Advertisement creative visual element intelligent generation method and device

Abstract

The invention relates to an advertisement creative visual element intelligent generation method and device. The method comprises the following steps: performing feature extraction and feature fusion on advertisement design materials to obtain a visual feature triple combination; performing multi-head attention coding and similarity calculation on the visual feature triple combination to obtain an element vector group; inputting the visual feature triple combination and the element vector group into a double-coding framework for bidirectional attention fusion to obtain a multi-modal feature group; performing depth separable convolution processing and feature separation on the multi-modal feature group to obtain a target feature group; and performing neighborhood screening and reconstruction calculation on the target feature group based on a local linear embedding algorithm to generate advertisement creative visual elements. According to the method, the quality and innovativeness of the generated visual elements are ensured, and the personalized customization capability of the advertisement creativity is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent generation of visual elements, and in particular to a method and device for intelligent generation of creative advertising visual elements. Background Art

[0002] With the rapid development of digital advertising, the design and generation of creative visual elements in advertising has become a key link in advertising production. Traditional advertising visual element design mainly relies on the experience and intuition of designers, which not only has low work efficiency, but also makes it difficult to ensure the consistency of design quality and the novelty of creativity.

[0003] Existing intelligent advertising design systems mainly use a single feature extraction method and a fixed model structure, which cannot effectively capture the diverse visual features in advertising design materials. In addition, the computational complexity is high when processing large-scale advertising materials, and the model parameters are redundant, which affects the practicality and promotion of the system. In addition, the existing methods have limited semantic understanding and feature representation capabilities of visual elements, making it difficult to accurately grasp the creative intent and visual effects of advertising design. In the process of processing feature fusion and generation of advertising visual elements, existing technologies often ignore the correlation and hierarchy between features, resulting in the lack of logic and artistry in the generated visual elements. At the same time, due to the lack of effective feature decoupling and knowledge transfer mechanisms, it is difficult to achieve personalized customization and style transfer of advertising creativity, which limits the application scenarios and innovation space of intelligent advertising design systems. Summary of the invention

[0004] The main purpose of the present invention is to provide a method and device for intelligently generating visual elements of advertising creativity. The present invention ensures the quality and innovation of generated visual elements and improves the personalized customization capability of advertising creativity.

[0005] To achieve the above object, the present invention provides a method for intelligently generating creative visual elements of advertisements, comprising the following steps: Perform feature extraction and feature fusion on the advertising design materials to obtain a triple combination of visual features; Performing multi-head attention encoding and similarity calculation on the visual feature triple combination to obtain an element vector group; Inputting the visual feature triple combination and the element vector group into a dual encoding framework for bidirectional attention fusion to obtain a multimodal feature group; Performing depth-separable convolution processing and feature separation on the multimodal feature group to obtain a target feature group; Based on the local linear embedding algorithm, the target feature group is subjected to neighborhood screening and reconstruction calculation to generate advertising creative visual elements.

[0006] The present invention also provides an intelligent generation device for creative advertising visual elements, comprising: A feature extraction module is used to extract and fuse features of the advertising design material to obtain a triple combination of visual features; A calculation module, used for performing multi-head attention encoding and similarity calculation on the visual feature triple combination to obtain an element vector group; An attention fusion module, used for inputting the visual feature triple combination and the element vector group into a dual encoding framework for bidirectional attention fusion to obtain a multimodal feature group; A feature separation module, used for performing deep separable convolution processing and feature separation on the multimodal feature group to obtain a target feature group; The generation module is used to perform neighborhood screening and reconstruction calculation on the target feature group based on a local linear embedding algorithm to generate advertising creative visual elements.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention effectively improves the ability to capture and understand visual features in advertising design materials and enhances the model's ability to express the semantic relationship of visual elements by constructing an extraction mechanism for visual feature triple combinations, combining multi-head attention encoding and bidirectional attention fusion; the lightweight design of deep separable convolution and temporal displacement modules significantly reduces the computational complexity and storage overhead of the model while maintaining the accuracy of feature extraction and improving the practicality of the system; the feature decoupling self-knowledge distillation network is introduced to achieve parameter compression and performance optimization of the model through the separation of target features and non-target features and knowledge transfer, thereby enhancing the generalization ability of the system; the adaptive neighborhood selection and reconstruction calculation mechanism based on the local linear embedding algorithm ensures the quality and innovation of the generated visual elements and improves the personalized customization ability of advertising creativity. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a schematic diagram of the steps of a method for intelligently generating creative visual elements of advertisements in one embodiment of the present invention; Figure 2 This is a structural block diagram of a device for intelligently generating creative visual elements of advertisements in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 This embodiment provides a method for intelligently generating creative visual elements of advertisements, comprising the following steps: S1, extract and fuse features of the advertising design materials to obtain a triple combination of visual features; Among them, the pixel size of the advertising design material is adjusted, and the original material is uniformly adjusted to a fixed-size RGB three-channel image to ensure the consistency and computational efficiency of subsequent processing. For the RGB three-channel image, a method based on image processing and deep learning is used to extract color distribution, texture features and edge features. The color distribution is obtained by calculating the distribution characteristics of pixels in the RGB space, and modeled using the histogram analysis method or the initial feature map of the convolutional neural network. The texture features are extracted through the local binary pattern (LBP), gray level co-occurrence matrix (GLCM) or the middle-level features of the deep network to capture the detailed information of the image surface. The extraction of edge features is combined with Canny edge detection or gradient analysis based on convolutional features to clarify the significant boundaries of the structure in the image. Through the above process, a first feature matrix containing color, texture and edge information is obtained to describe the basic visual characteristics of the image. Through the bidirectional feature propagation mechanism in the deep neural network, the first feature matrix is ​​forward feature transferred and backward feature transferred. The forward feature transfer uses the method of propagating features layer by layer to the deep layer to enhance the global characteristics, and the backward feature transfer strengthens the important local feature response through the back propagation mechanism. The bidirectional feature transfer mechanism helps to fully integrate features from different dimensions and obtain a second feature matrix with more global expression ability. The node correlation degree of the feature nodes in the second feature matrix is ​​calculated. The calculation of the node correlation degree is based on a specific similarity function or attention mechanism. By analyzing the mutual relationship and importance distribution between the feature nodes, an attention coefficient matrix is ​​generated to clarify the importance and relevance of different feature nodes in the overall feature space. The second feature matrix is ​​weighted fused using the attention coefficient matrix to integrate multiple feature information. Weighted fusion assigns different weights to each feature node, retains features with high relevance and high importance, and obtains the third feature matrix. The node correlation degree of the feature nodes in the third feature matrix is ​​calculated. The calculation method relies on the optimization of the attention mechanism to generate the importance score of the node. Based on the importance score of the node, the feature nodes are screened and combined, and nodes with higher importance scores are retained to reduce the interference of redundant features. Through screening, a candidate feature combination set is formed. The candidate feature combination set is classified and processed, and the features are divided into three categories: subject features, relationship features, and object features. The subject feature is the representation of the core visual elements in the image, the relationship feature describes the relationship between the subject and the background or other elements, and the object feature represents the background information or auxiliary elements. Based on the classification results, the candidate features are classified and combined in combination with the deep learning classification model or manually designed rules to finally form a complete visual feature triple combination.

[0014] S2, multi-head attention encoding and similarity calculation are performed on the combination of visual feature triplets to obtain an element vector group; Specifically, a visual feature encoding layer with 8 attention heads and 512-dimensional hidden layers is added to the initial S-BERT model to expand it into a target S-BERT model. The visual feature triples are combined and input into the target S-BERT model. The input features are mapped to the new feature space through a set of 512×512-dimensional linear projection matrices to obtain a 1024-dimensional initial feature vector. The conversion process strengthens the expressive power of the features through linear projection operations, and constructs a high-dimensional feature representation, namely the initial feature vector, based on the preliminary mapping. The initial feature vector is subjected to Query-Key-Value decomposition through three independent sets of 512×1024-dimensional linear transformation matrices to generate query features, key features, and value features. The query feature represents the model's focus on important information in the input features, the key feature is used to define the relationship between features, and the value feature saves the original information of the input features for subsequent weighted calculations. This decomposition mechanism is the core attention calculation method in Transformer, which effectively captures the correlation between features. The attention distribution matrix is ​​generated by performing a dot product operation on the query feature and the key feature, and applying a scaling factor to the result to normalize the denominator. The similarity of two feature vectors is calculated by dot product to capture the strength of the correlation between features, and the scaling operation is used to prevent gradient explosion or numerical instability in high-dimensional calculations. The resulting attention distribution matrix directly reflects the relative importance between the query feature and the key feature. In order to process the attention distribution matrix, the probability normalization method is applied to convert the values ​​in the matrix into attention weight coefficients. These weight coefficients are probability distributions after normalization, ensuring that the relative relationship between features can be quantified. Using these weight coefficients, the value features are weightedly combined to aggregate the most relevant information for each query feature. Since the target S-BERT model contains 8 attention heads, this operation is performed independently in each attention head to ensure that the model can capture the diversity of input features from multiple perspectives. The output features of each attention head are merged into a unified multi-head feature representation through a splicing operation. The multi-head features are reduced in dimension through a 4096×1024-dimensional linear layer. The dimensionality reduction operation compresses the multi-head features into a more compact space through linear projection while retaining the key information content. Layer Normalization is used for feature normalization. Layer Normalization effectively reduces the deviation problem between features by normalizing the features, improves the robustness of features and the convergence speed of model training, and finally generates an element vector group.

[0015] S3, input the visual feature triple combination and element vector group into the dual encoding framework for bidirectional attention fusion to obtain a multimodal feature group; It should be noted that the visual feature triples are input into the feature encoder in the dual encoding framework, and the element vector group is input into the element encoder in the dual encoding framework. The feature encoder and the element encoder are respectively composed of a stack of six layers of Transformer encoders. This design allows efficient feature extraction of the two inputs in independent encoding paths. The feature encoder extracts the global and local information of the visual feature triples, captures the hierarchical relationship within the features through a multi-head attention mechanism, and embeds and serializes the input multimodal data for subsequent processing. The element encoder performs similar processing on the element vector group, models the interaction relationship between elements through the self-attention mechanism of the Transformer, and generates an element encoding output with a high-dimensional embedding representation. The feature encoding output is linearly transformed and mapped to a new feature space to obtain a feature query matrix. Similarly, the element encoding output is linearly transformed to generate an element key value matrix. The linear transformation operation maps the original output features to the feature space of the same dimension through the weight matrix, ensuring that the two matrices are matched and fused in the subsequent interaction process. The feature query matrix is ​​used to represent the query relationship, and the element key value matrix stores both key and value information to support subsequent attention calculations. The feature query matrix and the element key value matrix are matrix multiplied to calculate the cross attention distribution. The cross attention distribution captures the similarity between the feature query and the element key value through the dot product operation and reflects the correlation structure between the two features at the matrix level. To optimize the result of the attention calculation, an activation function, such as softmax, is applied to the cross attention distribution to transform the original cross attention distribution into a probability distribution and generate the cross attention weights. Using these attention weights, the element key value matrix is ​​weighted summed to obtain the cross feature representation. The cross feature representation is input into the eight layers of self-attention layer. Each layer of self-attention layer mines the internal relationship between features through the multi-head attention mechanism and enhances the expressiveness of features through layer-by-layer modeling. After eight layers of self-attention processing, self-attention features are generated. Self-attention features integrate the high-order semantic information of visual features and element features in an adaptive way to improve the discriminative power of features. The self-attention features are dimensionally transformed and feature normalized. High-dimensional features are embedded into a low-dimensional space through linear projection, thereby reducing computational overhead and retaining key information. The feature normalization uses the Layer Normalization method to eliminate the deviation between feature dimensions and improve the stability of features and the convergence efficiency of the model. After processing, the feature mapping matrix is ​​obtained. The feature mapping matrix is ​​superimposed with the cross feature representation through residual connection. The residual connection directly adds the initial cross feature representation and the feature mapping matrix after multi-layer refinement to ensure that the original information is not lost in the deep network, while introducing the richness of nonlinear changes.In order to avoid overfitting and improve the generalization ability of the model, the dropout mechanism is introduced after the feature superposition to randomly discard the features. Through the dropout mechanism, some features are temporarily blocked in each training, prompting the model to learn more robust feature expressions. After the above steps, a multimodal feature group is finally generated.

[0016] S4, performing deep separable convolution processing and feature separation on the multimodal feature group to obtain a target feature group; Specifically, the multimodal feature group is input into the first Ghost module, which has the characteristics of lightweight design and efficient feature extraction capability. In the first Ghost module, a 3×3 depthwise separable convolution operation is used to extract the main branch features of the multimodal feature group to generate a main branch feature map. Depthwise separable convolution greatly reduces the amount of calculation by decomposing the traditional convolution into depthwise convolution and pointwise convolution, while retaining the integrity of feature expression. While the main branch feature is extracted, the auxiliary branch of the Ghost module performs specific sparse feature extraction on the multimodal feature group to generate a Ghost feature map. The generation of the Ghost feature map further compresses the feature representation through operations based on weight sharing and feature sparsity, so that the model can efficiently learn potential information. After completing the feature extraction of the main branch and the auxiliary branch, the main branch feature map is input into the TSM module for temporal feature shift operation. The TSM module generates a temporal shift feature map by shifting 1 / 4 of the channel features forward by one time step, 1 / 4 of the channel features backward by one time step, and the remaining channel features remain unchanged. Through the temporal shift operation, the TSM module can capture the dynamic relationship between features. The temporal shift feature map is concatenated with the Ghost feature map to complete the feature output of the first Ghost module. The feature concatenation operation integrates the features extracted by the main branch and the auxiliary branch, retains the association between global and local features, and enhances the diversity and expression of features. The output features of the first Ghost module are passed to the second to fifth Ghost modules in turn, and each Ghost module uses the same feature extraction structure. Through depthwise separable convolution and temporal shift operations, these modules further process the input features to generate the output features of five Ghost modules. As the network goes deeper layer by layer, the feature extraction capability of each module is gradually enhanced, and the input features are decomposed and represented more finely. After obtaining the output features of the five Ghost modules, these features are integrated at multiple scales by cross-layer feature fusion. Cross-layer feature fusion can effectively capture information of different scales by aligning and combining features at different levels, making the final feature expression richer and more comprehensive. The fused multi-scale features are input into the feature pyramid network (FPN) to complete the top-down transmission and lateral connection of features. FPN achieves multi-scale fusion of features by transferring high-level semantic information downward step by step and combining feature representations of different scales. The features output from FPN are expressed as lightweight feature representations. The lightweight feature representation is separated into target features and non-target features through feature decoupling self-knowledge distillation network. Self-knowledge distillation network is a technology that transfers knowledge through the model itself, which can improve the learning ability of the model while decoupling features. During the decoupling process, the network separates the lightweight feature representation into target features and non-target features.The target features represent the key visual elements and semantic information in the ad creative, while the non-target features represent the background or auxiliary information. By decoupling, the model can focus more on generating the core elements required in the ad creative and finally obtain the target feature group.

[0017] The lightweight feature representation is input into the teacher network for feature extraction. As a high-performance model, the teacher network can generate feature maps with rich semantic information and high resolution, namely the teacher feature map. The teacher feature map is input into the feature decoupling module for decomposition of target features and non-target features. The design of the feature decoupling module includes three 1×1 convolutional layers and a channel attention layer. The 1×1 convolutional layer is used to reduce the computational complexity and enhance the local expression ability of the features while retaining the important relationship between channels. The channel attention layer strengthens the expression ability of the target features and suppresses the interference of non-target features by assigning weights to each channel in the feature map. In the output of this module, the target features and non-target features are separated, representing the key visual elements and non-key background or auxiliary information in the advertising creativity, respectively. The target features are measured by the target feature loss function to quantify the difference between the target features and the expected features. The target feature loss function adopts a weighted combination of cross entropy loss and L2 regularization term. The cross entropy loss measures the performance of the target features in the classification task, while the L2 regularization term is used to constrain the parameters of the target features to avoid overfitting problems. At the same time, the non-target features are measured by the non-target feature loss function, which combines the mean square error loss and the L1 regularization term. The mean square error loss is used to evaluate the reconstruction error of the non-target features, while the L1 regularization term constrains the sparsity of the non-target features to ensure that they do not interfere with the target features during the generation process. In order to achieve efficient knowledge transfer, the lightweight feature representation is input into the student network for feature extraction. The student network is used as an optimization model, and its goal is to improve its own performance by imitating the behavior of the teacher network. By measuring the student feature graph with the knowledge distillation loss function, the feature transfer from the teacher network to the student network can be effectively achieved. The knowledge distillation loss function adopts the combination of Kullback-Leibler divergence and temperature coefficient. Kullback-Leibler divergence is used to measure the difference between the teacher feature distribution and the student feature distribution, while the temperature coefficient enhances the knowledge transfer effect by adjusting the smoothness of the distribution. The target feature loss value, non-target feature loss value and knowledge distillation loss value are weighted and combined to obtain the total loss value L. By comprehensively considering the accuracy of the target features, the separability of the non-target features, and the knowledge learning effect of the student network on the teacher network, the parameters of the student network are iteratively updated and optimized. During the optimization process, the gradient descent algorithm is used to minimize the total loss value and gradually adjust the parameters of the student network so that it can be closer to the characteristic performance of the teacher network while having higher efficiency. Through the collaborative training of the teacher network and the student network, combined with the efficient separation ability of the feature decoupling module, and the joint optimization of multiple loss functions, the target feature group is finally generated.

[0018] S5, based on the local linear embedding algorithm, performs neighborhood screening and reconstruction calculation on the target feature group to generate advertising creative visual elements.

[0019] Among them, the cosine similarity is calculated for each feature vector in the target feature group to quantify the similarity between feature vectors and construct a feature distance matrix. Cosine similarity can effectively reflect the relative relationship of feature points in high-dimensional space by calculating the cosine value of the angle between vectors. Based on the generated feature distance matrix, a K nearest neighbor search is performed on each feature point, and the K feature points with the closest distance are selected to form an optimal neighborhood set to ensure that the selected neighborhood can accurately characterize the local structure of the feature point. The feature points in the optimal neighborhood set are centralized in spatial coordinates, and the global offset effect in the feature distribution is eliminated by subtracting the central coordinates of each feature point to obtain a centralized feature matrix. The centralized feature matrix is ​​subjected to singular value decomposition to decompose the local geometric structure matrix. Singular value decomposition can extract the main characteristic modes of the centralized feature matrix and reveal the geometric distribution characteristics of the feature points in the local space. Based on the local geometric structure matrix, a reconstruction weight optimization objective function is constructed to determine the reconstruction weight of the feature point. The objective function approximates the original feature point by keeping the local structure of the feature point unchanged, that is, by linearly combining the neighborhood feature points. In order to optimize the objective function, the Lagrange multiplier method is introduced to solve the reconstruction weights and obtain the reconstruction weight matrix. The Lagrange multiplier method can achieve optimization under constraints, so that the reconstruction weight of each feature point not only meets the linear relationship of the neighborhood features, but also meets the constraint that the weight sum is 1, ensuring that the optimization result has good geometric consistency. The reconstruction weight matrix is ​​multiplied with the target feature group to generate a low-dimensional feature representation. The low-dimensional feature representation is completed by the local linear embedding method, which compresses the original high-dimensional features into a low-dimensional space while maintaining the local geometric relationship of the original features. The low-dimensional feature representation is feature classified to determine the visual element category label corresponding to each feature. The classification process combines support vector machines, decision trees or deep learning models to assign labels to low-dimensional features according to the annotated categories in the training data to ensure that the generated features have clear semantic meanings. According to the visual element category labels, the low-dimensional feature representation is feature reorganized, and features of different categories are reorganized into a new representation according to the category information. In order to convert the low-dimensional feature representation back to the original feature space, the generator module of the generative adversarial network is introduced. The generative adversarial network generator can map low-dimensional features back to high-dimensional feature representations that are consistent with the target feature group by learning the distribution of the original feature space. The training of the generator network is based on the distribution difference between the real features and the generated features, so that the generated features not only retain the local structure of the target feature group, but also maintain semantic consistency with the original data. Through the above feature classification, reorganization and generation process, the creative visual elements of the advertisement are obtained.

[0020] In one example, feature extraction and feature fusion are performed on the advertising design material to obtain a visual feature triple combination, including: The pixel size of the advertising design material is adjusted to obtain an RGB three-channel image, and the color distribution, texture features, and edge features of the RGB three-channel image are extracted to obtain a first feature matrix; Perform forward feature transfer and backward feature transfer on the first feature matrix to obtain a second feature matrix, and calculate the node association degree of the feature nodes in the second feature matrix to obtain an attention coefficient matrix; The second feature matrix is ​​weightedly fused based on the attention coefficient matrix to obtain a third feature matrix, and the correlation degree of the feature nodes in the third feature matrix is ​​calculated to obtain the node importance score; Feature nodes are screened and combined according to the node importance scores to obtain a candidate feature combination set, and the candidate feature combination set is classified and combined into subject features, relationship features and object features to obtain a visual feature triple combination.

[0021] In this example, the advertising design material is preprocessed and the input material image is adjusted to a uniform pixel size, for example Assume that the original image size is , pixel size adjustment uses bilinear interpolation algorithm, through the formula Completed, including Represents the adjusted pixel value, is the original pixel value, is the interpolation weight. The adjusted image is unified into RGB three-channel representation, and each pixel is represented by The three values ​​define the basis of the input features. Various feature extraction operations are performed on the adjusted RGB image to generate the first feature matrix The color distribution feature is calculated by calculating the histogram of the RGB channel Extract, where For the The frequency of the color value, is the total number of pixels, is the Kronecker function. Texture features are extracted using the gray level co-occurrence matrix (GLCM) to calculate the gray level co-occurrence relationship between each pair of pixels. Assume that the gray level is , the co-occurrence matrix The elements are defined as count ,in Represents the gray value of the pixel pair. For edge features, the Canny algorithm is used to calculate the gradient amplitude and direction ,in and The horizontal and vertical gradients calculated by the Sobel operator. After combining the above features, we can form , each element of which Represents pixels The multi-dimensional features of the first feature matrix Perform forward feature pass and backward feature pass to generate the second feature matrix The forward feature transfer is achieved through convolution operation, using the formula ,in represents the convolution kernel weight, is the bias. The backward feature transfer adjusts the weights through the back-propagation mechanism and bias , to enhance the significant information in the features. Calculate node association and construct attention coefficient matrix The attention coefficient is obtained by measuring the similarity of feature points, using the dot product attention mechanism, and the formula is: ,in and The query and key features are is the feature dimension, is the number of nodes. Used for Perform weighted fusion to generate the third feature matrix , the formula is .right The importance score of each node is calculated. Based on the weighted average attention weight calculation, the formula is ,in is the node connection weight. Filter the feature nodes according to the node importance score and select The nodes above the threshold form a candidate feature combination set. The candidate feature set is classified and combined, and the features are divided into subject features, relationship features and object features. Subject features represent the main visual elements of the image, such as the product or logo in the advertisement; relationship features describe the relative position or interaction between the subject and other elements, such as the contrast between the product and the background; object features include the features of the background and auxiliary elements. Through the above process, a visual feature triple combination is finally formed. ,in represents the subject feature matrix, is the relationship feature matrix, is the object feature matrix.

[0022] In one example, multi-head attention encoding and similarity calculation are performed on the combination of visual feature triplets to obtain an element vector group, including: Add 8 attention heads and a visual feature encoding layer with 512 hidden layers to the initial S-BERT model to obtain the target S-BERT model; The visual feature triples are combined and input into the target S-BERT model, and the feature space is mapped through a 512×512 dimensional linear projection matrix to obtain an initial feature vector of 1024 dimensions. The initial feature vector is decomposed into query-key-value by three independent 512×1024-dimensional linear transformation matrices to obtain query features, key features, and value features; Perform dot product operation on query features and key features, and normalize the denominator by the scaling factor to obtain the attention distribution matrix; The attention distribution matrix is ​​probability normalized to obtain the attention weight coefficient, the value features are weighted combined based on the attention weight coefficient, and the output features of the 8 attention heads are concatenated to obtain the multi-head features; The multi-head features are reduced in dimension through a 4096×1024-dimensional linear layer, and Layer Normalization is used to normalize the features to obtain an element vector group.

[0023] In this example, a visual feature encoding layer is introduced into the initial S-BERT model. This layer contains 8 attention heads and a 512-dimensional hidden layer. The purpose is to expand the model's feature processing capabilities to multimodal data, especially the field of visual features. The addition of attention heads allows the model to focus on different feature patterns at the same time, while the 512-dimensional design of the hidden layer balances computational complexity and expression ability to form the target S-BERT model. After the model expansion is completed, the visual feature triple combination is passed as input data to the target S-BERT model. This combination includes subject features, relationship features, and object features, each of which contains multidimensional feature descriptions to form a high-dimensional data set. The model maps the input features through a 512×512-dimensional linear projection matrix, as shown in the formula: ,in is the projected eigenvector, is the projection weight matrix, are input features, is the bias term. After this projection operation, the feature space is transformed from the original dimension into an initial feature vector of 1024 dimensions, which enhances the feature representation capability and provides a unified input format for the multi-head attention mechanism. dimensional linear transformation matrix, decomposing the initial feature vector into query features ( ), key features( ) and value characteristics This step defines how to model the relationship between features, where ,in , , are the weights of the linear transformation matrix respectively. The query feature is used to characterize the importance of the target feature, the key feature represents the relative relationship between the features, and the value feature carries the information of the original feature. After the feature decomposition is completed, the attention distribution matrix is ​​calculated by performing a dot product operation on the query feature and the key feature. The formula is: ,in represents the attention distribution matrix, is the dimension of the key feature. The dot product operation reflects the similarity between the query and the key, and the scaling factor It is used to prevent high-dimensional dot product values ​​from being too large and affecting gradient stability. The attention distribution matrix is ​​probability normalized using the Softmax function, and the formula is: ,in For the Line The attention weights of the columns, is the number of key features. This normalization step converts the attention distribution into a probability distribution so that the sum of the weights of each row is 1. The value features are weighted and combined using the attention weight coefficients. The formula is: ,in is the weighted feature representation. Each attention head completes this process independently, and the output features of 8 attention heads are obtained. These output features are combined into multi-head features through splicing operations, and the formula is ,in is the concatenated multi-head feature, which integrates the different pattern information captured by multiple attention heads. In order to optimize the feature representation and reduce the dimension, the multi-head feature is reduced in dimension through a 4096×1024-dimensional linear layer. The formula is ,in is the feature representation after dimensionality reduction, and are the weight and bias of the dimension reduction layer respectively. Layer Normalization is used for normalization, and the formula is ,in and are the mean and variance of the features, respectively. and are trainable scaling and offset parameters, is a smoothing term to prevent division by zero. After normalization, the element vector group is finally obtained as a high-quality expression of visual features.

[0024] In one example, the visual feature triple combination and the element vector group are input into the dual encoding framework for bidirectional attention fusion to obtain a multimodal feature group, including: The visual feature triple combination is input into the feature encoder in the dual encoding framework for six-layer Transformer encoding to obtain feature encoding output. At the same time, the element vector group is input into the element encoder in the dual encoding framework for six-layer Transformer encoding to obtain element encoding output. Perform a linear transformation on the feature encoding output to obtain a feature query matrix. At the same time, perform a linear transformation on the element encoding output to obtain an element key value matrix. Perform matrix multiplication on the feature query matrix and the element key value matrix to obtain the cross attention distribution, activate the cross attention distribution to obtain the cross attention weight, and perform weighted summation on the element key value matrix based on the cross attention weight to obtain the cross feature representation; The cross feature representation is refined through eight self-attention layers to obtain self-attention features, and the self-attention features are dimensionally transformed and normalized to obtain a feature mapping matrix. The feature mapping matrix is ​​superimposed with the cross-feature representation through residual connection, and the features are randomly discarded through the dropout mechanism to obtain a multimodal feature group.

[0025] In this example, the visual feature triples are input into the feature encoder in the dual encoding framework. The visual feature triples include subject features (such as the core elements of the advertisement), relation features (such as the relative layout between visual elements), and object features (such as background information). These features are converted into high-dimensional feature vectors through embedding operations and represented as matrices ,in is the number of eigenvectors, is the dimension of each vector. The feature encoder is composed of six layers of Transformer encoders, each of which includes a multi-head attention mechanism and a feedforward neural network module. In each layer of Transformer encoder, multi-head attention captures the global relationship of input features through query, key, and value mechanisms. The formula is: ; in, , , is a learnable weight matrix, is the dimension of each attention head. Through six layers of stacking, the feature encoder generates feature encoding output At the same time, the element vector group Input the element encoder in the dual encoding framework. The element vector group represents the characteristics of independent elements in the advertisement, such as fonts, icons or color information, and is also processed by the six-layer Transformer encoder to generate the element encoding output The feature encoder and element encoder share similar architectures, but are optimized for visual features and element features respectively. Perform linear transformation to generate feature query matrix ,in is the transformation weight of the query feature. Similarly, the element encoding output Perform linear transformation to generate key-value matrix and in are the transformation weights of key and value features respectively. Key features with elements Perform matrix multiplication to get the cross attention distribution : ; in Represents the attention distribution between features and elements. According to the cross attention weight Element-valued matrix Perform weighted summation to generate cross feature representation : ; in It is a representation that combines visual features and element features. Input eight layers of self-attention module. The self-attention module is based on the internal relationship of the input features, and captures local and global information through a similar attention mechanism, thereby strengthening the feature representation. After eight layers of refinement, the self-attention feature is generated. . For self-attention features Perform dimension transformation and feature normalization. Dimension transformation is achieved through linear projection, the formula is: ; in is the transformed feature representation, and are weights and biases. Normalization uses Layer Normalization, and the formula is: ; in and are the mean and variance of the feature, and is a learnable parameter. In order to integrate the self-attention feature and the cross-feature representation, the feature mapping matrix and Superposition through residual connections: ; In order to enhance the generalization ability of the model, the Dropout mechanism is used to randomly discard features and set some features to zero to prevent overfitting. The final output is a multimodal feature set. , which contains the deep fusion representation of visual feature triplets and element vector groups.

[0026] In one example, a multimodal feature group is subjected to a deep separable convolution process and feature separation to obtain a target feature group, including: The multimodal feature group is input into the first Ghost module, and the main branch feature is extracted from the multimodal feature group through 3×3 depth-separable convolution to obtain the main branch feature map, and the auxiliary branch feature is extracted from the multimodal feature group to obtain the Ghost feature map; The main branch feature map is input into the TSM module for temporal feature shift. The TSM module shifts 1 / 4 of the channel features forward by one time step, and 1 / 4 of the channel features backward by one time step, while the remaining channel features remain unchanged, to obtain a temporal shift feature map. The time series displacement feature map and the Ghost feature map are concatenated to obtain the output feature of the first Ghost module; The output features of the first Ghost module are sequentially input into the second to fifth Ghost modules. Each Ghost module adopts the same feature extraction structure and performs feature processing through depthwise separable convolution and temporal shift operations, finally obtaining the output features of five Ghost modules. The output features of the five Ghost modules are fused across layers to obtain multi-scale features, which are then input into the feature pyramid network. The features are fused through top-down feature transfer and lateral feature connection to obtain lightweight feature representation. The target features and non-target features are separated from the lightweight feature representation through feature decoupling self-knowledge distillation network to obtain the target feature group.

[0027] In this example, the multimodal feature group Enter the first Ghost module, which consists of a main branch and an auxiliary branch. In the main branch, through a The depth-separable convolution operation extracts the main features of the input features, and its calculation formula is: ; DepthConv is a deep convolution operation, which is performed on each channel separately. convolution, is the weight of the depth convolution, PointConv is the point convolution operation, through Convolution integrates channel information. is the weight of the point convolution. Depthwise separable convolution decouples the operations of spatial dimension and channel dimension, greatly reducing the amount of computation while extracting efficient main branch feature maps. In the auxiliary branch, the Ghost feature map is generated by sparse feature extraction , which is designed to approximate the main feature extraction process through a simple linear transformation. The calculation formula is: ; in is the sparse transformation weight matrix of the auxiliary branch, which can efficiently generate redundant features and supplement the main branch features. The two branches of the first Ghost module output and . The main branch feature map The input time series feature shift module (TSM) is used for time series processing. TSM simulates time dynamics by partially shifting the feature channel, which is defined as: The channel is shifted forward by one time step, and the other channels are shifted back one time step, and the rest The channels remain unchanged. The mathematical expression of the bit shift operation is: ; in represents the time step, is the channel index, is the spatial position. Time series displacement feature map It captures the temporal dynamic information in multimodal features. Ghost feature map Perform the splicing operation to obtain the output features of the first Ghost module: ; The concatenation operation combines two features along the channel dimension. The output features are passed to the second to fifth Ghost modules in sequence. Each Ghost module repeats the feature extraction structure of the main branch and auxiliary branch, and performs time series feature processing through TSM. The output features obtained from the five Ghost modules are respectively expressed as In order to fuse multi-level feature information, the output features of the five Ghost modules are cross-layer fused. Cross-layer fusion is achieved through the feature pyramid network, which transmits high-level semantic information layer by layer from top to bottom, and integrates features of different scales through horizontal connections. The fusion calculation formula of the pyramid network is: ; in Represents the pyramid network Layer features, UpSample is the upsampling operation, Output features for the corresponding Ghost module. After feature pyramid fusion, the generated multi-scale features Contains rich spatial and semantic information. Multi-scale features The input features are decoupled from the knowledge distillation network to separate the target features from the non-target features. High-quality target features are generated through the teacher network. and non-target features , and use the knowledge distillation loss function to optimize the student network, the formula is: ; in and The losses of target features and non-target features are respectively, using cross entropy loss and mean square error loss. is the weight factor. After the optimization is completed, the output target feature group That is the core visual element required for advertising creativity.

[0028] In one example, the target features and non-target features are separated from the lightweight feature representation by feature decoupling from the knowledge distillation network to obtain the target feature group, including: The lightweight feature representation is input into the teacher network for feature extraction to obtain the teacher feature map, and the teacher feature map is input into the feature decoupling module for target feature and non-target feature decomposition. The feature decoupling module contains three 1×1 convolutional layers and one channel attention layer to obtain target features and non-target features; The target feature is measured by the target feature loss function to obtain the target feature loss value. The target feature loss function adopts a weighted combination of the cross entropy loss and the L2 regularization term. The non-target feature is measured by the non-target feature loss function to obtain the non-target feature loss value. The non-target feature loss function adopts a weighted combination of mean square error loss and L1 regularization term; The lightweight feature representation is input into the student network for feature extraction to obtain the student feature map, and the feature transfer measurement of the student feature map is performed through the knowledge distillation loss function to obtain the knowledge distillation loss value. The knowledge distillation loss function adopts the combination of Kullback-Leibler divergence and temperature coefficient. The target feature loss value, non-target feature loss value and knowledge distillation loss value are weightedly combined to obtain the total loss value, and the student network parameters are optimized according to the total loss value L to obtain the target feature group.

[0029] In this example, the lightweight feature is represented by Input into the teacher network. As a high-performance model, the teacher network has stronger feature extraction capabilities and can Perform deep feature processing to generate teacher feature maps . The teacher characteristic map Input feature decoupling module, the design of this module includes three convolutional layer and a channel attention layer. Each The convolution layer reduces the computational complexity by performing channel-by-channel convolution operations while extracting the local relationship of features. Its mathematical expression is: ; in It is The convolution weights of the layer, For the Layer input features, is the bias term, is the convolution output. The feature map generated by the three-layer convolution operation It is then passed to the channel attention layer, which enhances the importance of the target feature by calculating the weight distribution of the feature channel. The formula for the channel attention mechanism is: ; in is the channel weight, is the weight matrix of the two-layer fully connected network, Represents the sigmoid function, which is used to generate normalized weights. , target features and non-target features Separated out: ; in Represents element-by-element multiplication. The target feature loss function is used to multiply the target feature. The target feature loss function combines the cross entropy loss and the L2 regularization term, and its formula is: ; in is the true label, is the predicted probability, is the sample size, is the weight associated with the target feature, is the regularization coefficient. The cross entropy term measures the classification accuracy, and the L2 regularization term avoids overfitting. At the same time, the non-target features are measured by the mean square error loss and the L1 regularization term. The non-target feature loss function formula is: ; in is the reference non-target feature, is the L1 regularization coefficient, and the L1 term is used to sparse non-target features. At the same time, the lightweight feature is represented as Input the student network, which is used as a lightweight model for performance optimization to generate a student feature map Through the knowledge distillation loss function and To measure feature transfer, the knowledge distillation loss formula is: ; in is the temperature coefficient, KL is the Kullback-Leibler divergence, is the Softmax function. The temperature coefficient enhances the stability of feature matching by smoothing the probability distribution. , non-target feature loss value and knowledge distillation loss Weighted combination to get the total loss value : ; in is the weighting coefficient. Using the total loss value Perform back propagation optimization on the student network, gradually adjust the parameters, and finally generate the target feature set .

[0030] In one example, neighborhood screening and reconstruction calculation are performed on the target feature group based on a local linear embedding algorithm to generate advertising creative visual elements, including: The cosine similarity is calculated for the feature vectors in the target feature group to obtain the feature distance matrix, and the K nearest neighbor search is performed on each feature point according to the feature distance matrix to obtain the optimal neighborhood set; The feature points in the optimal neighborhood set are processed by centering the spatial coordinates to obtain a centralized feature matrix, and the centralized feature matrix is ​​subjected to singular value decomposition to obtain a local geometric structure matrix; A reconstruction weight optimization objective function is constructed based on the local geometric structure matrix, and the weight coefficient of the reconstruction weight optimization objective function is solved by the Lagrange multiplier method to obtain a reconstruction weight matrix; Perform matrix multiplication operation on the reconstructed weight matrix and the target feature group to obtain a low-dimensional feature representation, and perform feature classification on the low-dimensional feature representation to obtain a visual element category label; The low-dimensional feature representation is reorganized according to the visual element category label, and the low-dimensional feature representation is reconstructed into the original feature space through the generator of the generative adversarial network to obtain the creative visual elements of the advertisement.

[0031] In this example, the cosine similarity is calculated for the feature vectors in the target feature group to generate a feature distance matrix. ,in is the number of feature points in the target feature group. The cosine similarity calculation formula is: ; By calculating the similarity between all feature vectors, we get the feature distance matrix ,in Representing feature points and The relative distance of , for each feature point conduct Nearest neighbor search, by selecting The smallest distance The optimal neighborhood set is formed by After obtaining the optimal neighborhood set, the feature points in each neighborhood set are centralized in spatial coordinates to remove the global offset in the feature point distribution. The centralization formula is: ; in is the centralised feature point, is the characteristic mean of the neighborhood set. The centralized feature points are formed into a matrix , and perform singular value decomposition on it, the decomposition formula is: ; in and is an orthogonal matrix, is a diagonal matrix containing the singular values ​​of the centered eigenvalue matrix. Matrix represents the local geometric structure matrix of feature points , which captures the main direction of the feature point distribution. Based on the local geometric structure matrix , construct the reconstruction weight optimization objective function to keep the local linear relationship of the feature points unchanged. The objective function formula is: ; in is the reconstruction weight matrix, It is a feature point and its neighboring points The weight of . To optimize the objective function , the Lagrange multiplier method is introduced to integrate the constraints into part of the objective function, and the optimization formula is: ; right Find the partial derivative and set it to 0 to get the optimal solution , as the reconstruction weight matrix. The reconstruction weight matrix With target feature group Perform matrix multiplication to generate low-dimensional feature representation :

[0032] in , represents the feature dimension after dimensionality reduction. Classify and generate visual element category labels , where each label Representing feature points The classification operation is completed based on support vector machines or deep learning classifiers. According to the visual element category label, the low-dimensional feature representation is reorganized and the features of the same category are merged into a new feature representation. The low-dimensional feature representation is reorganized through the generator of the Generative Adversarial Network (GAN). Reconstruct back to the original feature space to generate creative visual elements of the advertisement. The GAN generator learns the mapping from low-dimensional space to high-dimensional feature space, and its reconstruction formula is: ; in is the generator network, It is a reconstructed high-dimensional feature representation. Through adversarial training, the generator can generate advertising creative visual elements that are similar to the original features but have category information.

[0033] Reference Figure 2 This embodiment provides an intelligent generation device for creative advertising visual elements, including: Feature extraction module 1, used to extract and fuse features of the advertising design material to obtain a triple combination of visual features; Computation module 2, used for performing multi-head attention encoding and similarity calculation on the combination of visual feature triples to obtain an element vector group; Attention fusion module 3, used for inputting the visual feature triple combination and the element vector group into the dual encoding framework for bidirectional attention fusion to obtain a multimodal feature group; The feature separation module 4 is used to perform deep separable convolution processing and feature separation on the multimodal feature group to obtain a target feature group; The generation module 5 is used to perform neighborhood screening and reconstruction calculation on the target feature group based on the local linear embedding algorithm to generate advertising creative visual elements.

[0034] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0035] Reference Figure 3In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0036] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0037] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0038] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0039] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0040] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for intelligently generating creative visual elements of advertisements, characterized in that: The following steps are involved: Perform feature extraction and feature fusion on the advertising design materials to obtain a triple combination of visual features; Performing multi-head attention encoding and similarity calculation on the visual feature triple combination to obtain an element vector group; Inputting the visual feature triple combination and the element vector group into a dual encoding framework for bidirectional attention fusion to obtain a multimodal feature group; Performing depth-separable convolution processing and feature separation on the multimodal feature group to obtain a target feature group; Based on the local linear embedding algorithm, the target feature group is subjected to neighborhood screening and reconstruction calculation to generate advertising creative visual elements.

2. The method for intelligently generating creative visual elements of advertisements according to claim 1, characterized in that: The feature extraction and feature fusion of the advertising design material to obtain a visual feature triple combination includes: Adjusting the pixel size of the advertisement design material to obtain an RGB three-channel image, and extracting color distribution, texture features, and edge features from the RGB three-channel image to obtain a first feature matrix; Performing forward feature transfer and backward feature transfer on the first feature matrix to obtain a second feature matrix, and performing node association calculation on feature nodes in the second feature matrix to obtain an attention coefficient matrix; Based on the attention coefficient matrix, the second feature matrix is ​​weightedly fused to obtain a third feature matrix, and the correlation degree of the feature nodes in the third feature matrix is ​​calculated to obtain the node importance score; The feature nodes are screened and combined according to the node importance scores to obtain a candidate feature combination set, and the candidate feature combination set is classified and combined into subject features, relationship features and object features to obtain a visual feature triple combination.

3. The method for intelligently generating creative visual elements of advertisements according to claim 2, characterized in that: The multi-head attention encoding and similarity calculation are performed on the visual feature triple combination to obtain an element vector group, including: Add 8 attention heads and a visual feature encoding layer with 512 hidden layers to the initial S-BERT model to obtain the target S-BERT model; The visual feature triple combination is input into the target S-BERT model, and feature space mapping is performed through a 512×512 dimensional linear projection matrix to obtain a 1024-dimensional initial feature vector; Perform Query-Key-Value decomposition on the initial feature vector through three independent 512×1024 dimensional linear transformation matrices to obtain query features, key features and value features; Performing a dot product operation on the query feature and the key feature, and normalizing the denominator by a scaling factor to obtain an attention distribution matrix; Probability normalization is performed on the attention distribution matrix to obtain an attention weight coefficient, weighted combination is performed on the value features based on the attention weight coefficient, and output features of 8 attention heads are concatenated to obtain multi-head features; The multi-head features are reduced in dimension through a 4096×1024-dimensional linear layer, and the features are normalized using Layer Normalization to obtain an element vector group.

4. The method for intelligently generating creative visual elements of advertisements according to claim 3, characterized in that: The step of inputting the visual feature triple combination and the element vector group into a dual encoding framework for bidirectional attention fusion to obtain a multimodal feature group includes: Inputting the visual feature triple combination into a feature encoder in a dual encoding framework for six-layer Transformer encoding to obtain a feature encoding output, and at the same time, inputting the element vector group into an element encoder in the dual encoding framework for six-layer Transformer encoding to obtain an element encoding output; Performing a linear transformation on the feature encoding output to obtain a feature query matrix, and at the same time, performing a linear transformation on the element encoding output to obtain an element key value matrix; Performing a matrix multiplication operation on the feature query matrix and the element key value matrix to obtain a cross attention distribution, activating the cross attention distribution to obtain a cross attention weight, and performing a weighted summation on the element key value matrix based on the cross attention weight to obtain a cross feature representation; The cross feature representation is refined through eight self-attention layers to obtain self-attention features, and the self-attention features are dimensionally transformed and feature normalized to obtain a feature mapping matrix; The feature mapping matrix is ​​superimposed with the cross feature representation through residual connection, and features are randomly discarded through a dropout mechanism to obtain a multimodal feature group.

5. The method for intelligently generating creative visual elements of advertisements according to claim 4, characterized in that: The performing of deep separable convolution processing and feature separation on the multimodal feature group to obtain a target feature group includes: Inputting the multimodal feature group into the first Ghost module, performing main branch feature extraction on the multimodal feature group through 3×3 depthwise separable convolution to obtain a main branch feature map, and performing auxiliary branch feature extraction on the multimodal feature group to obtain a Ghost feature map; The main branch feature map is input into the TSM module for time series feature shifting. The TSM module shifts 1 / 4 of the channel features forward by one time step, and shifts 1 / 4 of the channel features backward by one time step, while the remaining channel features remain unchanged, to obtain a time series shift feature map; Performing feature splicing on the time series displacement feature map and the Ghost feature map to obtain a first Ghost module output feature; The output features of the first Ghost module are sequentially input into the second Ghost module to the fifth Ghost module. Each Ghost module adopts the same feature extraction structure, performs feature processing through depth-separable convolution and temporal shift operations, and finally obtains the output features of five Ghost modules; The output features of the five Ghost modules are cross-layer fused to obtain multi-scale features, and the multi-scale features are input into the feature pyramid network. The features are fused through top-down feature transfer and lateral feature connection to obtain lightweight feature representation; The target features and non-target features of the lightweight feature representation are separated through feature decoupling from the knowledge distillation network to obtain a target feature group.

6. The method for intelligently generating creative visual elements of advertisements according to claim 5, characterized in that: The feature decoupling self-knowledge distillation network is used to separate the target features and non-target features of the lightweight feature representation to obtain a target feature group, including: Input the lightweight feature representation into the teacher network for feature extraction to obtain a teacher feature map, and input the teacher feature map into a feature decoupling module for target feature and non-target feature decomposition, wherein the feature decoupling module includes three 1×1 convolutional layers and one channel attention layer to obtain target features and non-target features; The target feature is measured by a target feature loss function to obtain a target feature loss value, wherein the target feature loss function adopts a weighted combination of a cross entropy loss and an L2 regularization term; Measuring the non-target feature by a non-target feature loss function to obtain a non-target feature loss value, wherein the non-target feature loss function adopts a weighted combination of a mean square error loss and an L1 regularization term; Input the lightweight feature representation into the student network for feature extraction to obtain a student feature graph, and perform feature transfer measurement on the student feature graph through a knowledge distillation loss function to obtain a knowledge distillation loss value, wherein the knowledge distillation loss function adopts a combination of Kullback-Leibler divergence and a temperature coefficient; The target feature loss value, the non-target feature loss value and the knowledge distillation loss value are weightedly combined to obtain a total loss value, and the student network is parameter optimized according to the total loss value L to obtain a target feature group.

7. The method for intelligently generating creative visual elements of advertisements according to claim 6, characterized in that: The method of performing neighborhood screening and reconstruction calculation on the target feature group based on the local linear embedding algorithm to generate the creative visual elements of the advertisement includes: Calculating the cosine similarity of the feature vectors in the target feature group to obtain a feature distance matrix, and performing a K-nearest neighbor search on each feature point according to the feature distance matrix to obtain an optimal neighborhood set; Performing spatial coordinate centering processing on the feature points in the optimal neighborhood set to obtain a centralized feature matrix, and performing singular value decomposition on the centralized feature matrix to obtain a local geometric structure matrix; Constructing a reconstruction weight optimization objective function based on the local geometric structure matrix, and solving the weight coefficients of the reconstruction weight optimization objective function by Lagrange multiplier method to obtain a reconstruction weight matrix; Performing a matrix multiplication operation on the reconstructed weight matrix and the target feature group to obtain a low-dimensional feature representation, and performing feature classification on the low-dimensional feature representation to obtain a visual element category label; The low-dimensional feature representation is reorganized according to the visual element category label, and the low-dimensional feature representation is reconstructed into the original feature space through a generator of a generative adversarial network to obtain an advertising creative visual element.

8. An intelligent device for generating creative visual elements for advertisements, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the device comprises: A feature extraction module is used to extract and fuse features of the advertising design material to obtain a triple combination of visual features; A calculation module, used for performing multi-head attention encoding and similarity calculation on the visual feature triple combination to obtain an element vector group; An attention fusion module, used for inputting the visual feature triple combination and the element vector group into a dual encoding framework for bidirectional attention fusion to obtain a multimodal feature group; A feature separation module, used for performing deep separable convolution processing and feature separation on the multimodal feature group to obtain a target feature group; The generation module is used to perform neighborhood screening and reconstruction calculation on the target feature group based on a local linear embedding algorithm to generate advertising creative visual elements.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images