter that controls the sensitivity of the gradient penalty. Represents the gradient of the generated image and the conditional features;

[0078] Step S444: Design the discriminator loss function, the formula used is as follows:

[0079] ;

[0080] In the formula, represents the discriminator loss function, Represents the discriminator for the real image and the corresponding conditional features The output, Represents the weight parameter of the total penalty term of the discriminator;

[0081] Step S45: Perform model training, specifically using the standardized artwork generation data as training data for the digital artwork generation model, by minimizing the generator loss function and minimizing the discriminator loss function, the generator and the discriminator use an adversarial training method to perform model training, and obtain a trained digital artwork generation model.

[0082] Furthermore, in step S5, the Jingchu style digital artwork is generated, specifically based on the trained digital artwork generation model, by inputting optimized Jingchu style comprehensive features and random noise vectors to obtain Jingchu style digital artwork.

[0083] The technical solution adopted by the present invention is as follows: the present invention provides a Jingchu style digital artwork generation system based on artificial intelligence, including a Jingchu artwork collection module, a data preprocessing module, a Jingchu style feature combination optimization module, a digital artwork generation model construction module and a Jingchu style digital artwork generation module;

[0084] The Jingchu artwork collection module obtains the original data of Jingchu artworks by collecting image data of Jingchu artworks and historical data of Jingchu culture from different historical periods from museums and art galleries, and sends the original data of Jingchu artworks to the data preprocessing module;

[0085] The data preprocessing module receives the data sent by the data acquisition module, and performs image data preprocessing, image data annotation and image feature extraction on the original data of the Jingchu artwork to obtain standardized artwork generation data, and sends the standardized artwork generation data to the Jingchu style feature combination optimization module and the digital artwork generation model construction module;

[0086] The Jingchu style feature combination optimization module receives the standardized artwork generation data sent by the data preprocessing module, and obtains the optimal feature weight by designing the objective function and the improved optimization algorithm, obtains the optimal weight of the Jingchu style feature, adjusts the Jingchu cultural artwork feature according to the Jingchu style feature optimal weight, obtains the Jingchu style comprehensive feature, and sends the Jingchu style comprehensive feature to the digital artwork generation model construction module and the Jingchu style digital artwork generation module;

[0087] The module for constructing a digital artwork generation model receives data sent by the data preprocessing module and the Jingchu style feature combination optimization module, constructs a generator and a discriminator model based on a conditional generative network, optimizes the generator and the discriminator using an adversarial training method, obtains a trained digital artwork generation model, and sends the trained digital artwork generation model to the Jingchu style digital artwork generation module;

[0088] The Jingchu style digital artwork generation module receives the results sent by the digital artwork generation model construction module and the Jingchu style feature combination optimization module, and obtains the Jingchu style digital artwork based on the trained digital artwork generation model and input data.

[0089] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0090] (1) In view of the problem that in the traditional style digital artwork generation process, the extraction of special cultural style features usually needs to be performed dynamically during each generation, resulting in low generation efficiency, poor style consistency and excessive computational burden, this scheme creatively proposes to obtain and optimize the Jingchu style feature combination in advance, and use the optimized features as input to avoid repeated calculations in each generation process, thereby improving the generation efficiency, ensuring the style consistency of the generated works, and reducing the consumption of computing resources. The pre-optimized Jingchu style features can also enhance the control of the generation model over the style details, making the generated Jingchu style digital artwork more in line with expectations and having higher artistic and cultural accuracy.

[0091] (2) In the process of optimizing the existing feature combination, there is an improper setting of feature weights, which leads to the fact that the artwork generated by the model does not meet the expected technical problem. This solution proposes to improve the particle swarm optimization algorithm, combine the method of adjusting the particle velocity weight value and the dynamic adjustment of the learning factor, and optimize the selection of feature weights. This can effectively obtain the optimal feature weight combination, thereby ensuring that the generated artwork is more in line with expectations, achieving consistency in style and improved artistic effect.

[0092] (3) In view of the technical problems in the existing digital artwork generation models, such as poor style consistency, lack of diversity in the generated results, and unstable generation process, this scheme innovatively optimizes them by introducing diversity loss terms and discriminator penalty terms. The diversity loss term ensures that the generated images have sufficient variability in style and details, effectively avoiding the phenomenon of generating repeated similar images. The introduction of the discriminator penalty term enhances the stability of the discriminator and promotes the generator to learn richer and more realistic artistic features, so that the generated Jingchu style digital artworks have been significantly improved in terms of style consistency, artistic value and creative diversity. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 A schematic diagram of a process for generating digital artworks in the Jingchu style based on artificial intelligence provided by the present invention;

[0094] Figure 2 A schematic diagram of a module of a Jingchu style digital artwork generation system based on artificial intelligence provided by the present invention;

[0095] Figure 3 This is a schematic diagram of the process of optimizing the combination of Jingchu style features in step S3;

[0096] Figure 4 A schematic diagram of the process of constructing a digital artwork generation model in step S4;

[0097] Figure 5 A schematic diagram of a process for obtaining the optimal feature weight in step S32;

[0098] Figure 6 A schematic diagram of a process for designing a loss function for step S44;

[0099] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0100] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0101] In the description of the present invention, it is necessary to understand that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred system or element must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0102] Example 1, see Figure 1 The present invention provides a method for generating digital artworks in Jingchu style based on artificial intelligence, the method comprising the following steps:

[0103] Step S1: Collection of Jingchu artworks;

[0104] Step S2: data preprocessing;

[0105] Step S3: Jingchu style feature combination optimization;

[0106] Step S4: constructing a digital artwork generation model;

[0107] Step S5: Generate Jingchu style digital artwork.

[0108] By performing the above operations, in order to address the problem that in the traditional style digital artwork generation process, the extraction of special cultural style features usually needs to be performed dynamically during each generation, resulting in low generation efficiency, poor style consistency and excessive computational burden, this scheme creatively proposes to obtain and optimize the Jingchu style feature combination in advance, and use the optimized features as input to avoid repeated calculations in each generation process, thereby improving the generation efficiency, ensuring the style consistency of the generated works, and reducing the consumption of computing resources. The pre-optimized Jingchu style features can also enhance the control of the generation model over the style details, so that the generated Jingchu style digital artwork is more in line with expectations and has higher artistic and cultural accuracy.

[0109] Example 2, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S1, the Jingchu artwork collection is specifically to obtain the original data of Jingchu artworks from museums and art galleries through data collection; the original data of Jingchu artworks include image data of Jingchu artworks in different historical periods and historical data of Jingchu culture; the image data of Jingchu artworks include image data of Jingchu artworks bronzes, image data of Jingchu artworks pottery, image data of Jingchu culture paintings, image data of Jingchu culture sculptures and image data of Jingchu folk artworks; the image data of Jingchu artworks bronzes include image data of tripods, image data of jars, image data of bells, image data of inscription vessels image data and weapon image data; the Jingchu artwork pottery image data includes pottery image data, pottery figurine image data, pottery sacrificial vessel image data and pottery pot image data; the Jingchu culture painting data includes murals, silk paintings and portraits; the Jingchu culture sculpture art image data includes stone carving image data, wood carving image data, bronze carving image data and stone tablet image data; the Jingchu folk art image data includes embroidery image data, fabric image data, folk pottery image data and handicraft image data; the Jingchu cultural history data includes the historical background of Jingchu artwork, the interpretation of cultural symbols and the artistic style evolution data.

[0110] Example 3, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, in step S2, the data preprocessing is specifically used to perform image data preprocessing, image data annotation and image feature extraction on the original data of Jingchu artworks to obtain standardized artwork generation data;

[0111] The image data preprocessing includes image denoising, image enhancement, image expansion, image alignment and image standardization;

[0112] The image denoising specifically comprises removing noise from the image by using a median filtering algorithm and enhancing image details by sharpening processing;

[0113] The image enhancement is specifically performed by normalizing and unifying the contrast, color saturation and brightness of the image;

[0114] The image expansion specifically includes performing random geometric transformation of the image to balance the image data of each category;

[0115] The image alignment specifically includes removing the blank and irrelevant areas of the image to ensure that the main part of the artwork is centered;

[0116] The image standardization specifically refers to adjusting the size of the image to a uniform specification;

[0117] The image data annotation specifically includes annotating each artwork in detail according to the Jingchu cultural history data, indicating the artwork type, historical period, artistic style characteristics and cultural symbols of each image in the Jingchu artwork image data;

[0118] The image feature extraction specifically uses a convolutional neural network to extract the features of Jingchu cultural artworks in the image, and the features of Jingchu cultural artworks include color, texture, form, pattern and cultural symbol features.

[0119] Example 4, see Figure 1 , Figure 2 , Figure 3 and Figure 5 This embodiment is based on the above embodiment. In step S3, the Jingchu style feature combination optimization includes designing an objective function, obtaining an optimal feature weight and a Jingchu style feature combination; specifically, the following steps are included:

[0120] Step S31: Designing an objective function, specifically combining the characteristics of Jingchu culture to construct an objective function that comprehensively considers style conformity and cultural fit; including the following steps:

[0121] Step S311: Jingchu style conformity, specifically measuring the feature difference between the generated artwork and the Jingchu style image, the formula used is as follows:

[0122] ;

[0123] In the formula, represents the style conformity calculation function, represents the number of features, It means that under weight w, the feature In the performance of existing artwork images, Indicates the image of Jingchu style artwork in performance on a characteristic;

[0124] Step S312: Compatibility of Jingchu culture, the formula used is as follows:

[0125] ;

[0126] In the formula, represents the cultural fit calculation function, Indicates that culture meets quantity, It indicates the performance of the qth cultural symbol in the existing artwork image under the weight w. Indicates the expression of the qth cultural symbol in the Jingchu style image;

[0127] Step S313: Obtain the comprehensive objective function, the formula used is as follows:

[0128] ;

[0129] In the formula, represents the comprehensive objective function, The weight coefficient representing the conformity of the Jingchu style, The weight coefficient indicating the compatibility of Jingchu culture;

[0130] Step S32: obtaining the optimal feature weight, specifically obtaining the optimal weight of the Jingchu style feature by improving the optimization algorithm, specifically including the following steps:

[0131] Step S321: Initialize parameters, specifically by constructing initial algorithm parameters; the initial algorithm parameters include the number of particles N and the maximum number of iterations ;

[0132] Step S322: Initialize the particle swarm, specifically randomly generate particle positions, and the particle positions represent a set of Jingchu style feature weights; the formula used is as follows:

[0133] ;

[0134] In the formula, represents the position of the ith particle, represents the color feature weight parameter, represents the texture feature weight parameter, represents the shape feature weight parameter, represents the pattern feature weight parameter, represents the weight parameter of cultural symbol feature;

[0135] Step S323: Calculate the fitness value, specifically, calculate the fitness value of the particles in the particle swarm through the comprehensive objective function , evaluates the quality of the current particle position;

[0136] Step S324: Obtain the particle velocity weight value, using the following formula:

[0137] ;

[0138] In the formula, represents the particle velocity weight value of the tth iteration, Indicates the maximum value of the particle velocity weight. indicates the minimum value of the particle velocity weight, t indicates the current number of iterations, Represents the adjustment factor, which is used to control the rate at which the particle velocity weight decreases. Indicates the adjustment amplitude used to enhance the particle velocity weight value;

[0139] Step S325: Obtain learning factors, specifically, obtain individual learning factors and group learning factors, and the formula used is as follows:

[0140] ;

[0141] ;

[0142] In the formula, represents the individual learning factor of the tth iteration, represents the population learning factor of the tth iteration, and represents the minimum and maximum values ​​of the individual learning factor, and Indicates the minimum and maximum values ​​of the group learning factor; and Used to control and The decay rate of and Respectively represent control and The decay rate of and They are used to adjust the phase of the sine function and the cosine function respectively;

[0143] Step S326: Update the particle velocity using the following formula:

[0144] ;

[0145] In the formula, represents the velocity of the i-th particle in the t+1th iteration, represents the velocity of the i-th particle in the t-th iteration, represents the position of the i-th particle in the t-th iteration, represents the local optimal position of individual particles, represents the global optimal position of the particle, and Represents a random number in the range [0,1], represents the noise term parameter;

[0146] Step S327: Update the particle position. The formula used is as follows:

[0147] ;

[0148] In the formula, represents the position of the i-th particle in the t+1-th iteration;

[0149] Step S328: Update the optimal position of the particle. The formula used is as follows:

[0150] ;

[0151] ;

[0152] In the formula, represents the updated local optimal position of the individual particle, represents the updated global optimal position of the particle;

[0153] Step S329: search and determination, specifically, by constructing a search termination condition, searching and determining the global optimal position of the particle, and obtaining the data setting of the global optimal position of the particle;

[0154] The search termination conditions include threshold termination and iteration termination;

[0155] The threshold termination is specifically to set the fitness threshold, when the particle fitness value f i When it is above the fitness threshold, the search is completed;

[0156] The iteration termination specifically refers to terminating the iteration and obtaining the global optimal position of the particle when the maximum number of iterations is reached;

[0157] The global optimal position of the particle specifically refers to the optimal weight of the Jingchu style feature;

[0158] Step S33: Obtaining the comprehensive features of the Jingchu style, specifically adjusting the features of the Jingchu cultural artworks according to the optimal weights of the Jingchu style features to obtain the comprehensive features of the Jingchu style, the formula used is as follows:

[0159] ;

[0160] In the formula, It represents the comprehensive characteristics of Jingchu style. Indicates color characteristics, Represents texture features, Represents shape characteristics, Represents pattern features, Represents cultural symbolic characteristics.

[0161] By performing the above operations, in the process of optimizing the existing feature combination, there is an improper setting of feature weights, which leads to the fact that the artwork generated by the model does not meet the expected technical problem. This solution proposes to improve the particle swarm optimization algorithm, combine the method of adjusting the particle velocity weight value and the dynamic adjustment of the learning factor, and optimize the selection of feature weights. It can effectively obtain the optimal feature weight combination, thereby ensuring that the generated artwork is more in line with expectations, achieving consistency in style and improved artistic effects.

[0162] Example 5, see Figure 1 , Figure 2 , Figure 4 and Figure 6 This embodiment is based on the above embodiment. In step S4, the digital artwork generation model is specifically constructed based on a conditional generative adversarial network, including the following steps:

[0163] Step S41: define the conditional input of the generator, specifically including conditional features and noise , the conditional feature is the comprehensive feature of Jingchu style, and the noise input is a 128-dimensional random noise vector that conforms to uniform distribution;

[0164] Step S42: Construct a generator for converting the conditional features and the noise vector into a digital artwork image that conforms to the Jingchu style. Specifically, the generator receives the conditional features and noise The input is mapped to a feature space of size 1024 through a fully connected layer, and then the image is gradually generated through four convolutional layers and upsampling operations. Each convolution operation can be nonlinearly transformed through the ReLU activation function. The output of the generator is a size of RGB image , the output image represents a Jingchu style digital artwork that matches the given conditional features;

[0165] Step S43: Construct a discriminator, specifically the discriminator accepts the output image of the generator and conditional features As input data, the input image is gradually feature extracted through four convolutional layers, downsampled after the convolution operation, and finally converted into a single output through a fully connected layer. The output is a probability value representing the authenticity of the image, which indicates whether the image conforms to the given Jingchu style. The discriminator feeds back the authenticity of the image to the generator;

[0166] Step S44: Designing a loss function, specifically including the following steps:

[0167] Step S441: introduce a diversity loss term to measure the diversity between generated images. The formula used is as follows:

[0168] ;

[0169] In the formula, represents the diversity loss function, Indicates that the generator is The generated image, Indicates that the generator is The generated image, represents the i-th noise, represents the jth noise;

[0170] Step S442: Design the generator loss function, the formula used is as follows:

[0171] ;

[0172] In the formula, represents the generator loss function, D represents the discriminator function, Represents the weight parameter of the diversity loss term;

[0173] Step S443: Design the discriminator penalty term, the formula used is as follows:

[0174] ;

[0175] ;

[0176] ;

[0177] In the formula, Represents the penalty term of the discriminator in processing real images, represents the penalty term of the discriminator when processing the generated image, represents the total penalty term of the discriminator, represents the control penalty intensity factor, Represents the real image and the corresponding conditional features The gradient of represents the L2 norm, represents a hyperparameter that controls the sensitivity of the gradient penalty. Represents the gradient of the generated image and the conditional features;

[0178] Step S444: Design the discriminator loss function, the formula used is as follows:

[0179] ;

[0180] In the formula, represents the discriminator loss function, Represents the discriminator for the real image and the corresponding conditional features The output, Represents the weight parameter of the total penalty term of the discriminator;

[0181] Step S45: Perform model training, specifically using the standardized artwork generation data as training data for the digital artwork generation model, by minimizing the generator loss function and minimizing the discriminator loss function, the generator and the discriminator use an adversarial training method to perform model training, and obtain a trained digital artwork generation model.

[0182] By performing the above operations, this solution innovatively introduces diversity loss terms and discriminator penalty terms to optimize the existing digital artwork generation models, which have technical problems such as poor style consistency, lack of diversity in generation results, and unstable generation process. The diversity loss terms ensure that the generated images have sufficient variability in style and details, and effectively avoid the repetitive generation of similar images. The introduction of the discriminator penalty term enhances the stability of the discriminator and promotes the generator to learn richer and more realistic artistic features, so that the generated Jingchu style digital artworks are significantly improved in terms of style consistency, artistic value and creative diversity.

[0183] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the Jingchu style digital artwork is generated, specifically based on the trained digital artwork generation model, by inputting optimized Jingchu style comprehensive features and random noise vectors to obtain Jingchu style digital artwork.

[0184] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the technical solution adopted by the present invention is as follows: the present invention provides a Jingchu style digital artwork generation system based on artificial intelligence, including a Jingchu artwork collection module, a data preprocessing module, a Jingchu style feature combination optimization module, a digital artwork generation model construction module and a Jingchu style digital artwork generation module;

[0185] The Jingchu artwork collection module obtains the original data of Jingchu artworks by collecting image data of Jingchu artworks and historical data of Jingchu culture from different historical periods from museums and art galleries, and sends the original data of Jingchu artworks to the data preprocessing module;

[0186] The data preprocessing module receives the data sent by the data acquisition module, and performs image data preprocessing, image data annotation and image feature extraction on the original data of the Jingchu artwork to obtain standardized artwork generation data, and sends the standardized artwork generation data to the Jingchu style feature combination optimization module and the digital artwork generation model construction module;

[0187] The Jingchu style feature combination optimization module receives the standardized artwork generation data sent by the data preprocessing module, and obtains the optimal feature weight by designing the objective function and the improved optimization algorithm, obtains the optimal weight of the Jingchu style feature, adjusts the Jingchu cultural artwork feature according to the Jingchu style feature optimal weight, obtains the Jingchu style comprehensive feature, and sends the Jingchu style comprehensive feature to the digital artwork generation model construction module and the Jingchu style digital artwork generation module;

[0188] The module for constructing a digital artwork generation model receives data sent by the data preprocessing module and the Jingchu style feature combination optimization module, constructs a generator and a discriminator model based on a conditional generative network, optimizes the generator and the discriminator using an adversarial training method, obtains a trained digital artwork generation model, and sends the trained digital artwork generation model to the Jingchu style digital artwork generation module;

[0189] The Jingchu style digital artwork generation module receives the results sent by the digital artwork generation model construction module and the Jingchu style feature combination optimization module, and obtains the Jingchu style digital artwork based on the trained digital artwork generation model and input data.

[0190] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0191] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0192] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for generating digital artworks in Jingchu style based on artificial intelligence, characterized by: The method comprises the following steps: Step S1: collecting Jingchu artworks, specifically collecting Jingchu artworks from museums and art galleries to obtain original data; Step S2: data preprocessing, specifically, performing image data preprocessing, image data annotation and image feature extraction on the original data of Jingchu artworks to obtain standardized artwork generation data; Step S3: Jingchu style feature combination optimization, used to obtain the comprehensive features of Jingchu style after feature combination optimization, specifically, designing an objective function that comprehensively considers style conformity and cultural fit, and improving the particle swarm optimization algorithm by combining the method of particle velocity weight value adjustment and learning factor dynamic adjustment to obtain the optimal weight of Jingchu style features, and adjusting the features of Jingchu cultural artworks according to the optimal weight of Jingchu style features to obtain the comprehensive features of Jingchu style; comprising the following steps: Step S31: designing an objective function; Step S32: obtaining optimal feature weights; Step S33: obtaining comprehensive features of the Jingchu style; Step S4: constructing a digital artwork generation model, specifically, building a generator and a discriminator model based on a conditional generation network, designing a generator loss function and a discriminator loss function in combination with a diversity loss term and a discriminator penalty term in the model design, and finally optimizing the parameters of the generator and the discriminator through an adversarial training method to obtain a trained digital artwork generation model; comprising the following steps: Step S41: define conditional input of the generator; Step S42: construct the generator; Step S43: construct the discriminator; Step S44: design the loss function; Step S45: perform model training; Step S5: generating Jingchu style digital artwork, specifically, based on the trained digital artwork generation model, inputting data to obtain Jingchu style digital artwork.

2. The method for generating digital artwork in Jingchu style based on artificial intelligence according to claim 1, characterized in that: In step S3, the Jingchu style feature combination optimization includes designing an objective function, obtaining an optimal feature weight and a Jingchu style feature combination; specifically, the following steps are included: Step S31: Designing an objective function, specifically combining the characteristics of Jingchu culture to construct an objective function that comprehensively considers style conformity and cultural fit; including the following steps: Step S311: Jingchu style conformity, specifically measuring the feature difference between the generated artwork and the Jingchu style image, the formula used is as follows: ； In the formula, represents the style conformity calculation function, represents the number of features, It means that under weight w, the feature In the performance of existing artwork images, Indicates the image of Jingchu style artwork in performance on a characteristic; Step S312: Compatibility of Jingchu culture, the formula used is as follows: ； In the formula, represents the cultural fit calculation function, Indicates that culture meets quantity, It indicates the performance of the qth cultural symbol in the existing artwork image under the weight w. Indicates the expression of the qth cultural symbol in the Jingchu style image; Step S313: Obtain the comprehensive objective function, the formula used is as follows: ； In the formula, represents the comprehensive objective function, The weight coefficient representing the conformity of the Jingchu style, The weight coefficient indicating the compatibility of Jingchu culture; Step S32: obtaining the optimal feature weight, specifically obtaining the optimal weight of the Jingchu style feature by improving the optimization algorithm, specifically including the following steps: Step S321: Initialize parameters, specifically by constructing initial algorithm parameters; the initial algorithm parameters include the number of particles N and the maximum number of iterations ; Step S322: Initialize the particle swarm, specifically randomly generate particle positions, and the particle positions represent a set of Jingchu style feature weights; the formula used is as follows: ； In the formula, represents the position of the ith particle, represents the color feature weight parameter, represents the texture feature weight parameter, represents the shape feature weight parameter, represents the pattern feature weight parameter, represents the weight parameter of cultural symbol feature; Step S323: Calculate the fitness value, specifically, calculate the fitness value of the particles in the particle swarm through the comprehensive objective function , evaluates the quality of the current particle position; Step S324: Obtain the particle velocity weight value, using the following formula: ； In the formula, represents the particle velocity weight value of the tth iteration, Indicates the maximum value of the particle velocity weight. indicates the minimum value of the particle velocity weight, t indicates the current number of iterations, Represents the adjustment factor, which is used to control the rate at which the particle velocity weight decreases. Indicates the adjustment amplitude used to enhance the particle velocity weight value; Step S325: Obtain learning factors, specifically, obtain individual learning factors and group learning factors, and the formula used is as follows: ； ； In the formula, represents the individual learning factor of the tth iteration, represents the population learning factor of the tth iteration, and represents the minimum and maximum values ​​of the individual learning factor, and Indicates the minimum and maximum values ​​of the group learning factor; and Used to control and The decay rate of and Respectively represent control and The decay rate of and They are used to adjust the phase of the sine function and the cosine function respectively; Step S326: Update the particle velocity using the following formula: ； In the formula, represents the velocity of the i-th particle in the t+1th iteration, represents the velocity of the i-th particle in the t-th iteration, represents the position of the i-th particle in the t-th iteration, represents the local optimal position of individual particles, represents the global optimal position of the particle, and represents a random number in the range [0,1], represents the noise term parameter; Step S327: Update the particle position. The formula used is as follows: ； In the formula, represents the position of the i-th particle in the t+1-th iteration; Step S328: Update the optimal position of the particle. The formula used is as follows: ； ； In the formula, represents the updated local optimal position of the individual particle, represents the updated global optimal position of the particle; Step S329: search and determination, specifically, by constructing a search termination condition, searching and determining the global optimal position of the particle, and obtaining the data setting of the global optimal position of the particle; The search termination conditions include threshold termination and iteration termination; The threshold termination is specifically to set a fitness threshold, when the particle fitness value f i When it is above the fitness threshold, the search is completed; The iteration termination specifically refers to terminating the iteration and obtaining the global optimal position of the particle when the maximum number of iterations is reached; The global optimal position of the particle specifically refers to the optimal weight of the Jingchu style feature; Step S33: Obtaining the comprehensive features of the Jingchu style, specifically adjusting the features of the Jingchu cultural artworks according to the optimal weights of the Jingchu style features to obtain the comprehensive features of the Jingchu style, the formula used is as follows: ； In the formula, It represents the comprehensive characteristics of Jingchu style. Indicates color characteristics, Represents texture features, Represents shape characteristics, Represents pattern features, Represents cultural symbolic characteristics.

3. The method for generating digital artwork in Jingchu style based on artificial intelligence according to claim 1, characterized in that: In step S4, the digital artwork generation model is specifically constructed based on a conditional generative adversarial network, including the following steps: Step S41: define the conditional input of the generator, specifically including conditional features and noise , the conditional feature is the comprehensive feature of Jingchu style, and the noise input is a 128-dimensional random noise vector that conforms to uniform distribution; Step S42: Construct a generator for converting the conditional features and the noise vector into a digital artwork image that conforms to the Jingchu style. Specifically, the generator receives the conditional features and noise The input is mapped to a feature space of size 1024 through a fully connected layer, and then the image is gradually generated through four convolutional layers and upsampling operations. Each convolution operation can be nonlinearly transformed through the ReLU activation function. The output of the generator is a size of RGB image , the output image represents a digital artwork in the Jingchu style that matches the given conditional features; Step S43: Construct a discriminator, specifically the discriminator accepts the output image of the generator and conditional features As input data, the input image is gradually feature extracted through four convolutional layers, downsampled after the convolution operation, and finally converted into a single output through a fully connected layer. The output is a probability value representing the authenticity of the image, which indicates whether the image conforms to the given Jingchu style. The discriminator feeds back the authenticity of the image to the generator; Step S44: Designing a loss function, specifically including the following steps: Step S441: introduce a diversity loss term to measure the diversity between generated images. The formula used is as follows: ； In the formula, represents the diversity loss function, Indicates that the generator is The generated image, Indicates that the generator is The generated image, represents the i-th noise, represents the jth noise; Step S442: Design the generator loss function, the formula used is as follows: ； In the formula, represents the generator loss function, D represents the discriminator function, Represents the weight parameter of the diversity loss term; Step S443: Design the discriminator penalty term, the formula used is as follows: ； ； ； In the formula, Represents the penalty term of the discriminator in processing real images, represents the penalty term of the discriminator when processing the generated image, represents the total penalty term of the discriminator, represents the control penalty intensity factor, Represents the real image and the corresponding conditional features The gradient of represents the L2 norm, represents a hyperparameter that controls the sensitivity of the gradient penalty. Represents the gradient of the generated image and the conditional features; Step S444: Design the discriminator loss function, the formula used is as follows: ； In the formula, represents the discriminator loss function, Represents the discriminator for the real image and the corresponding conditional features The output, Represents the weight parameter of the total penalty term of the discriminator; Step S45: Perform model training, specifically using the standardized artwork generation data as training data for the digital artwork generation model, by minimizing the generator loss function and minimizing the discriminator loss function, the generator and the discriminator use an adversarial training method to perform model training, and obtain a trained digital artwork generation model.

4. The method for generating digital artwork in Jingchu style based on artificial intelligence according to claim 1, characterized in that: In step S5, the Jingchu style digital artwork is generated, specifically based on the trained digital artwork generation model, by inputting optimized Jingchu style comprehensive features and random noise vectors to obtain Jingchu style digital artwork.

5. The method for generating digital artwork in Jingchu style based on artificial intelligence according to claim 1 is characterized by: In step S1, the Jingchu artwork collection is specifically to obtain the original data of Jingchu artworks through data collection from museums and art galleries; the original data of Jingchu artworks include image data of Jingchu artworks from different historical periods and historical data of Jingchu culture; the image data of Jingchu artworks include image data of bronze ware of Jingchu artworks, image data of pottery of Jingchu artworks, painting data of Jingchu culture, image data of sculpture art of Jingchu culture and image data of Jingchu folk artworks.

6. The method for generating digital artwork in Jingchu style based on artificial intelligence according to claim 1, characterized in that: In step S2, the data preprocessing is specifically used to perform image data preprocessing, image data annotation and image feature extraction on the original data of Jingchu artworks to obtain standardized artwork generation data; The image data preprocessing includes image denoising, image enhancement, image expansion, image alignment and image standardization; The image denoising specifically comprises removing noise from the image by using a median filtering algorithm and enhancing image details by sharpening processing; The image enhancement is specifically performed by normalizing and unifying the contrast, color saturation and brightness of the image; The image expansion specifically includes performing random geometric transformation of the image to balance the image data of each category; The image alignment specifically includes removing the blank and irrelevant areas of the image to ensure that the main part of the artwork is centered; The image standardization specifically refers to adjusting the size of the image to a uniform specification; The image data annotation specifically includes annotating each artwork in detail according to the Jingchu cultural history data, indicating the artwork type, historical period, artistic style characteristics and cultural symbols of each image in the Jingchu artwork image data; The image feature extraction specifically uses a convolutional neural network to extract the features of Jingchu cultural artworks in the image, and the features of Jingchu cultural artworks include color, texture, form, pattern and cultural symbol features.

7. A system for generating digital artworks in the style of Jingchu based on artificial intelligence, used to implement a method for generating digital artworks in the style of Jingchu based on artificial intelligence as described in any one of claims 1 to 6, characterized in that: It includes Jingchu artwork collection module, data preprocessing module, Jingchu style feature combination optimization module, digital artwork generation model building module and Jingchu style digital artwork generation module.

8. The artificial intelligence-based Jingchu style digital artwork generation system according to claim 7 is characterized by: The Jingchu artwork collection module obtains the original data of Jingchu artworks by collecting image data of Jingchu artworks and historical data of Jingchu culture from different historical periods from museums and art galleries, and sends the original data of Jingchu artworks to the data preprocessing module; The data preprocessing module receives the data sent by the data acquisition module, and performs image data preprocessing, image data annotation and image feature extraction on the original data of the Jingchu artwork to obtain standardized artwork generation data, and sends the standardized artwork generation data to the Jingchu style feature combination optimization module and the digital artwork generation model construction module; The Jingchu style feature combination optimization module receives the standardized artwork generation data sent by the data preprocessing module, and obtains the optimal feature weight by designing the objective function and the improved optimization algorithm, obtains the optimal weight of the Jingchu style feature, adjusts the Jingchu cultural artwork feature according to the Jingchu style feature optimal weight, obtains the Jingchu style comprehensive feature, and sends the Jingchu style comprehensive feature to the digital artwork generation model construction module and the Jingchu style digital artwork generation module; The module for constructing a digital artwork generation model receives data sent by the data preprocessing module and the Jingchu style feature combination optimization module, constructs a generator and a discriminator model based on a conditional generative network, optimizes the generator and the discriminator using an adversarial training method, obtains a trained digital artwork generation model, and sends the trained digital artwork generation model to the Jingchu style digital artwork generation module; The Jingchu style digital artwork generation module receives the results sent by the digital artwork generation model construction module and the Jingchu style feature combination optimization module, and obtains the Jingchu style digital artwork based on the trained digital artwork generation model and input data.

Images