Information processing method, system and equipment for image content filtering and feature classification

Abstract

The invention relates to an information processing method, system and device for image content filtering and feature classification, and the method comprises the steps: carrying out the denoising of an input image, extracting the multi-level features of the denoised image through a convolutional neural network, and generating a high-dimensional feature vector; performing dimension reduction processing on the high-dimensional feature vector to obtain a feature vector set after dimension reduction; calculating a feature vector center point of each classification category and constructing a category center point set; according to the Euclidean distance between the center points of each category, the classification interval is dynamically adjusted to optimize the classification boundary; evaluating the importance of each feature in the high-dimensional feature vector to classification by using an information gain method, generating a feature weight set, and performing weighting processing on the high-dimensional feature vector; and in the weighted feature space, a classification hyperplane is constructed by adopting a support vector machine algorithm, and image classification is carried out in combination with a dynamically adjusted classification interval, so that the purposes of accurately and efficiently classifying images and adaptively coping with new category and content changes in a complex scene are achieved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an information processing method, system and device for image content filtering and feature classification. Background Art

[0002] With the rapid development of information technology, image data is increasingly used in various fields. However, existing related technologies have many problems in practical application. In terms of image denoising, some denoising methods will inevitably damage the original details of the image while removing noise, affecting the accuracy of subsequent processing. In the feature extraction stage, some traditional methods often find it difficult to extract effective features when facing complex images, and may generate a large amount of redundant data, increasing the processing burden. For the dimensionality reduction of high-dimensional feature vectors, current methods are difficult to retain key information well while reducing the dimension, which easily affects the accuracy of classification.

[0003] In the process of calculating the center points of classification categories and constructing the center point sets, existing technologies are prone to calculation deviations when dealing with large-scale data or data with complex distribution, resulting in unreasonable classification. In terms of optimizing classification intervals and adjusting classification boundaries, traditional methods are relatively fixed and single, and cannot be flexibly adjusted according to the actual distribution of data. When evaluating the importance of features and weighting feature vectors, existing methods are also inaccurate and unreasonable, making it difficult for classification models to fully play their role. Current image content filtering and feature classification technologies have certain limitations, and a more advanced information processing method is urgently needed to overcome these problems and meet the ever-increasing requirements for image information processing in different fields. Summary of the invention

[0004] The main purpose of the present invention is to provide an information processing method, system and device for image content filtering and feature classification, so as to achieve the purpose of accurately and efficiently classifying images in complex scenarios and adaptively responding to new categories and content changes.

[0005] To achieve the above object, the present invention provides an information processing method for image content filtering and feature classification, comprising the following steps: De-noise the input image, extract the multi-level features of the denoised image through a convolutional neural network, and generate a high-dimensional feature vector; Performing dimensionality reduction processing on the high-dimensional feature vector to obtain a feature vector set after dimensionality reduction; Based on the feature vector set after dimension reduction, the feature vector center point of each classification category is calculated, and a category center point set is constructed; According to the Euclidean distance between the center points of each category, the classification interval is dynamically adjusted to optimize the classification boundary; Using the information gain method to evaluate the importance of each feature in the high-dimensional feature vector to classification, generating a feature weight set, and performing weighted processing on the high-dimensional feature vector; In the weighted feature space, the support vector machine algorithm is used to construct the classification hyperplane, and the image classification is performed in combination with the dynamically adjusted classification interval.

[0006] Furthermore, the step of denoising the input image, extracting multi-level features of the denoised image through a convolutional neural network, and generating a high-dimensional feature vector includes: Performing Gaussian filtering and denoising on the input image, including reading the input image pixel data, selecting the corresponding Gaussian kernel size according to the image size, generating a Gaussian kernel weight matrix and performing normalization, calculating the weighted average value through convolution operation to update the original pixel value, and outputting and storing the denoised image; Taking the denoised image as input data, setting a first convolution layer at the front end of the network, and defining the first convolution kernel size and step size parameters; Extracting shallow edge features of the denoised image according to the set first-layer convolution kernel size to generate a first feature map; A second convolutional layer is set in the middle layer of the network to extract the texture abstract features of the first feature map and generate a second feature map; Setting a third convolutional layer at the back end of the network to extract the semantic high-level features of the second feature map and generate a third feature map; Converting the third feature map into a high-dimensional vector set through a fully connected layer to obtain a high-dimensional feature value; The high-dimensional feature values ​​are stored as data streams to obtain high-dimensional feature vectors.

[0007] Furthermore, the step of performing dimensionality reduction processing on the high-dimensional feature vector to obtain a reduced-dimensional feature vector set includes: Obtain the data value of the high-dimensional feature vector and determine the information content of the feature quantity using the principal component analysis method; Perform dimensionality reduction on high-dimensional data and calculate the dimensionality value of the principal component; Extract the vector data after dimensionality reduction, transform the feature vector through linear transformation, and obtain the feature vector set after dimensionality reduction.

[0008] Furthermore, the step of calculating the feature vector center point of each classification category based on the feature vector set after dimension reduction and constructing a category center point set includes: According to the feature vector set after dimension reduction, the mean point of all samples in each classification category is calculated, and the mean point is the center point of the feature vector of the corresponding category; When the number of classification categories exceeds a preset threshold, the K-means clustering algorithm is used to classify the feature vector set; For each divided category, recalculate the corresponding feature vector mean point; Construct a set of category center points to store the mean point data of all categories; When new samples or categories are added, the mean points of each category are recalculated based on the updated feature vector set, and the category center point set is updated synchronously.

[0009] Furthermore, the step of dynamically adjusting the classification interval to optimize the classification boundary according to the Euclidean distance between the center points of each category includes: Based on the category center point set, the coordinates of the center points of all categories are obtained, and the Euclidean distance between every two center points is calculated to obtain a distance matrix; According to the minimum distance value in the distance matrix, determine the benchmark value of the initial classification interval; If the distance between the center points of two categories in the distance matrix is ​​less than the preset threshold, the dynamic weight adjustment algorithm is used to calculate the boundary offset of each category; Regenerating the classification boundary line according to the boundary offset, and optimizing the position of the classification hyperplane using a support vector machine algorithm; The optimized classification boundary is verified by the silhouette coefficient. If the silhouette coefficient is lower than the preset threshold, the classification boundary is iteratively adjusted again until the requirements are met.

[0010] Furthermore, the step of using the information gain method to evaluate the importance of each feature in the high-dimensional feature vector to classification, generating a feature weight set, and performing weighted processing on the high-dimensional feature vector includes: Obtain high-dimensional feature vector data containing multiple samples, each sample corresponds to a set of feature values ​​and category labels; The information gain method is used to calculate the importance of each feature in the classification task and obtain a preliminary feature weight set; Normalizing the preliminary feature weight set to generate a normalized feature weight set; According to the normalized feature weight set, the original high-dimensional feature vector is weighted to obtain the weighted feature vector.

[0011] Furthermore, the step of constructing a classification hyperplane using a support vector machine algorithm in the weighted feature space and performing image classification in combination with a dynamically adjusted classification interval includes: Obtain the original feature values ​​of the image set, calculate the weighted value of each feature in the feature space according to the feature weight set, and generate a weighted feature vector; Using a support vector machine algorithm, the weighted feature vector is input into a classifier to construct an initial classification hyperplane; According to the distribution density of the samples in the feature space, the adjustment coefficient of the classification interval is dynamically calculated to optimize the position of the classification hyperplane; If the classification accuracy is lower than the preset threshold, the feature weight value is iteratively adjusted based on the gradient descent algorithm to regenerate the classification hyperplane; The classification results are compared with the preset classification labels. If they are inconsistent, the weight values ​​are readjusted and the classification hyperplane is iteratively optimized until the classification accuracy reaches the standard.

[0012] Furthermore, after the step of constructing a classification hyperplane using a support vector machine algorithm in the weighted feature space and classifying the image using the dynamically adjusted classification interval, the method further includes: Extract image features from the input image and determine the matching degree between the current feature vector and the category center point set; If the matching degree is lower than the preset threshold, it is judged that the image content has changed or a new category has been added; Perform cluster analysis on the feature vectors corresponding to the newly added categories or changed content, and calculate the center point of the newly added categories; According to the updated set of category center points, the support vector machine classifier is retrained to generate a new classification hyperplane; The new classification hyperplane is used to classify image features. If the classification error exceeds a preset range, the classification hyperplane parameters are fine-tuned through an incremental learning algorithm.

[0013] The present invention also provides an information processing system for image content filtering and feature classification, comprising: A feature extraction unit is used to perform denoising on the input image, extract multi-level features of the denoised image through a convolutional neural network, and generate a high-dimensional feature vector; A dimension reduction unit, used for performing dimension reduction processing on the high-dimensional feature vector to obtain a feature vector set after dimension reduction; A calculation unit, used to calculate the feature vector center point of each classification category based on the feature vector set after dimension reduction, and construct a category center point set; An interval adjustment unit is used to dynamically adjust the classification interval and optimize the classification boundary according to the Euclidean distance between the center points of each category; A weight processing unit, used to evaluate the importance of each feature in the high-dimensional feature vector to classification using an information gain method, generate a feature weight set, and perform weighted processing on the high-dimensional feature vector; The classification unit is used to construct a classification hyperplane in the weighted feature space using a support vector machine algorithm and classify images in combination with a dynamically adjusted classification interval.

[0014] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned information processing method for image content filtering and feature classification when executing the computer program.

[0015] The information processing method, system and device for image content filtering and feature classification provided by the present invention have the following beneficial effects: the present invention can more accurately identify various types of images by first denoising the image and then extracting features using a convolutional neural network. For example, in satellite image and medical image analysis scenarios, this method can effectively reduce misjudgments; and classification judgment based on the center point of each category feature vector can make the results of tasks such as fruit image classification more accurate and reliable. Dimensionality reduction of high-dimensional feature vectors not only speeds up the processing speed of systems such as face recognition, but also reduces storage requirements; using the information gain method to evaluate and weight processing features can significantly improve the classification efficiency of industrial product image detection. When new image categories appear, such as new product images on e-commerce platforms, this method can quickly achieve classification; and multiple methods work together to make it widely applicable to different types of images and multiple application fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of an information processing method for image content filtering and feature classification in one embodiment of the present invention; Figure 2 is a structural block diagram of an information processing system for image content filtering and feature classification 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.

[0017] 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

[0018] 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.

[0019] Reference Figure 1 , is a flow chart of an information processing method for image content filtering and feature classification proposed by the present invention, comprising the following steps: S1, denoise the input image, extract the multi-level features of the denoised image through a convolutional neural network, and generate a high-dimensional feature vector; S2, performing dimensionality reduction processing on the high-dimensional feature vector to obtain a feature vector set after dimensionality reduction; S3, based on the feature vector set after dimension reduction, calculating the feature vector center point of each classification category, and constructing a category center point set; S4, dynamically adjusts the classification interval to optimize the classification boundary according to the Euclidean distance between the center points of each category; S5, using the information gain method to evaluate the importance of each feature in the high-dimensional feature vector to the classification, generating a feature weight set, and performing weighted processing on the high-dimensional feature vector; S6, in the weighted feature space, the support vector machine algorithm is used to construct the classification hyperplane, and the image classification is performed in combination with the dynamically adjusted classification interval.

[0020] As described in step S1 above, the input image is denoised, and the multi-level features of the denoised image are extracted through a convolutional neural network to generate a high-dimensional feature vector. First, the pixel data of the input image is read, and the appropriate Gaussian kernel size is selected according to the image size. A large kernel is selected for a large image to smooth the whole, and a small kernel is selected for a small image to preserve the details. A Gaussian kernel weight matrix is ​​generated and normalized. When performing a convolution operation on the image, mirror symmetry or 0 padding is used for edge pixels to ensure that the convolution is normal. After denoising, the image enters the convolutional neural network. At the front end of the network, the first convolution layer defines the convolution kernel size and step size parameters to extract shallow edge features to generate the first feature map; the second convolution layer in the middle layer uses a smaller convolution kernel to extract texture abstract features to generate the second feature map; the third convolution layer at the back end uses a smaller convolution kernel to extract semantic high-level features to generate the third feature map, and finally it is converted into a high-dimensional vector set through a fully connected layer and stored as a high-dimensional feature vector, making the feature extraction more accurate and comprehensive.

[0021] As described in step S2 above, the high-dimensional feature vector is subjected to dimensionality reduction processing to obtain a set of feature vectors after dimensionality reduction. Dimensionality reduction is performed using the principal component analysis method. First, the high-dimensional feature vector data value is obtained to determine the amount of feature information. The covariance matrix is ​​calculated and the eigenvalue is decomposed, and the dimension reduction dimension is determined based on the cumulative contribution rate. If the cumulative contribution rate of the first few principal components is more than 80%, the dimensions corresponding to these principal components are selected. The dimensionality reduction vector data is extracted, and the dimensionality reduction feature vector set is obtained through linear transformation to reduce data redundancy and improve processing efficiency. For example, when processing a large amount of image feature data, dimensionality reduction can make subsequent calculations more efficient.

[0022] As described in step S3 above, based on the feature vector set after dimension reduction, the feature vector center point of each classification category is calculated, and a category center point set is constructed. According to the feature vector set after dimension reduction, the mean point of the samples in each classification category is calculated as the center point. If the number of categories exceeds the preset threshold, the K-means clustering algorithm is used to divide the categories and recalculate the mean point. A set is constructed to store the mean point data of all categories. When there are new samples or categories, the mean point is recalculated based on the updated feature vector set and the set is updated synchronously. For example, in the classification of fruit images, when new fruit varieties are added, the center point can be adjusted in time to ensure accurate classification.

[0023] As described in the above step S4, according to the Euclidean distance between the center points of each category, the classification interval is dynamically adjusted to optimize the classification boundary. The center point coordinates are obtained based on the category center point set, and the Euclidean distance between each two is calculated to obtain the distance matrix, and the initial classification interval reference value is determined by the minimum distance value. If the distance between the center points of two categories is less than the preset threshold, the boundary offset is calculated using the dynamic weight adjustment algorithm, and the classification boundary line is regenerated accordingly, and the classification hyperplane position is optimized using the support vector machine algorithm. It is verified by the contour coefficient, and it is iteratively adjusted if it does not meet the standard. For example, when distinguishing similar flower varieties, the classification boundary can be optimized accordingly to improve the classification accuracy.

[0024] As described in step S5 above, the information gain method is used to evaluate the importance of each feature in the high-dimensional feature vector to the classification, a feature weight set is generated, and the high-dimensional feature vector is weighted. High-dimensional feature vector data containing multiple samples is obtained, each sample has a feature value and a category label. The information gain method is used to calculate the importance of each feature to obtain a preliminary feature weight set, which is normalized to generate a normalized feature weight set. Based on this, the original high-dimensional feature vector is weighted to obtain a weighted feature vector, highlighting key features and suppressing minor features. For example, in image recognition, important features can play a greater role.

[0025] As described in step S6 above, in the weighted feature space, the support vector machine algorithm is used to construct a classification hyperplane, and the image classification is performed in combination with the dynamically adjusted classification interval. The original eigenvalues ​​of the image set are obtained to calculate the weighted value to obtain the weighted eigenvector, which is input into the support vector machine classifier to construct the initial classification hyperplane. According to the distribution of samples in the feature space, the concept of the distance from the sample to the hyperplane is introduced to calculate the classification interval adjustment amount to optimize the hyperplane. If the classification deviation exceeds the threshold, the weighted value is iteratively adjusted to determine the final classification surface. For example, after the hyperplane is constructed for the first time, the average distance from the sample to the plane is calculated, and the adjustment coefficient is set according to the distance and incorporated into the objective function to retrain the SVM. If the classification accuracy does not reach the preset threshold, adjust the weight of the wrongly classified sample and retrain until the classification deviation is less than the threshold, complete the image classification and ensure the classification accuracy.

[0026] Determine the matching degree between the image features and the set of category center points. If the matching degree is low, process the changes in image content or the addition of new categories, perform cluster analysis on the relevant feature vectors to calculate the center points of the new categories, retrain the support vector machine to generate a new hyperplane, and use the incremental learning algorithm to fine-tune the hyperplane parameters when the classification error exceeds the range, so that the model can adapt to the changes and improve the classification accuracy. For example, in a scene where new image types are constantly appearing, the model can continuously optimize the classification effect.

[0027] In one embodiment, an intelligent security monitoring system processes images with a resolution of 640×480 pixels collected by a monitoring camera to identify objects such as pedestrians, cars, and motorcycles.

[0028] Perform Gaussian filtering to denoise the noisy image. Since the image size is larger than the preset threshold (set to 300×300), select a 7×7 Gaussian kernel and set the standard deviation . The Gaussian kernel weights are calculated and normalized by Gaussian function, and the image is denoised by convolution operation. The edge pixels are filled in a mirror-symmetric manner. The denoised image is input into the convolutional neural network. The first convolution layer at the front end of the network sets the convolution kernel to 5×5 and the step size to 2 to extract shallow edge features to generate the first feature map; the second convolution layer in the middle layer uses a 3×3 convolution kernel to extract texture abstract features to generate the second feature map; the third convolution layer at the back end uses a 1×1 convolution kernel to extract semantic high-level features to generate the third feature map. Finally, the third feature map is converted into a 2048-dimensional high-dimensional feature vector through a fully connected layer.

[0029] Use principal component analysis to reduce the dimensionality of the 2048-dimensional vector. Calculate the covariance matrix (Here n=100, indicating processing 100 frames of images), after eigenvalue decomposition, if the cumulative contribution rate of the first 100 principal components reaches 95% (i.e. ), the data is reduced to 100 dimensions, and the reduced-dimensional feature vector set is obtained after transformation.

[0030] Perform operations based on the feature vector after dimensionality reduction. Assume that there are 40 pedestrian samples, 30 car samples, and 30 motorcycle samples in the current data set. Calculate the mean point of the feature vector of each category as the center point. For example, for the pedestrian category, sum the dimensions of the 100-dimensional feature vector of the 40 samples and divide by 40 to get the center point of the pedestrian category. Construct these center points into a set for storage. When new samples are added, recalculate the mean point of each category and update the set.

[0031] Calculate the Euclidean distance between the category center points to get the distance matrix. Suppose the category center point set is (corresponding to the center points of pedestrian, car, and motorcycle categories respectively), the Euclidean distance matrix is ​​D, Represents the Euclidean distance between the center points of categories i and j. Here, the adjustment coefficient k = 0.5 is introduced (the k value can be adjusted according to the actual data distribution and classification effect), and the classification interval is calculated using the formula:

[0032] Formula represents classification interval Based on the minimum distance Make adjustments. represents the adjustment amount based on the difference between the current distance and the minimum distance, and k plays a role in controlling the adjustment range. (distance between pedestrian and car category center point) is 10, is 8, then the adjusted classification interval The classification boundary line is regenerated according to the adjusted classification interval, and the hyperplane is optimized with the help of support vector machine. It is verified by the silhouette coefficient (the preset threshold is 0.6). If it does not meet the standard, it is iteratively adjusted.

[0033] Information gain is used to evaluate the feature importance of 100 samples (including 100-dimensional feature values ​​and category labels), and the initial weight set is normalized to obtain a standardized set, and then the original vector is weighted. With the input of new data, in order to dynamically update the feature weights, the learning rate is set . Assume that at the tth update, the weight of the i-th feature is ω t i , the information gain is , the correlation between the i-th feature in the new sample and the classification result It is obtained by calculating the correlation between the new sample feature value and the corresponding category label (for example, using the Pearson correlation coefficient). Use the formula to update the feature weight:

[0034] Among them, ω t i is the weight of the current feature, is the learning rate, which is used to control the step size of weight update to avoid the update amplitude being too large or too small, resulting in model instability or slow convergence. It reflects the closeness of the correlation between the feature and the classification result in the new sample. Indicates the importance of the feature in the classification task. Multiply the three and add the current weight ω t i , we get the updated weight For example, the current weight of a feature , information gain , the correlation between the feature and the classification result in the new sample , then the updated weight .

[0035] In the weighted feature space, support vector machine is used to construct classification hyperplane for classification. The weighted feature vector is calculated and input into the classifier to obtain the initial hyperplane, and the classification interval coefficient is adjusted according to the sample distribution density. The classification accuracy threshold is set to 85%. If the threshold is not reached, the gradient descent algorithm is used to adjust the weight and update the hyperplane. The classification results are compared with the preset labels. If there is inconsistency, readjustment is performed until the accuracy reaches the standard.

[0036] After completing the above image classification, extract the image features of the newly input image, calculate the Euclidean distance between the current feature vector and each center point in the category center point set, and judge the matching degree. If the matching degree is lower than the preset threshold (set to 10), it is determined that the image content may have changed or there are new categories. Use the K-means clustering algorithm for clustering analysis to calculate the center point of the new category. Retrain the support vector machine classifier based on the updated category center point set to generate a new classification hyperplane. Use the new classification hyperplane to classify the image features. If the classification error exceeds the preset range (such as the classification accuracy is less than 80%), use the incremental learning algorithm such as the online support vector machine (Online SVM) to fine-tune the classification hyperplane parameters. By continuously iterating this process, the model can continuously adapt to the newly emerging object categories and maintain a high classification accuracy.

[0037] The accuracy, recall, and F1-score are used to evaluate the model performance. The accuracy is calculated on the test set (containing 200 frames of images). , if the model accurately identifies 170 objects, the accuracy is Taking the pedestrian category as an example, the recall rate , if there are actually 50 pedestrians and 40 are correctly identified, then the pedestrian recall rate is F1 value , the F1 value of the pedestrian category is These indicators are used to evaluate the model performance. If the requirements are not met (such as the F1 value is lower than 80%), the model parameters (such as convolution kernel size, step length, etc.) are adjusted or the training data is increased and retrained to improve the practicality and reliability of the model in security monitoring.

[0038] Reference Figure 2 , is a structural block diagram of an information processing system for image content filtering and feature classification in one embodiment of the present invention, comprising: A feature extraction unit is used to perform denoising on the input image, extract multi-level features of the denoised image through a convolutional neural network, and generate a high-dimensional feature vector; A dimension reduction unit, used for performing dimension reduction processing on the high-dimensional feature vector to obtain a feature vector set after dimension reduction; A calculation unit, used to calculate the feature vector center point of each classification category based on the feature vector set after dimension reduction, and construct a category center point set; An interval adjustment unit is used to dynamically adjust the classification interval and optimize the classification boundary according to the Euclidean distance between the center points of each category; A weight processing unit, used to evaluate the importance of each feature in the high-dimensional feature vector to classification using an information gain method, generate a feature weight set, and perform weighted processing on the high-dimensional feature vector; The classification unit is used to construct a classification hyperplane in the weighted feature space using a support vector machine algorithm and classify images in combination with a dynamically adjusted classification interval.

[0039] For the specific implementation of each unit in the above device example, please refer to the above method embodiment, which will not be repeated here.

[0040] Reference Figure 3 In 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.

[0041] 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.

[0042] In summary, the input image is denoised, and the multi-level features of the denoised image are extracted through a convolutional neural network to generate a high-dimensional feature vector; the high-dimensional feature vector is subjected to dimensionality reduction processing to obtain a feature vector set after dimensionality reduction; based on the feature vector set after dimensionality reduction, the feature vector center point of each classification category is calculated, and a category center point set is constructed; according to the Euclidean distance between the center points of each category, the classification interval is dynamically adjusted to optimize the classification boundary; the information gain method is used to evaluate the importance of each feature in the high-dimensional feature vector to the classification, a feature weight set is generated, and the high-dimensional feature vector is weighted; in the weighted feature space, the support vector machine algorithm is used to construct a classification hyperplane, and the image is classified in combination with the dynamically adjusted classification interval, so as to achieve the purpose of accurately and efficiently classifying images in complex scenes and adaptively responding to new categories and content changes. 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.

[0043] 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.

[0044] 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. An information processing method for image content filtering and feature classification, characterized in that: The following steps are involved: De-noise the input image, extract the multi-level features of the denoised image through a convolutional neural network, and generate a high-dimensional feature vector; Performing dimensionality reduction processing on the high-dimensional feature vector to obtain a feature vector set after dimensionality reduction; Based on the feature vector set after dimension reduction, the feature vector center point of each classification category is calculated, and a category center point set is constructed; According to the Euclidean distance between the center points of each category, the classification interval is dynamically adjusted to optimize the classification boundary; Using the information gain method to evaluate the importance of each feature in the high-dimensional feature vector to classification, generating a feature weight set, and performing weighted processing on the high-dimensional feature vector; In the weighted feature space, the support vector machine algorithm is used to construct the classification hyperplane, and the image classification is performed in combination with the dynamically adjusted classification interval.

2. The information processing method for image content filtering and feature classification according to claim 1, characterized in that: The step of denoising the input image, extracting multi-level features of the denoised image through a convolutional neural network, and generating a high-dimensional feature vector includes: Performing Gaussian filtering and denoising on the input image, including reading the input image pixel data, selecting the corresponding Gaussian kernel size according to the image size, generating a Gaussian kernel weight matrix and performing normalization, calculating the weighted average value through convolution operation to update the original pixel value, and outputting and storing the denoised image; Taking the denoised image as input data, setting a first convolution layer at the front end of the network, and defining the first convolution kernel size and step size parameters; Extracting shallow edge features of the denoised image according to the set first-layer convolution kernel size to generate a first feature map; A second convolutional layer is set in the middle layer of the network to extract the texture abstract features of the first feature map and generate a second feature map; Setting a third convolutional layer at the back end of the network to extract the semantic high-level features of the second feature map and generate a third feature map; Converting the third feature map into a high-dimensional vector set through a fully connected layer to obtain a high-dimensional feature value; The high-dimensional feature values ​​are stored as data streams to obtain high-dimensional feature vectors.

3. The information processing method for image content filtering and feature classification according to claim 1, characterized in that: The step of performing dimensionality reduction processing on the high-dimensional feature vector to obtain a feature vector set after dimensionality reduction includes: Obtain the data value of the high-dimensional feature vector and determine the information content of the feature quantity using the principal component analysis method; Perform dimensionality reduction on high-dimensional data and calculate the dimensionality value of the principal component; Extract the vector data after dimensionality reduction, transform the feature vector through linear transformation, and obtain the feature vector set after dimensionality reduction.

4. The information processing method for image content filtering and feature classification according to claim 1, characterized in that: The step of calculating the feature vector center point of each classification category based on the feature vector set after dimension reduction and constructing a category center point set includes: According to the feature vector set after dimension reduction, the mean point of all samples in each classification category is calculated, and the mean point is the center point of the feature vector of the corresponding category; When the number of classification categories exceeds a preset threshold, the K-means clustering algorithm is used to classify the feature vector set; For each divided category, recalculate the corresponding feature vector mean point; Construct a set of category center points to store the mean point data of all categories; When new samples or categories are added, the mean points of each category are recalculated based on the updated feature vector set, and the category center point set is updated synchronously.

5. The information processing method for image content filtering and feature classification according to claim 1, characterized in that: The step of dynamically adjusting the classification interval and optimizing the classification boundary according to the Euclidean distance between the center points of each category includes: Based on the category center point set, the coordinates of the center points of all categories are obtained, and the Euclidean distance between every two center points is calculated to obtain a distance matrix; According to the minimum distance value in the distance matrix, determine the benchmark value of the initial classification interval; If the distance between the center points of two categories in the distance matrix is ​​less than the preset threshold, the dynamic weight adjustment algorithm is used to calculate the boundary offset of each category; Regenerating the classification boundary line according to the boundary offset, and optimizing the position of the classification hyperplane using a support vector machine algorithm; The optimized classification boundary is verified by the silhouette coefficient. If the silhouette coefficient is lower than the preset threshold, the classification boundary is iteratively adjusted again until the requirements are met.

6. The information processing method for image content filtering and feature classification according to claim 1, characterized in that: The step of using the information gain method to evaluate the importance of each feature in the high-dimensional feature vector to classification, generating a feature weight set, and performing weighted processing on the high-dimensional feature vector includes: Obtain high-dimensional feature vector data containing multiple samples, each sample corresponds to a set of feature values ​​and category labels; The information gain method is used to calculate the importance of each feature in the classification task and obtain a preliminary feature weight set; Normalizing the preliminary feature weight set to generate a normalized feature weight set; According to the normalized feature weight set, the original high-dimensional feature vector is weighted to obtain the weighted feature vector.

7. The information processing method for image content filtering and feature classification according to claim 1, characterized in that: The step of constructing a classification hyperplane using a support vector machine algorithm in the weighted feature space and performing image classification in combination with a dynamically adjusted classification interval includes: Obtain the original feature values ​​of the image set, calculate the weighted value of each feature in the feature space according to the feature weight set, and generate a weighted feature vector; Using a support vector machine algorithm, the weighted feature vector is input into a classifier to construct an initial classification hyperplane; According to the distribution density of the samples in the feature space, the adjustment coefficient of the classification interval is dynamically calculated to optimize the position of the classification hyperplane; If the classification accuracy is lower than the preset threshold, the feature weight value is iteratively adjusted based on the gradient descent algorithm to regenerate the classification hyperplane; The classification results are compared with the preset classification labels. If they are inconsistent, the weight values ​​are readjusted and the classification hyperplane is iteratively optimized until the classification accuracy reaches the standard.

8. The information processing method for image content filtering and feature classification according to claim 1, characterized in that: After the step of constructing a classification hyperplane using a support vector machine algorithm in the weighted feature space and classifying the image using the dynamically adjusted classification interval, the method further includes: Extract image features from the input image and determine the matching degree between the current feature vector and the category center point set; If the matching degree is lower than the preset threshold, it is judged that the image content has changed or a new category has been added; Perform cluster analysis on the feature vectors corresponding to the newly added categories or changed contents, and calculate the center points of the newly added categories; According to the updated set of category center points, the support vector machine classifier is retrained to generate a new classification hyperplane; The new classification hyperplane is used to classify image features. If the classification error exceeds a preset range, the classification hyperplane parameters are fine-tuned through an incremental learning algorithm.

9. An information processing system for image content filtering and feature classification, characterized in that: include: A feature extraction unit is used to perform denoising on the input image, extract multi-level features of the denoised image through a convolutional neural network, and generate a high-dimensional feature vector; A dimension reduction unit, used for performing dimension reduction processing on the high-dimensional feature vector to obtain a feature vector set after dimension reduction; A calculation unit, used to calculate the feature vector center point of each classification category based on the feature vector set after dimension reduction, and construct a category center point set; An interval adjustment unit is used to dynamically adjust the classification interval and optimize the classification boundary according to the Euclidean distance between the center points of each category; A weight processing unit, used to evaluate the importance of each feature in the high-dimensional feature vector to classification using an information gain method, generate a feature weight set, and perform weighted processing on the high-dimensional feature vector; The classification unit is used to construct a classification hyperplane in the weighted feature space using a support vector machine algorithm and classify images in combination with a dynamically adjusted classification interval.

10. 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 information processing method for image content filtering and feature classification described in any one of claims 1 to 7 are implemented.

Images