Advertisement image optimization system based on neural network

Abstract

The invention provides an advertisement image optimization system based on a neural network. The advertisement image optimization system comprises an image acquisition module, a feature extraction module and an optimization processing module, the image acquisition module is used for acquiring original data of an advertisement image from an advertisement image data source through a web crawler or local reading to obtain basic data of the advertisement image, and preprocessing the acquired image, and the preprocessing comprises normalization, cutting processing, obtaining of preprocessed advertisement image data, and obtaining of the advertisement image data; a transmitter of a data transmission layer of the image acquisition module is used for sending various advertisement image data to the optimization processing module; the feature extraction module is used for loading the image acquisition module by using a neural network technology when the system is started; and the optimization processing module is used for analyzing and optimizing the received advertisement image data and displaying an optimization result. According to the invention, the optimization of the advertisement image can be realized, the visual effect and potential market attraction of the advertisement are improved, and the click rate and conversion rate of the advertisement are increased.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and neural network technology, and more specifically, to an advertising image optimization system based on a neural network. Background Art

[0002] Existing advertising image optimization technology mainly relies on manual design and adjustment, which is not only time-consuming and labor-intensive, but also difficult to adapt to the rapidly changing market demands and diverse advertising delivery platforms. In the advertising field, the quality of the image directly affects the attractiveness and effectiveness of the advertisement. Therefore, how to quickly and accurately optimize the advertising image to adapt it to different delivery environments and target audiences has become an urgent problem to be solved.

[0003] Traditional image processing techniques, such as color adjustment, cropping, and normalization, can improve image quality to a certain extent, but these methods often require professional image processing knowledge and experience, and are difficult to automate on a large scale. In addition, with the development of big data and deep learning technology, people have begun to try to use machine learning models to automatically analyze and optimize images, but these models often require a large amount of labeled data and complex training processes, and may have insufficient generalization capabilities in practical applications.

[0004] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: the existing advertising image optimization system lacks deep learning and understanding of advertising image features, resulting in limited optimization effects and inability to effectively adapt to the needs of different advertising scenarios and audiences. Summary of the invention

[0005] The present invention provides an advertising image optimization system based on a neural network, comprising: Image acquisition module, feature extraction module and optimization processing module; The image acquisition module is used to obtain the original data of the advertising image from the advertising image data source through a web crawler or local reading to obtain the basic data of the advertising image, pre-process the collected image, the pre-processing includes normalization and cropping to obtain the pre-processed advertising image data, and send various types of advertising image data to the optimization processing module using the transmitter of the data transmission layer of the image acquisition module; The feature extraction module is used to load the image acquisition module using the neural network technology when the system is started; The optimization processing module is used to analyze and optimize the received advertising image data and display the optimization results. Furthermore, the image acquisition module also includes a basic library layer for providing general image classes and interfaces, the general image classes including image pixel data classes, image color mode classes, image size classes and image acquisition abstract classes, as well as defining an image context to manage the data context of each image at different processing stages, storing the image data objects of each image in the preprocessing layer, and defining an image monitor to perform periodic monitoring of the image data objects in the image context, and immediately sending the image data that meets the optimization conditions to the optimization processing module. Furthermore, the original data of the advertisement image is obtained from the advertisement image data source by a web crawler or local reading to obtain basic data of the advertisement image, including: In the advertising image data source, create image acquisition interfaces for the online advertising platform, local advertising library, and social media advertising library, all of which inherit the common image acquisition abstract class of the basic library layer. In each image acquisition interface, collect the original data of the advertising image according to the configuration of each data source; The universal acquisition manager of the basic library layer uniformly scans and loads the image acquisition interfaces of various data sources. Each image acquisition interface runs independently, encapsulates the collected original advertising image data into a unified image data object, and puts it into the image data set of the acquisition manager to obtain the basic data of the advertising image.

[0006] Furthermore, the normalization and cropping of the collected images includes: Normalize the image so that the image pixel value range is between [0,1]. According to the normalization formula

[0007] in, is the original pixel value, and are the minimum and maximum pixel values, respectively. is the normalized pixel value; Crop the image according to the preset cropping size ; in, is the cutting height, For the crop width, crop the image from the center of the image or the specified position to obtain the cropped image data; The data after normalization and cropping are integrated to obtain pre-processed advertising image data. Furthermore, the normalization of the image so that the image pixel value range is between [0, 1] also includes: When there are outliers in the image pixel values, the outliers are first corrected to the nearest normal boundary value and then normalized; When an image has multiple RGB channels, each channel is normalized separately. Furthermore, the preprocessing also includes: Adding tag information to the cropped image for identifying the image source and processing history in subsequent processing, and storing the tag information in metadata of the image data object in a specific format; Furthermore, the resolution change rate is calculated according to the original resolution of the image and the resolution after cropping, which is used to evaluate the degree of loss of image information. Furthermore, the image acquisition module is loaded using the neural network technology, including: When the system starts, based on the neural network framework, the customized advertising image loader is used to obtain the codes of each layer in the image acquisition module from the system's image storage directory; Among them, the construction of the advertising image loader is as follows: Suppose the pixel matrix of the advertising image is , whose dimensions are ;in, is the image height, is the image width, is the number of image channels, each pixel in the advertising image The pixel value is ,satisfy , , and each pixel value is a dimensional vector; For an advertisement image, define its advertisement feature vector ; in, It is the dimension of advertising features, which includes the color features of advertising elements. , Ad text features , Advertised product features ; The loader first obtains the basic parameters of the image through the format parsing formula according to the format information of the advertisement image; The format parsing formula is: ; in, For the parsed format parameters, obtain the basic parameters of the image, including color mode and resolution; Then construct an advertising feature index based on the metadata of the advertising image

[0008] in, Metadata includes ad category labels and delivery platform information; Matching the advertisement image with a specific advertisement feature library through an advertisement feature matching formula to determine the initial attributes of the image in the advertisement scene; The advertisement feature matching formula is:

[0009] in, For the matching results; Through the convolutional layer and pooling layer of the neural network, the image data corresponding to each layer of code is extracted to create new feature vectors corresponding to each layer. After adding the extended image features in each layer of code to the new feature vectors, a new feature representation is generated. In the convolution layer, the convolution kernel is used With image Perform convolution operation to extract local features of the image; In the pooling layer, the local features are reduced in dimension using the maximum pooling formula. Instantiate the transmitter of the data transmission layer, the acquisition manager and image monitor of the basic library layer; The instantiated acquisition manager periodically executes each image acquisition interface of the advertisement image data source. Furthermore, using the convolution kernel With image Perform convolution operation, the formula is:

[0010] in, is the convolution result, is the image coordinate, is the convolution kernel coordinate, extracting local features of the image; The maximum pooling formula is:

[0011] in, is the pooling result, is the pooling area.

[0012] Furthermore, the feature extraction module also includes training optimization of the neural network model, including: Construct training and testing datasets based on the collected large amount of advertising image data and the corresponding optimization target labels; The back propagation algorithm is used to update the parameters of the neural network model, and the loss function is:

[0013] in is the true label, is the predicted probability. Furthermore, the optimization processing module is used to analyze and optimize the received advertising image data and display the optimization results, including: performing feature analysis on the received advertising image data using a trained neural network model, and determining optimization targets according to the target audience and delivery platform of the advertising image; According to the optimization goal, the image is color-adjusted through the color balance formula, and the color balance formula is:

[0014] in, is the adjusted color value, is the original color value, Adjust the amount for color; The optimized advertising image is displayed to the user, and the image evaluation indicators before and after optimization are recorded for evaluating the optimization effect. The image evaluation indicators include click-through rate.

[0015] The above-mentioned embodiments of the present invention have at least the following beneficial effects: by adopting the advertising image optimization system based on neural network, the advertising image data can be automatically collected and preprocessed, including normalization and cropping operations, thereby improving the efficiency and accuracy of image processing. The system can use neural network technology to extract and analyze the features of advertising images, intelligently determine the optimization target according to the target audience and delivery platform of the advertising images, and then perform optimization operations such as color adjustment. This method can not only reduce manual intervention and reduce costs, but also improve the attractiveness and adaptability of advertising images, making them more in line with market demand.

[0016] In addition, through the training and optimization of the neural network model, the system can automatically update the model parameters based on a large amount of advertising image data and the corresponding optimization target labels to improve the accuracy of the optimization effect. The system can also record image evaluation indicators before and after optimization, such as click-through rate, to evaluate the optimization effect and further guide the adjustment of the optimization strategy. This data-driven optimization method can continuously improve the performance of advertising images, enhance the actual effect of advertising, and bring advertisers a higher return on investment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which: Figure 1 A schematic diagram of the structure of a neural network-based advertising image optimization system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0019] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0020] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0021] Reference below Figure 1 , Figure 1 FIG. 1 is a schematic diagram of the structure of an advertising image optimization system based on a neural network provided by an embodiment of the present invention. Figure 1 As shown, a neural network-based advertising image optimization system 100 includes: Image acquisition module 101, feature extraction module 102 and optimization processing module 103; The image acquisition module is used to obtain the original data of the advertising image from the advertising image data source through a web crawler or local reading to obtain the basic data of the advertising image, pre-process the collected image, the pre-processing includes normalization and cropping to obtain the pre-processed advertising image data, and send various types of advertising image data to the optimization processing module using the transmitter of the data transmission layer of the image acquisition module; The feature extraction module is used to load the image acquisition module using the neural network technology when the system is started; The optimization processing module is used to analyze and optimize the received advertising image data and display the optimization results. It should be noted that the system includes an image acquisition module, a feature extraction module, and an optimization processing module. These three modules work together to achieve automatic optimization of advertising images. Here, the image acquisition module refers to the component responsible for obtaining the original data of the advertising image. It can obtain images from different data sources through web crawlers or local reading. The feature extraction module uses neural network technology to identify and extract key features in the image. The optimization processing module analyzes these features and optimizes the image accordingly.

[0022] Specifically, when the system starts, the image acquisition module will load the common image classes and interfaces provided by the basic library layer, including image pixel data class, image color mode class, image size class and image acquisition abstract class. The basic library layer is also responsible for defining image context management, that is, the storage and management of image data in different processing stages.

[0023] More specifically, the image acquisition module also includes a transmitter of the data transmission layer, which is responsible for sending the pre-processed image data to the optimization processing module. When the system is started, the feature extraction module will load the neural network technology to facilitate subsequent feature recognition and extraction.

[0024] Preferably, the preprocessing step of the image acquisition module may include normalizing the image so that the image pixel value range is between [0,1], which can be achieved by the formula: normalized pixel value = (original pixel value - minimum pixel value) / (maximum pixel value - minimum pixel value).

[0025] Furthermore, the image can be cropped, and the image can be cropped from the center of the image or a specified position according to a preset cropping size to obtain cropped image data. These preprocessing steps help to improve the accuracy of subsequent feature extraction and the effect of optimization processing.

[0026] In some embodiments, the image acquisition module also includes a basic library layer for providing general image classes and interfaces, the general image classes including image pixel data classes, image color mode classes, image size classes and image acquisition abstract classes, as well as defining an image context to manage the data context of each image at different processing stages, storing the image data objects of each image in the preprocessing layer, and defining an image monitor to perform periodic monitoring of the image data objects in the image context, and immediately send the image data that meets the optimization conditions to the optimization processing module. It should be noted that the image acquisition module further includes a basic library layer, which provides a series of general image classes and interfaces to support various operations in the image acquisition process. Here, the basic library layer refers to a component within the system that contains the basic tools and functions required for image processing. The general image classes include image pixel data class, image color mode class, image size class and image acquisition abstract class, which are abstract representations of basic data structures and operations that may be used when processing images. At the same time, image context management refers to the way to manage and maintain image data at different stages of image processing to ensure data consistency and integrity.

[0027] Specifically, the image pixel data class in the basic library layer is responsible for storing and processing image pixel information, the image color mode class manages the image color space, the image size class handles the size change of the image, and the image acquisition abstract class defines the basic operations and interfaces of image acquisition. Image context management ensures that image data objects can be correctly stored and transferred in different stages such as image preprocessing, feature extraction and optimization processing.

[0028] More specifically, the image monitor acts as a timed monitoring tool that checks the image data objects in the image context. Once image data that meets the optimization conditions is found, it will be immediately sent to the optimization processing module.

[0029] Preferably, the image acquisition abstract class in the basic library layer can be implemented by creating specific image acquisition interfaces. These interfaces can be image acquisition interfaces of online advertising platforms, local advertising libraries, and social media advertising libraries, all of which inherit from the general image acquisition abstract class of the basic library layer.

[0030] Furthermore, these interfaces collect the original data of advertising images according to the configuration of their respective data sources. The acquisition manager is responsible for uniformly scanning and loading the image acquisition interfaces of these data sources, and runs independently to encapsulate the collected original data of advertising images into a unified image data object and store it in the image data set of the acquisition manager. This design enables the system to flexibly adapt to different data sources while maintaining the consistency and efficiency of data processing.

[0031] In some embodiments, the step of obtaining the original data of the advertisement image by crawling a web or reading locally from the advertisement image data source to obtain the basic data of the advertisement image includes: In the advertising image data source, create image acquisition interfaces for the online advertising platform, local advertising library, and social media advertising library, all of which inherit the common image acquisition abstract class of the basic library layer. In each image acquisition interface, collect the original data of the advertising image according to the configuration of each data source; The universal acquisition manager of the basic library layer uniformly scans and loads the image acquisition interfaces of various data sources. Each image acquisition interface runs independently, encapsulates the collected original advertising image data into a unified image data object, and puts it into the image data set of the acquisition manager to obtain the basic data of the advertising image.

[0032] It should be noted that in the process of obtaining the original data of advertising images through web crawlers or local reading, this system involves creating image acquisition interfaces from different sources. These interfaces are inherited from the general image acquisition abstract class of the basic library layer. Here, web crawler refers to an automated network data collection tool, and local reading refers to directly obtaining data from the system or a connected storage device. Advertising image data sources include online advertising platforms, local advertising libraries, social media advertising libraries, etc., each of which corresponds to an image acquisition interface for collecting original advertising image data from a specific source.

[0033] Specifically, each image acquisition interface will collect the original data of the advertising image according to the specific configuration of its data source. For example, the interface of the online advertising platform may be configured with a specific URL and access frequency, while the interface of the social media advertising library may require a specific API key and access rights. The acquisition manager of the basic library layer is responsible for uniformly scanning and loading these image acquisition interfaces and coordinating their operations. The collected original data of the advertising image will be encapsulated into a unified image data object and placed in the image data collection of the acquisition manager to obtain the basic data of the advertising image.

[0034] Preferably, in order to improve the efficiency and accuracy of data collection, the collection manager can be optimized so that it can dynamically adjust the collection strategy according to the update frequency and importance of the data source. For example, for frequently updated and important data sources, the collection frequency can be increased; while for less updated data sources, the collection frequency can be appropriately reduced.

[0035] Furthermore, an error handling mechanism can be introduced to deal with situations where network requests fail or data sources are unavailable. In this case, the system can automatically retry or switch to an alternative data source to ensure the continuity and stability of data collection. These optimization measures help improve the robustness and adaptability of the system, enabling it to better cope with different data collection scenarios.

[0036] In some embodiments, the normalizing and cropping the acquired image includes: Normalize the image so that the image pixel value range is between [0,1]. According to the normalization formula

[0037] in, is the original pixel value, and are the minimum and maximum pixel values, respectively. is the normalized pixel value; Crop the image according to the preset cropping size ; in, is the cutting height, For the crop width, crop the image from the center of the image or the specified position to obtain the cropped image data; The data after normalization and cropping are integrated to obtain pre-processed advertising image data. It should be noted that this system normalizes and crops the collected images to adapt to the subsequent feature extraction and optimization processing. Normalization refers to adjusting the pixel value range of the image to between [0,1] to facilitate the processing of algorithms such as neural networks. Cropping refers to cropping a specific area from the image according to a preset size to focus on the key part of the image. These two steps are the key links of image preprocessing and are crucial to improving the effect of subsequent processing.

[0038] Specifically, normalization is achieved through a formula: Normalized pixel value = (original pixel value - minimum pixel value) / (maximum pixel value - minimum pixel value). In this process, it is necessary to determine the maximum and minimum values ​​of each pixel value, and then apply the above formula to convert each pixel value. Cropping involves setting the height and width of the crop, that is, the crop size, which can be determined based on the content of the image and the needs of subsequent processing. The cropping operation can be performed from the center of the image or a specified position to obtain the cropped image data.

[0039] Preferably, the normalization process may further include processing of outliers, that is, correcting outliers to the nearest normal boundary values ​​before normalization. This can be achieved by statistically analyzing the image data, determining the normal range of pixel values, and then adjusting pixel values ​​outside this range to the boundary values. For the cropping process, in addition to cropping according to a preset size, an intelligent cropping algorithm may also be introduced, such as cropping based on image content recognition, so as to ensure that the cropped image area contains the most critical visual information.

[0040] Furthermore, the cropped image can be added with tag information for identifying the image source and processing history in subsequent processing, and the tag information can be stored in the metadata of the image data object in a specific format. These optimization measures can further improve the quality and efficiency of image preprocessing.

[0041] In some embodiments, the normalizing the image so that the image pixel value range is between [0, 1] further includes: When there are outliers in the image pixel values, the outliers are first corrected to the nearest normal boundary value and then normalized; When an image has multiple RGB channels, each channel is normalized separately. It should be noted that this system pays special attention to the processing of outliers in the image pixel values ​​when normalizing the image. Outliers here refer to those pixel values ​​that deviate from the normal data range, which may be caused by noise, damage or other abnormal factors. These outliers are processed to ensure the accuracy of the normalization process and the quality of the image data. Before normalization, the system will first correct these outliers to the nearest normal boundary value, and then perform normalization.

[0042] Specifically, outlier processing can be achieved by setting a threshold range, which defines the boundaries of normal pixel values. If a pixel value exceeds this range, the system will correct it to the nearest boundary value. For example, if a pixel value is abnormally high, the system may correct it to the maximum pixel value; if it is abnormally low, it will be corrected to the minimum pixel value.

[0043] More specifically, when an image has multiple RGB channels, the system will normalize each channel separately to ensure that the pixel values ​​of each channel are correctly adjusted to the range of [0,1].

[0044] Preferably, the system can use a more sophisticated method to identify and handle outliers. For example, statistical methods such as standard deviation or interquartile range (IQR) can be used to determine outliers. This method can more accurately identify outliers that are far away from the majority of data.

[0045] Furthermore, for the normalization of multi-channel images, the system can use an inter-channel comparison method, that is, before normalization, the pixel value distributions of different channels are compared to determine the minimum and maximum values ​​of each channel. This method can ensure that the normalization of each channel is accurate even when there are differences between different channels. These optimization measures can further improve the quality of image preprocessing and the robustness of the system.

[0046] In some embodiments, pre-processing further comprises: Adding tag information to the cropped image for identifying the image source and processing history in subsequent processing, and storing the tag information in metadata of the image data object in a specific format; Furthermore, the resolution change rate is calculated according to the original resolution of the image and the resolution after cropping, which is used to evaluate the degree of loss of image information. It should be noted that, in addition to normalization and cropping, this system also includes the step of adding tag information to the cropped image when preprocessing the acquired image. Tag information refers to metadata used to identify the source and processing history of the image, which can be stored in the metadata of the image data object in a specific format. This step helps to track the source and history of the image in subsequent processing and ensure the traceability of image processing.

[0047] Specifically, the tag information may include the image acquisition time, acquisition location, acquisition equipment information, and any related processing records. This information can be stored in the metadata field of the image, such as EXIF ​​(Exchangeable Image File Format) data, which is a widely used standard for storing digital photo information. Adding tag information to the cropped image can be implemented programmatically in image processing software or libraries, such as using Python's PIL (Python Imaging Library) library to add and modify the image's EXIF ​​data.

[0048] Preferably, the system can further refine the content of the marking information, for example, including the calculation of the resolution change rate of the image. The resolution change rate refers to the change ratio between the original resolution of the image and the resolution after cropping. This indicator is used to evaluate the degree of loss of image information. The resolution change rate can be calculated by the following formula: (original resolution - resolution after cropping) / original resolution. This calculation result can help the system evaluate the impact of the cropping operation on the image quality and provide a reference for subsequent image optimization.

[0049] Furthermore, the system can also use more advanced image processing techniques, such as image hashing or digital watermarking, to enhance the security and integrity of the image. These techniques can add difficult-to-tamper mark information to the image without significantly affecting the visual quality of the image.

[0050] In some embodiments, using neural network technology to load an image acquisition module includes: When the system starts, based on the neural network framework, the customized advertising image loader is used to obtain the codes of each layer in the image acquisition module from the system's image storage directory; Among them, the construction of the advertising image loader is as follows: Suppose the pixel matrix of the advertising image is , whose dimensions are ;in, is the image height, is the image width, is the number of image channels, each pixel in the advertising image The pixel value is ,satisfy , , and each pixel value is a dimensional vector; For an advertisement image, define its advertisement feature vector ; in, It is the dimension of advertising features, which includes the color features of advertising elements. , Ad text features , Advertised product features ; The loader first obtains the basic parameters of the image through the format parsing formula according to the format information of the advertisement image; The format parsing formula is: ; in, For the parsed format parameters, obtain the basic parameters of the image, including color mode and resolution; Then construct an advertising feature index based on the metadata of the advertising image

[0051] in, Metadata includes ad category labels and delivery platform information; Matching the advertisement image with a specific advertisement feature library through an advertisement feature matching formula to determine the initial attributes of the image in the advertisement scene; The advertisement feature matching formula is:

[0052] in, For the matching results; Through the convolutional layer and pooling layer of the neural network, the image data corresponding to each layer of code is extracted to create new feature vectors corresponding to each layer. After adding the extended image features in each layer of code to the new feature vectors, a new feature representation is generated. In the convolution layer, the convolution kernel is used With image Perform convolution operation to extract local features of the image; In the pooling layer, the local features are reduced in dimension using the maximum pooling formula. Instantiate the transmitter of the data transmission layer, the acquisition manager and image monitor of the basic library layer; The instantiated acquisition manager periodically executes each image acquisition interface of the advertisement image data source. It should be noted that this system uses neural network technology to load the image acquisition module, which involves operations based on the neural network framework when the system starts. The neural network framework refers to the software architecture used to build and train neural network models, which provides a series of tools and algorithms to process image data. In this process, the system will obtain various layers of code from the image storage directory through a custom advertising image loader. These codes are the basis for the operation of the system and involve the processing and feature extraction of image data.

[0053] Specifically, the construction of the advertisement image loader involves processing the advertisement image pixel matrix, where the dimension of the pixel matrix is ​​height × width × number of channels. In this dimension, the pixel value of each pixel is a multidimensional vector, representing the color information of the pixel.

[0054] More specifically, the system first needs to define an advertising feature vector, which includes dimensions such as the color features of advertising elements, advertising text features, and advertising product features. The loader will obtain the basic parameters of the image, such as color mode and resolution, based on the format information of the advertising image through the format parsing formula. Then, an advertising feature index is constructed based on the metadata of the advertising image, which includes advertising category labels and delivery platform information. Through the advertising feature matching formula, the system matches the advertising image with a specific advertising feature library to determine the initial attributes of the image in the advertising scene.

[0055] Preferably, the system can further refine the advertisement feature matching formula to improve the accuracy and efficiency of matching. For example, more complex machine learning algorithms, such as Deep Convolutional Neural Networks (DCNNs), can be introduced to extract deep features of images.

[0056] Furthermore, the system can use pre-trained models to speed up the feature extraction process. These models have been trained on a large amount of image data and can quickly identify key features in the image. In the operation of the convolution layer and the pooling layer, the system can adjust the size and step size of the convolution kernel, as well as the window size and step size of the pooling layer to optimize the effect of feature extraction. These adjustments can be customized according to specific application scenarios and performance requirements to achieve the best image processing effect.

[0057] In some embodiments, the convolution kernel is used With image Perform convolution operation, the formula is:

[0058] in, is the convolution result, is the image coordinate, is the convolution kernel coordinate, extracting local features of the image; The maximum pooling formula is:

[0059] in, is the pooling result, is the pooling area.

[0060] It should be noted that when this system uses neural network technology for feature extraction, it involves the process of using convolution kernels to perform convolution operations with images. Convolution operation is a mathematical operation used to extract local features in an image. In this process, the convolution kernel slides on the image, and the dot product between the convolution kernel and the local area of ​​the image is calculated to generate a new feature map. This process helps to identify features such as edges and textures in the image.

[0061] As you can understand, the formula for the convolution operation is: Convolution result = ∑(convolution kernel value * pixel value of the corresponding point in the local area of ​​the image). In this formula, each element of the convolution kernel is multiplied by the pixel value at the corresponding position in the image, and then the products are summed to get the convolution result.

[0062] More specifically, this process is repeated over the entire image, with the convolution kernel sliding over the image, moving a certain step size each time until the entire image is covered. Parameters such as the size, step size, and padding of the convolution kernel can be set as needed to meet different feature extraction requirements.

[0063] Preferably, the system can further refine the parameter settings of the convolution operation, such as selecting convolution kernels of different sizes to capture features of different scales. Small convolution kernels are suitable for capturing detailed features, while large convolution kernels can capture more extensive features.

[0064] Furthermore, the system can use multiple convolution kernels in parallel, each of which extracts different features, and then merge the results to obtain a richer feature representation. In the maximum pooling operation, the system can use pooling windows of different sizes, such as 2x2 or 3x3, and different step sizes to reduce the spatial dimension of the feature map while retaining the most important information. The parameters of these operations can be adjusted according to the specific application scenario and performance requirements to achieve the best image feature extraction effect.

[0065] In some embodiments, the feature extraction module further includes training optimization of the neural network model, including: Construct training and testing datasets based on the collected large amount of advertising image data and the corresponding optimization target labels; The back propagation algorithm is used to update the parameters of the neural network model, and the loss function is:

[0066] in is the true label, is the predicted probability. It should be noted that the feature extraction module of this system not only performs feature extraction, but is also responsible for training and optimizing the neural network model. Training optimization refers to adjusting the parameters of the neural network model through a large amount of data and corresponding labels to improve the prediction accuracy of the model. This process involves building training and test data sets, and using specific algorithms to update model parameters.

[0067] Specifically, the training dataset and the test dataset are constructed from a large amount of collected advertising image data and the corresponding optimized target labels. These datasets are used to train and evaluate the performance of the neural network model.

[0068] More specifically, during the training process, the system uses the back-propagation algorithm to update the parameters of the neural network model. The back-propagation algorithm is a commonly used technique in supervised learning that adjusts the model weights by calculating the gradient of the loss function to minimize the prediction error. The loss function is a function that measures the difference between the model's predicted value and the true label. The loss function used in this system is the cross-entropy loss, and its formula is: Loss = -∑(true label * log(predicted probability)). In this formula, the true label is the known correct label, and the predicted probability is the model's predicted probability for each label.

[0069] Preferably, the system can further refine the training optimization process, for example, by adjusting parameters such as learning rate, batch size, and number of iterations to improve training efficiency and model performance. The learning rate determines the step size of each parameter update, the batch size determines the number of data samples used in each training iteration, and the number of iterations determines the number of complete cycles of the entire training process.

[0070] Furthermore, the system can also use regularization techniques, such as L1 or L2 regularization, to prevent model overfitting and improve the generalization ability of the model. In practical applications, data enhancement techniques, such as random cropping, flipping or color transformation, can also be considered to increase the diversity of training data and thus improve the robustness of the model. These optimization measures help improve the performance and reliability of neural network models in actual advertising image optimization tasks.

[0071] In some embodiments, the optimization processing module is used to analyze and optimize the received advertising image data and display the optimization results, including: performing feature analysis on the received advertising image data using a trained neural network model, and determining optimization targets according to the target audience and delivery platform of the advertising image; According to the optimization goal, the image is color-adjusted through the color balance formula, and the color balance formula is:

[0072] in, is the adjusted color value, is the original color value, Adjust the amount for color; The optimized advertising image is displayed to the user, and the image evaluation indicators before and after optimization are recorded for evaluating the optimization effect. The image evaluation indicators include click-through rate.

[0073] It should be noted that in the optimization processing module, this system analyzes and optimizes the received advertising image data and displays the optimization results. The optimization processing module refers to the part of the system that is responsible for adjusting the image according to the analysis results, with the purpose of improving the visual effect of the image to adapt to specific advertising goals and delivery platforms. This process involves using the trained neural network model to perform feature analysis and determine the optimization target based on the analysis results.

[0074] Specifically, the optimization processing module first uses a neural network model to analyze the features of the advertising image. This model has been trained on a large amount of advertising image data and can identify the key features of the image and predict the best optimization strategy.

[0075] More specifically, based on the results of these feature analyses, the system will determine the optimization goal, such as improving the image contrast, adjusting the color balance, or enhancing the image clarity. Next, the system will adjust the image color according to the optimization goal and use the color balance formula to calculate the adjusted color value. The color balance formula is: adjusted color value = original color value + color adjustment amount. In this formula, the original color value is the current color value of the pixel in the image, and the color adjustment amount is the adjustment parameter calculated based on the optimization goal.

[0076] Preferably, the system can further refine the optimization process steps, such as introducing a multi-stage color adjustment strategy to gradually approach the optimal color effect. In addition, the system can dynamically adjust the color adjustment amount according to the target audience of the advertising image and the characteristics of the delivery platform to achieve a more personalized optimization effect. When displaying the optimized image, the system can provide a user interface to allow users to intuitively compare the images before and after optimization and provide feedback.

[0077] Furthermore, the system can also record image evaluation indicators before and after optimization, such as click-through rate, to evaluate the optimization effect and further adjust the optimization strategy based on these indicators. These refined steps and alternatives help improve the optimization effect of the system and ensure that advertising images can achieve the best display effect on different platforms.

[0078] The above-mentioned embodiments of the present invention have the following beneficial effects: the system can realize the automatic processing and optimization of advertising image data by integrating image acquisition, feature extraction and optimization processing modules. The image acquisition module can obtain the original data of the advertising image on the network or locally, and perform preprocessing, including normalization and cropping, to adapt to the subsequent optimization processing. The feature extraction module uses neural network technology to load the image data, while the optimization processing module performs in-depth analysis and optimization of these data, and finally displays the optimized results. This integrated method can improve the efficiency and effect of advertising image processing, make the advertising image more attractive to the target audience, and improve the click-through rate and conversion rate of the advertisement.

[0079] In addition, the system design also takes into account the contextual management and monitoring of image data, as well as the addition of label information to images and the calculation of resolution change rate. These functions can enhance the management capabilities of image data and the targeted optimization. Through the use of the neural network framework, the system can more accurately extract the features of advertising images and perform targeted optimization based on these features. This deep learning-based method can significantly improve the personalization and adaptability of advertising images, making them more in line with the needs of specific advertising scenarios, thereby improving the overall advertising effect and market competitiveness.

[0080] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0081] The above descriptions are only some preferred embodiments of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. An advertising image optimization system based on neural network, characterized in that: include: Image acquisition module, feature extraction module and optimization processing module; The image acquisition module is configured to acquire raw advertising image data from an advertising image data source by crawling a web page or reading locally to obtain basic advertising image data, preprocess the acquired image, including normalization and cropping, to obtain preprocessed advertising image data, and transmit various types of advertising image data to the optimization processing module using a transmitter of the data transmission layer of the image acquisition module; The feature extraction module is used to load the image acquisition module using neural network technology when the system starts; The optimization processing module is used to analyze and optimize the received advertising image data and display the optimization results.

2. The advertising image optimization system based on neural network according to claim 1, characterized in that: The image acquisition module also includes a basic library layer for providing general image classes and interfaces. The general image classes include image pixel data classes, image color mode classes, image size classes and image acquisition abstract classes, as well as defining image contexts to manage the data contexts of each image at different processing stages, storing image data objects of each image in the preprocessing layer, and defining image monitors to perform periodic monitoring of image data objects in the image contexts, and immediately sending image data that meets the optimization conditions to the optimization processing module.

3. The neural network-based advertising image optimization system according to claim 2, characterized in that: The method of obtaining the original data of the advertisement image by crawling a web or reading locally from the advertisement image data source to obtain the basic data of the advertisement image includes: In the advertising image data source, create image acquisition interfaces for the online advertising platform, local advertising library, and social media advertising library. All of them inherit the common image acquisition abstract class of the basic library layer. In each image acquisition interface, the original advertising image data is collected according to the configuration of the respective data source. The universal acquisition manager of the basic library layer uniformly scans and loads the image acquisition interfaces of various data sources. Each image acquisition interface runs independently, encapsulates the collected original advertising image data into a unified image data object, and puts it into the image data collection of the acquisition manager to obtain the basic data of the advertising image.

4. The neural network-based advertising image optimization system according to claim 2, characterized in that: The normalization and cropping of the collected image includes: Normalize the image so that the image pixel value range is between [0,1]. The normalization formula is: in, is the original pixel value, and are the minimum and maximum pixel values, respectively. is the normalized pixel value; Crop the image according to the preset cropping size ; in, is the cutting height, For cropping width, crop the image from the center or specified position to obtain the cropped image data; The data after normalization and cropping are integrated to obtain pre-processed advertising image data.

5. The neural network-based advertising image optimization system according to claim 4, characterized in that: The normalization process of the image so that the pixel value of the image is within the range of [0, 1] further includes: When there are outliers in the image pixel values, the outliers are first corrected to the nearest normal boundary value and then normalized; When an image has multiple RGB channels, each channel is normalized separately.

6. The neural network-based advertising image optimization system according to claim 4, characterized in that: Preprocessing also includes: Adding tag information to the cropped image for identifying the image source and processing history in subsequent processing, and storing the tag information in metadata of the image data object in a specific format; Furthermore, the resolution change rate is calculated based on the original resolution of the image and the resolution after cropping to evaluate the degree of image information loss.

7. The neural network-based advertising image optimization system according to claim 2, characterized in that: Use neural network technology to load the image acquisition module, including: When the system starts, based on the neural network framework, the customized advertising image loader is used to obtain the codes of each layer in the image acquisition module from the system's image storage directory; The ad image loader is constructed as follows: Assume the pixel matrix of the advertising image is , whose dimensions are ;in, is the image height, is the image width, is the number of image channels, each pixel in the advertising image The pixel value is ,satisfy , , and each pixel value is a dimensional vector; For an advertisement image, define its advertisement feature vector ; in, Ad feature dimension, including the color features of ad elements , ad text features , Advertisement product characteristics ; The loader obtains the basic parameters of the image based on the format information of the advertisement image through the format parsing formula; The format parsing formula is: ; in, For the parsed format parameters, obtain the basic parameters of the image, including color mode and resolution; Build an ad feature index based on the metadata of the ad image: in, Metadata includes ad category labels and delivery platform information; Matching the advertising image with a specific advertising feature library through an advertising feature matching formula to determine the initial attributes of the image in the advertising scene; The ad feature matching formula is: in, For matching results; Through the convolutional and pooling layers of the neural network, the image data corresponding to each layer of code is extracted to create new feature vectors. After adding the expanded image features in each layer of code to the new feature vectors, a new feature representation is generated. In the convolution layer, the convolution kernel is used With image Perform convolution operation to extract local features of the image; In the pooling layer, the local features are reduced in dimension by the maximum pooling formula. Instantiate the transmitter of the data transmission layer, the acquisition manager and image monitor of the basic library layer; The instantiated acquisition manager periodically executes each image acquisition interface of the advertisement image data source.

8. The neural network-based advertising image optimization system according to claim 7, characterized in that: Using convolution kernel With image Perform convolution operation, the formula is: in, is the convolution result, are image coordinates, is the convolution kernel coordinate, extracting local features of the image; The maximum pooling formula is: in, is the pooling result, is the pooling area.

9. The neural network-based advertising image optimization system according to claim 7, characterized in that: The feature extraction module also includes training optimization of the neural network model, including: Construct training and test datasets based on the large amount of collected advertising image data and the corresponding optimization target labels; The back propagation algorithm is used to update the parameters of the neural network model, and the loss function is: in is the true label, is the predicted probability.

10. The neural network-based advertising image optimization system according to claim 3, characterized in that: The optimization processing module is used to analyze and optimize the received advertising image data and display the optimization results, including: performing feature analysis on the received advertising image data using a trained neural network model and determining optimization targets based on the target audience and delivery platform of the advertising image; According to the optimization goal, the image is adjusted by the color balance formula. The color balance formula is: in, is the adjusted color value, is the original color value, Adjust the amount for color; The optimized advertising image is displayed to the user, and the image evaluation indicators before and after optimization are recorded for evaluating the optimization effect. The image evaluation indicators include click-through rate.

Images