A method and device for analyzing the probability of hot-selling clothing styles based on big data

Abstract

The present invention relates to the technical field of clothing recommendation, and invents a method for analyzing the probability of clothing styles becoming popular based on big data, comprising: collecting clothing images of clothing styles to be analyzed, and obtaining historical hot spot information corresponding to the clothing images; performing hierarchical feature extraction and style feature fusion operations on clothing images to obtain clothing style features; performing hot spot feature extraction, feature clustering, and hot spot feature fusion operations on historical hot spot information in turn to obtain clothing hot spot features; using historical hot spot information to time-mark clothing hot spot features to obtain periodic hot spot features, and extracting hot spot time series features from periodic hot spot features; performing kernel function mapping and linear regression analysis on clothing style features and hot spot time series features to obtain the probability of style becoming popular. The present invention also proposes a device for analyzing the probability of clothing styles becoming popular based on big data. The present invention can improve the accuracy of the probability analysis of clothing styles becoming popular.

Description

Technical Field

[0001] The present invention relates to the technical field of clothing recommendation, and in particular to a method and device for analyzing the probability of popular clothing styles based on big data. Background Art

[0002] Competition in the apparel industry is fierce, and consumer demand and fashion trends change rapidly. Therefore, understanding which clothing items are likely to become hits is crucial to the success of a business. Through hit probability analysis, brands and retailers can optimize inventory management and increase sales profits in the apparel field.

[0003] Existing analysis methods for hot-selling clothing products are often based on clothing color style, that is, based on the color matching of clothing, the probability of hot-selling clothing is inferred according to historical sales data. In actual applications, the probability of clothing hot-selling products is greatly affected by factors such as fashion trends. Traditional analysis methods based on color style cannot be combined with current fashion trends for analysis, and hot-selling products based on color style cannot represent or accurately measure clothing style information, which may lead to low accuracy in clothing hot-selling analysis. Summary of the invention

[0004] The present invention provides a method and device for analyzing the probability of hot-selling clothing styles based on big data, the main purpose of which is to solve the problem of low accuracy in analyzing hot-selling clothing styles.

[0005] To achieve the above purpose, the present invention provides a method for analyzing the probability of popular clothing styles based on big data, comprising:

[0006] Collect clothing images of the clothing styles to be analyzed, and obtain historical hot spot information corresponding to the clothing images;

[0007] Performing hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features, wherein the performing hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features includes: performing image standardization and multi-level scale splitting on the clothing image to obtain a local clothing image set; performing pixel enhancement operations on the local clothing image set using the following clothing pixel enhancement algorithm to obtain an enhanced clothing image set: in, The pixel coordinates in the enhanced clothing image set are The gray value of The pixel coordinates in the local clothing image set are The gray value of The pixel coordinates in the clothing pixel enhancement algorithm are The horizontal enhancement weight corresponding to the pixel is The pixel coordinates in the clothing pixel enhancement algorithm are The vertical enhancement weight corresponding to the pixel is is the preset enhancement factor, is the absolute value symbol, , is the image horizontal axis coordinate index, , is the image vertical axis coordinate index; a hierarchical clothing image set is generated according to the enhanced clothing image set and the clothing image; contour extraction and contour fitting are performed on the hierarchical clothing image set to obtain clothing contour features; morphological convolution and geometric feature extraction are performed on the hierarchical clothing image set to obtain clothing cutting features; pattern detection and texture extraction are performed on the hierarchical clothing image set to obtain clothing style features; feature dimension reduction screening and feature fusion are performed on the clothing contour features, the clothing cutting features and the clothing style features to obtain clothing style features;

[0008] Extracting hotspot features and clustering the features of the historical hotspot information in turn to obtain a clothing hotspot feature set;

[0009] Using the historical hot spot information, time-marking the clothing hot spot feature set to obtain a periodic hot spot feature set, and performing temporal hot spot analysis on the periodic hot spot feature set to obtain analysis hot spot features;

[0010] Kernel function mapping and linear regression analysis are performed on the clothing style features and the analysis hot spot features to obtain the probability of a style becoming a hit.

[0011] Optionally, the performing image standardization and multi-level scale splitting on the clothing image to obtain a local clothing image set includes:

[0012] Gray-scaling and filtering the clothing image to obtain a denoised clothing image;

[0013] Performing image enhancement and brightness compensation on the denoised clothing image to obtain a compensated clothing image;

[0014] Performing image convolution on the compensated clothing image to obtain clothing features;

[0015] Performing a clothing key point recognition operation on the compensated clothing image according to the clothing features to obtain a clothing key point set;

[0016] The clothing image is positioned and locally scaled according to the clothing key point set to obtain a local clothing image set.

[0017] Optionally, performing contour extraction and contour fitting on the hierarchical clothing image set to obtain clothing contour features includes:

[0018] Performing primary contour filtering on the hierarchical clothing image set to obtain a primary clothing contour set;

[0019] The pixel change rate of each pixel in the hierarchical clothing image set is calculated using the following pixel change rate algorithm: in, The pixel coordinates in the hierarchical clothing image set are The pixel change rate, is the symbol of partial derivative, The pixel coordinates in the hierarchical clothing image set are The gray value of is the image horizontal axis coordinate index, The vertical axis coordinate index of the image;

[0020] Performing contour pixel screening on the hierarchical clothing image set according to the pixel change rate to obtain a secondary clothing contour set;

[0021] Generating a clothing outline set according to the primary clothing outline set and the secondary clothing outline set;

[0022] Performing contour splitting on the clothing contour set to obtain a split clothing contour set;

[0023] Performing polynomial contour fitting on the split garment contour set to obtain a fitted garment contour set;

[0024] The fitted clothing contour set is subjected to contour splicing to obtain clothing contour features.

[0025] Optionally, performing feature dimension reduction screening and feature fusion on the clothing outline feature, the clothing cutting feature and the clothing style feature to obtain the clothing style feature includes:

[0026] Performing feature standardization and feature matrix operations on the clothing outline features, the clothing cutting features, and the clothing style features to obtain a clothing feature matrix;

[0027] Performing a covariance operation on the clothing feature matrix to obtain a clothing covariance feature matrix;

[0028] Performing eigendecomposition on the clothing covariance feature matrix to obtain clothing eigenvalues ​​and clothing eigenvectors;

[0029] Filtering out dimension-reduced clothing features from the clothing feature vector according to the clothing feature value;

[0030] Performing a feature matrix operation on the reduced-dimensional clothing features to obtain a reduced-dimensional clothing feature matrix;

[0031] The reduced-dimensional clothing feature matrix is ​​used to perform feature projection on the clothing feature matrix to obtain clothing style features.

[0032] Optionally, the hot spot feature extraction and feature clustering are performed on the historical hot spot information in sequence to obtain a clothing hot spot feature set, including:

[0033] Performing text segmentation and text vectorization operations on the historical hot spot information to obtain a hot spot text feature set;

[0034] Performing position encoding on the hotspot text feature set to obtain a primary hotspot encoding set;

[0035] Performing semantic encoding on the primary hotspot code set to obtain a secondary hotspot code set;

[0036] Extracting semantic features from the secondary hotspot coding set to obtain hotspot semantic features;

[0037] A feature clustering operation is performed on the hot semantic feature set to obtain a clothing hot cluster set.

[0038] Optionally, performing a feature clustering operation on the hotspot semantic feature set to obtain a hotspot feature set includes:

[0039] The hotspot semantic feature set is randomly divided into a plurality of hotspot semantic feature groups, and a primary hotspot center feature is randomly selected for each hotspot semantic feature group;

[0040] The feature correlation distance between each hotspot semantic feature in the hotspot semantic feature set and each of the primary hotspot center features is calculated using the following correlation distance formula: in, is the feature association distance, refers to the hotspot semantic feature, refers to the primary hot spot center characteristics, is the transpose symbol, is the covariance sign, refers to the covariance of the hotspot semantic features, refers to the covariance between the hotspot semantic feature and the primary hotspot center feature, refers to the covariance between the primary hotspot center feature and the hotspot semantic feature, refers to the covariance of the central characteristics of the primary hotspot;

[0041] updating the hotspot semantic feature groups into standard hotspot semantic feature groups one by one according to the feature association distances;

[0042] Calculate the standard hotspot central features of each of the standard hotspot semantic feature groups, and calculate the central feature association distances between the standard hotspot central features and the corresponding primary hotspot central features one by one;

[0043] According to all the central feature association distances, each of the standard hotspot semantic feature groups is iteratively updated into a corresponding clothing hotspot feature class, and all clothing hotspot feature classes are aggregated into a clothing hotspot feature class set.

[0044] Optionally, the using the historical hot spot information to time-mark the clothing hot spot feature set to obtain a periodic hot spot feature set includes:

[0045] Select the features in the clothing hotspot feature set one by one as target hotspot features, and use the clothing hotspot feature class corresponding to the target hotspot feature in the clothing hotspot feature set as the target feature class;

[0046] Using the cluster center of the target feature class as the target center feature;

[0047] Filtering the time information corresponding to the target hotspot feature from the historical hotspot information as the target time;

[0048] Using the target time to time-mark the target center feature to obtain a periodic hot spot feature;

[0049] The periodic hotspot features of all target hotspot features in the clothing hotspot feature set are aggregated into a target hotspot feature set.

[0050] Optionally, performing a time series hotspot analysis on the periodic hotspot feature set to obtain an analysis hotspot feature includes:

[0051] Sorting each periodic hotspot feature in the periodic hotspot feature set in time sequence to obtain a periodic hotspot feature sequence;

[0052] Performing a stationarity check on the periodic hotspot feature sequence to obtain a stationarity check result;

[0053] Performing differential processing on the periodic hotspot feature sequence according to the stable check result to obtain a stable hotspot feature sequence;

[0054] Performing autocorrelation time series fitting on the stationary hotspot feature sequence to obtain a fitting periodic model;

[0055] The fitting periodic model is used to perform periodic rolling analysis on the stationary hotspot feature sequence to obtain analytical hotspot features.

[0056] Optionally, performing kernel function mapping and linear regression analysis on the clothing style features and the analysis hotspot features to obtain the probability of a style becoming a hit includes:

[0057] Generate a style hotspot feature according to the clothing style feature and the analysis hotspot feature;

[0058] Performing kernel function mapping on the style hotspot features to obtain mapped style hotspot features;

[0059] Linearly combine and bias-weight the mapped style hot spot features to obtain a hot item analysis result;

[0060] The hot-selling analysis results are regressed and outputted to obtain the probability of the style becoming a hot-selling item.

[0061] In order to solve the above problems, the present invention also provides a device for analyzing the probability of popular clothing styles based on big data, the device comprising:

[0062] A data acquisition module, used to collect clothing images of clothing styles to be analyzed, and obtain historical hot spot information corresponding to the clothing images;

[0063] The style extraction module is used to perform hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features, wherein the hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features include: performing image standardization and multi-level scale splitting on the clothing image to obtain a local clothing image set; performing pixel enhancement operations on the local clothing image set using the following clothing pixel enhancement algorithm to obtain an enhanced clothing image set: in, The pixel coordinates in the enhanced clothing image set are The gray value of The pixel coordinates in the local clothing image set are The gray value of The pixel coordinates in the clothing pixel enhancement algorithm are The horizontal enhancement weight corresponding to the pixel is The pixel coordinates in the clothing pixel enhancement algorithm are The vertical enhancement weight corresponding to the pixel is is the preset enhancement factor, is the absolute value symbol, , is the image horizontal axis coordinate index, , is the image vertical axis coordinate index; a hierarchical clothing image set is generated according to the enhanced clothing image set and the clothing image; contour extraction and contour fitting are performed on the hierarchical clothing image set to obtain clothing contour features; morphological convolution and geometric feature extraction are performed on the hierarchical clothing image set to obtain clothing cutting features; pattern detection and texture extraction are performed on the hierarchical clothing image set to obtain clothing style features; feature dimension reduction screening and feature fusion are performed on the clothing contour features, the clothing cutting features and the clothing style features to obtain clothing style features;

[0064] A hotspot clustering module is used to extract hotspot features and perform feature clustering on the historical hotspot information in turn to obtain a clothing hotspot feature set;

[0065] A hot spot analysis module, used to use the historical hot spot information to time-mark the clothing hot spot feature set to obtain a periodic hot spot feature set, and perform a temporal hot spot analysis on the periodic hot spot feature set to obtain an analysis hot spot feature;

[0066] The hot-selling analysis module is used to perform kernel function mapping and linear regression analysis on the clothing style features and the analysis hot spot features to obtain the probability of a style being a hot-selling product.

[0067] By collecting clothing images and obtaining historical hot spot information, the present invention can combine the image information of styles and real-time hot events to extract fashion trends, thereby improving the accuracy of the probability analysis of hot items. By performing hierarchical feature extraction and style feature fusion operations, style features of different levels and scales can be extracted from clothing images, and important style features can be screened out for fusion, thereby enhancing the feature details of clothing style features, thereby improving the accuracy of the probability analysis of hot items.

[0068] By performing hot feature extraction and feature clustering, a hot feature set is obtained, and hot features related to clothing semantics in historical hot information can be extracted, and correlation separation and clustering dimension reduction for hot features can be achieved, reducing the data dimension of hot features, facilitating the subsequent extraction of temporal features, and by performing time labeling and temporal hot spot analysis, the feature dimension of clothing hot feature set can be reduced, the efficiency of extracting temporal features can be improved, and the hot feature changes in future time periods can be further analyzed, thereby improving the accuracy of the probability analysis of explosive models, and by performing kernel function mapping and linear regression analysis, the algorithm of support vector machine can be used to perform the probability analysis of explosive models of styles according to the trend hot information and clothing styles in future time periods, thereby realizing the probability analysis of explosive models in the sales cycle and improving the accuracy of the probability analysis of explosive models. Therefore, the method and device for analyzing the probability of explosive models of clothing styles based on big data proposed in the present invention can solve the problem of low accuracy when performing explosive model analysis of clothing. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A schematic diagram of a flow chart of a method for analyzing the probability of a hit clothing style based on big data provided by an embodiment of the present invention;

[0070] Figure 2 A schematic flow chart of a method for extracting a partial clothing image set provided by an embodiment of the present invention;

[0071] Figure 3 A schematic flow chart of a method for extracting periodic hotspot feature sets provided by an embodiment of the present invention;

[0072] Figure 4 A functional module diagram of a device for analyzing the probability of a hit clothing style based on big data provided by an embodiment of the present invention;

[0073] 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

[0074] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0075] The embodiment of the present application provides a method for analyzing the probability of a clothing style explosion based on big data. The execution subject of the method for analyzing the probability of a clothing style explosion based on big data includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiment of the present application. In other words, the method for analyzing the probability of a clothing style explosion based on big data can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0076] Reference Figure 1 FIG. 1 is a flow chart of a method for analyzing the probability of a clothing style hit based on big data provided by an embodiment of the present invention. In this embodiment, the method for analyzing the probability of a clothing style hit based on big data includes:

[0077] S1. Collect clothing images of clothing styles to be analyzed, and obtain historical hot spot information corresponding to the clothing images.

[0078] In detail, the clothing style to be analyzed refers to the clothing style that needs to be analyzed for the probability of becoming a hit, and the hit probability refers to the probability of a clothing style becoming a hit with sales volume higher than a certain value, wherein an optical camera can be used to capture images of the clothing of the clothing style to be analyzed to obtain clothing images.

[0079] Specifically, the historical hot information refers to real-time hot information when the clothing image is obtained, such as fashion trends in media such as short videos, news about clothing brands, and real-time economic-related news reports, etc., wherein the historical hot information can be obtained by crawling data from an authorized database.

[0080] In the embodiment of the present invention, by collecting clothing images and obtaining historical hot spot information, the trend can be extracted by combining the image information of the style and the real-time hot events, thereby improving the accuracy of the hot item probability analysis.

[0081] S2. Performing hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features.

[0082] In detail, the clothing style features in the field of clothing design refer to the design style, appearance and cutting methods of clothing. The clothing style features include the overall outline of the clothes, cutting methods, collars, sleeves, waist designs, hems and other design features.

[0083] In the embodiment of the present invention, the step of performing hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features includes:

[0084] Performing image standardization and multi-level scale splitting on the clothing image to obtain a local clothing image set;

[0085] Performing pixel enhancement operation on the local clothing image set to obtain an enhanced clothing image set;

[0086] generating a hierarchical clothing image set according to the enhanced clothing image set and the clothing image;

[0087] Performing contour extraction and contour fitting on the hierarchical clothing image set to obtain clothing contour features;

[0088] Performing morphological convolution and geometric feature extraction on the hierarchical clothing image set to obtain clothing cutting features;

[0089] Performing pattern detection and texture extraction on the hierarchical clothing image set to obtain clothing style features;

[0090] The clothing outline features, the clothing cutting features and the clothing style features are subjected to feature dimension reduction screening and feature fusion to obtain clothing style features.

[0091] In detail, the hierarchical clothing image set refers to an image set consisting of global and local images of the clothing images of different sizes after standardization, and the hierarchical clothing image set includes the enhanced clothing image set and the clothing images.

[0092] Specifically, refer to Figure 2 As shown, the clothing image is subjected to image standardization and multi-level scale splitting to obtain a local clothing image set, including:

[0093] S21, graying the clothing image and filtering and denoising the image to obtain a denoised clothing image;

[0094] S22, performing image enhancement and brightness compensation on the denoised clothing image to obtain a compensated clothing image;

[0095] S23, performing image convolution on the compensated clothing image to obtain clothing features;

[0096] S24, performing a clothing key point recognition operation on the compensated clothing image according to the clothing features to obtain a clothing key point set;

[0097] S25, performing image positioning and local scale splitting on the clothing image according to the clothing key point set to obtain a local clothing image set.

[0098] In detail, the image can be grayscaled using a mean grayscale algorithm, the filtering and denoising method can be bilateral filtering or median filtering, the image can be enhanced using a histogram equalization or adaptive histogram equalization method, and brightness compensation can be performed using a gamma correction or contrast stretching method.

[0099] Specifically, the Histogram of Oriented Gradients (HOG) or Scale-Invariant Feature Transform (SIFT) method can be used to locate clothing key points, and each clothing key point in the clothing key point set is an image point at a key position of a clothing collar, cuffs, etc.

[0100] In detail, the image positioning refers to positioning on the clothing image according to the position of each clothing key point in the clothing key point set, and the local scale splitting refers to performing an image splitting operation on the compensated clothing image obtained according to the preset mask clothing key point positions of multiple different scale sizes.

[0101] Specifically, the following clothing pixel enhancement algorithm is used to perform pixel enhancement operations to obtain an enhanced clothing image set: in, The pixel coordinates in the enhanced clothing image set are The gray value of The pixel coordinates in the local clothing image set are The gray value of The pixel coordinates in the clothing pixel enhancement algorithm are The horizontal enhancement weight corresponding to the pixel is The pixel coordinates in the clothing pixel enhancement algorithm are The vertical enhancement weight corresponding to the pixel is is the preset enhancement factor, is the absolute value symbol, , is the image horizontal axis coordinate index, , The vertical axis coordinate index of the image.

[0102] In detail, the clothing pixel enhancement algorithm can enhance the pixel scale of each local clothing image in the local clothing image set, retain more image detail features of the local clothing image set, and improve the feature details of clothing style features.

[0103] Specifically, the contour extraction and contour fitting of the hierarchical clothing image set to obtain clothing contour features includes:

[0104] Performing primary contour filtering on the hierarchical clothing image set to obtain a primary clothing contour set;

[0105] Calculating the pixel change rate of each pixel in the hierarchical clothing image set;

[0106] Performing contour pixel screening on the hierarchical clothing image set according to the pixel change rate to obtain a secondary clothing contour set;

[0107] Generating a clothing outline set according to the primary clothing outline set and the secondary clothing outline set;

[0108] Performing contour splitting on the clothing contour set to obtain a split clothing contour set;

[0109] Performing polynomial contour fitting on the split garment contour set to obtain a fitted garment contour set;

[0110] The fitted clothing contour set is subjected to contour splicing to obtain clothing contour features.

[0111] Specifically, the pixel change rate is calculated using the following pixel change rate algorithm: in, The pixel coordinates in the hierarchical clothing image set are The pixel change rate, is the symbol of partial derivative, The pixel coordinates in the hierarchical clothing image set are The gray value of is the image horizontal axis coordinate index, The vertical axis coordinate index of the image.

[0112] In detail, primary contour filtering can be performed using methods such as Canny edge detection and Sobel operator. The pixel change rate algorithm can identify pixels with a larger change rate as contour pixels based on the change amplitude of the pixel grayscale in the hierarchical clothing image set, thereby improving the accuracy of clothing contour extraction and contour details. The contour pixel screening refers to taking pixels in the hierarchical clothing image set whose pixel change rate is greater than a preset change rate threshold as contour pixels, and splicing the contour pixels into a secondary clothing contour set.

[0113] Specifically, the clothing contour set is a contour set composed of intersection pixels of the primary clothing contour set and the secondary clothing contour set, and the contour splitting refers to splitting into multiple split clothing contour sets according to the distance between contours, wherein the distance between contours can be calculated using the Euclidean distance algorithm.

[0114] In detail, polynomial contour fitting can be performed using a RANSAC contour fitting or active contour model method, and the contour splicing refers to connecting contours whose distances between contours are less than a preset distance threshold into a clothing contour to obtain a clothing contour set.

[0115] Specifically, morphological convolution can be performed using a wavelet transform method or an image convolution method. The geometric feature extraction refers to calculating the length and width of each clothing part in the hierarchical image set. Geometric feature extraction can be performed by performing region recognition and length analysis on the clothing contour set.

[0116] Specifically, the Histogram of Oriented Gradients (HOG) or Scale-Invariant Feature Transform (SIFT) method may be used for pattern detection, and the Gabor filter or image convolution method may be used for texture extraction.

[0117] Specifically, the clothing outline feature, the clothing cutting feature and the clothing style feature are subjected to feature dimension reduction screening and feature fusion to obtain clothing style features, including:

[0118] Performing feature standardization and feature matrix operations on the clothing outline features, the clothing cutting features, and the clothing style features to obtain a clothing feature matrix;

[0119] Performing a covariance operation on the clothing feature matrix to obtain a clothing covariance feature matrix;

[0120] Performing eigendecomposition on the clothing covariance feature matrix to obtain clothing eigenvalues ​​and clothing eigenvectors;

[0121] Filtering out dimension-reduced clothing features from the clothing feature vector according to the clothing feature value;

[0122] Performing a feature matrix operation on the reduced-dimensional clothing features to obtain a reduced-dimensional clothing feature matrix;

[0123] The reduced-dimensional clothing feature matrix is ​​used to perform feature projection on the clothing feature matrix to obtain clothing style features.

[0124] In detail, the feature standardization is to make the mean value of each feature of the clothing contour feature, the clothing cutting feature and the clothing style feature be 0 and the standard deviation be 1, and the feature matrixization operation refers to splicing the standardized feature vectors of the clothing contour feature, the clothing cutting feature and the clothing style feature as the row vectors of the matrix into a clothing feature matrix.

[0125] Specifically, feature decomposition can be performed using a direct eigenvalue decomposition method, a singular value decomposition method or a QR decomposition method. The reduced-dimensional clothing features are composed of several features with larger clothing eigenvalues ​​in the clothing feature vector, and the feature projection refers to calculating the matrix product of the reduced-dimensional clothing feature matrix and the clothing feature matrix.

[0126] In the embodiment of the present invention, by performing hierarchical feature extraction and style feature fusion operations, style features of different levels and scales can be extracted for clothing images, and important style features can be screened out for fusion, thereby enhancing the feature details of clothing style features, thereby improving the accuracy of hot-selling probability analysis.

[0127] S3. Extract hotspot features and cluster features of the historical hotspot information in sequence to obtain a clothing hotspot feature set.

[0128] In the embodiment of the present invention, the hot spot feature extraction and feature clustering are performed on the historical hot spot information in sequence to obtain a clothing hot spot feature set, including:

[0129] Performing text segmentation and text vectorization operations on the historical hot spot information to obtain a hot spot text feature set;

[0130] Performing position encoding on the hotspot text feature set to obtain a primary hotspot encoding set;

[0131] Performing semantic encoding on the primary hotspot code set to obtain a secondary hotspot code set;

[0132] Extracting semantic features from the secondary hotspot coding set to obtain hotspot semantic features;

[0133] A feature clustering operation is performed on the hot semantic feature set to obtain a clothing hot cluster set.

[0134] In detail, the text segmentation method may be the Jieba segmentation method, the text vectorization method may be the text vectorization method of the Word2Vec or BERT model, and the position encoding refers to encoding the position of each feature in the hot text feature set in the text.

[0135] Specifically, a pre-trained Transformer model or BERT model may be used for semantic encoding, and methods such as sentence embedding or attention mechanism may be used for semantic feature extraction. The hotspot semantic feature set is a set of pre-labeled clothing-related semantic features.

[0136] In detail, the hotspot semantic feature set is subjected to a feature clustering operation to obtain a hotspot feature set, including:

[0137] The hotspot semantic feature set is randomly divided into a plurality of hotspot semantic feature groups, and a primary hotspot center feature is randomly selected for each hotspot semantic feature group;

[0138] Calculate the feature association distance between each hotspot semantic feature in the hotspot semantic feature set and each of the primary hotspot center features;

[0139] updating the hotspot semantic feature groups into standard hotspot semantic feature groups one by one according to the feature association distances;

[0140] Calculate the standard hotspot central features of each of the standard hotspot semantic feature groups, and calculate the central feature association distances between the standard hotspot central features and the corresponding primary hotspot central features one by one;

[0141] According to all the central feature association distances, each of the standard hotspot semantic feature groups is iteratively updated into a corresponding clothing hotspot feature class, and all clothing hotspot feature classes are aggregated into a clothing hotspot feature class set.

[0142] Specifically, the feature correlation distance is calculated using the following correlation distance formula: in, is the feature association distance, refers to the hotspot semantic feature, refers to the primary hot spot center characteristics, is the transpose symbol, is the covariance sign, refers to the covariance of the hotspot semantic features, refers to the covariance between the hotspot semantic feature and the primary hotspot center feature, refers to the covariance between the primary hotspot center feature and the hotspot semantic feature, refers to the covariance of the primary hotspot center features.

[0143] In detail, the correlation distance formula can cluster according to the correlation relationship between each feature in the hot spot semantic feature set, thereby reducing the hot spot feature dimension and facilitating subsequent correlation feature fusion operations. Updating the hot spot semantic feature group into a standard hot spot semantic feature group one by one means reallocating each hot spot semantic feature in the hot spot semantic feature group according to the method of nearest allocation of the feature association distance to obtain a standard hot spot semantic feature group.

[0144] Specifically, the standard hotspot center feature refers to the mean feature of the standard hotspot feature group. The method for calculating the center feature association distance is consistent with the method for calculating the feature association distance between each hotspot semantic feature in the hotspot semantic feature set and each of the primary hotspot center features, which will not be repeated here.

[0145] In detail, the center distance-based iterative method can be used to iteratively update each of the standard hotspot semantic feature groups into a corresponding hotspot feature class according to all center feature association distances. The hotspot center feature set is a set of cluster centers of each hotspot feature class in the hotspot feature class set.

[0146] In an embodiment of the present invention, by performing hot spot feature extraction and feature clustering, a hot spot feature set is obtained, which can extract hot spot features related to clothing semantics in historical hot spot information, and realize correlation separation and clustering dimensionality reduction for hot spot features, thereby reducing the data dimension of hot spot features and facilitating the subsequent extraction of time series features.

[0147] S4. Use the historical hot spot information to time-mark the clothing hot spot feature set to obtain a periodic hot spot feature set, and perform temporal hot spot analysis on the periodic hot spot feature set to obtain analysis hot spot features.

[0148] For details, refer to Figure 3 As shown, the clothing hot spot feature set is time-labeled using the historical hot spot information to obtain a periodic hot spot feature set, including:

[0149] S31, selecting features in the clothing hotspot feature set one by one as target hotspot features, and taking the clothing hotspot feature class corresponding to the target hotspot feature in the clothing hotspot feature set as the target feature class;

[0150] S32, taking the cluster center of the target feature class as the target center feature;

[0151] S33, filtering out the time information corresponding to the target hotspot feature from the historical hotspot information as the target time;

[0152] S34, using the target time to time-mark the target center feature to obtain a periodic hot spot feature;

[0153] S35, aggregating the periodic hotspot features of all target hotspot features in the clothing hotspot feature set into a target hotspot feature set.

[0154] In the embodiment of the present invention, the step of performing a temporal hotspot analysis on the periodic hotspot feature set to obtain the analyzed hotspot features includes:

[0155] Sorting each periodic hotspot feature in the periodic hotspot feature set in time sequence to obtain a periodic hotspot feature sequence;

[0156] Performing a stationarity check on the periodic hotspot feature sequence to obtain a stationarity check result;

[0157] Performing differential processing on the periodic hotspot feature sequence according to the stable check result to obtain a stable hotspot feature sequence;

[0158] Performing autocorrelation time series fitting on the stationary hotspot feature sequence to obtain a fitting periodic model;

[0159] The fitting periodic model is used to perform periodic rolling analysis on the stationary hotspot feature sequence to obtain analytical hotspot features.

[0160] In detail, the ADF verification algorithm or the KPSS verification algorithm may be used to perform a stationarity check, and the stationarity verification result is a result used to indicate whether the periodic hotspot feature sequence is stationary, and the stationarity verification result includes stationary and non-stationary.

[0161] Specifically, performing differential processing on the periodic hotspot feature sequence according to the stationary check result means performing differential processing when the stationary check result is not stationary, and treating the periodic hotspot feature sequence as a stationary hotspot feature sequence when the stationary check result is stationary. The differential processing method may be a first differential or a second differential.

[0162] In detail, the autocorrelation function or the partial autocorrelation function may be used to perform autocorrelation time series fitting, and the period rolling analysis refers to performing rolling prediction using the fitting period model to obtain analysis hotspot features.

[0163] In the embodiment of the present invention, by performing time labeling and time series hot spot analysis, the feature dimension of the clothing hot spot feature set can be reduced, the efficiency of extracting time series features can be improved, and the changes in hot spot features in future time periods can be further analyzed, thereby improving the accuracy of the probability analysis of hot products.

[0164] S5. Perform kernel function mapping and linear regression analysis on the clothing style features and the analysis hot spot features to obtain the probability of a style becoming a hit.

[0165] In detail, the probability of a style becoming a hit refers to the probability that the sales volume of the clothing style to be analyzed exceeds a preset hit sales volume threshold in a future time period.

[0166] In the embodiment of the present invention, performing kernel function mapping and linear regression analysis on the clothing style features and the analysis hot spot features to obtain the probability of a style being a hit includes:

[0167] Generate a style hotspot feature according to the clothing style feature and the analysis hotspot feature;

[0168] Performing kernel function mapping on the style hotspot features to obtain mapped style hotspot features;

[0169] Linearly combine and bias-weight the mapped style hot spot features to obtain a hot item analysis result;

[0170] The hot-selling analysis results are regressed and outputted to obtain the probability of the style becoming a hot-selling item.

[0171] In detail, generating the style hotspot feature refers to combining the clothing style feature and the analysis hotspot feature into a style hotspot feature vector, and performing a standardization operation on the style hotspot feature vector to obtain the style hotspot feature.

[0172] Specifically, the kernel function of the support vector machine model that has been pre-trained with a large number of hot style features marked with the probability of a hit can be used for kernel function mapping, the support vector weights of the support vector machine model can be used for linear combination, and the bias item of the support vector machine model can be used for bias weighting, and normalization functions such as softmax or sigmoid can be used for regression output, wherein the kernel function can be a linear kernel function, a Gaussian kernel function, or a polynomial kernel function.

[0173] In the embodiment of the present invention, by performing kernel function mapping and linear regression analysis, the support vector machine algorithm can be used to perform a hit probability analysis on styles based on trend hot information and clothing styles in the future time period, thereby realizing the hit probability analysis of the sales cycle and improving the accuracy of the hit probability analysis.

[0174] By collecting clothing images and obtaining historical hot spot information, the present invention can combine the image information of styles and real-time hot events to extract fashion trends, thereby improving the accuracy of the probability analysis of hot items. By performing hierarchical feature extraction and style feature fusion operations, style features of different levels and scales can be extracted from clothing images, and important style features can be screened out for fusion, thereby enhancing the feature details of clothing style features, thereby improving the accuracy of the probability analysis of hot items.

[0175] By performing hot feature extraction and feature clustering, a hot feature set is obtained, and hot features related to clothing semantics in historical hot information can be extracted, and correlation separation and clustering dimension reduction for hot features can be achieved, reducing the data dimension of hot features, facilitating the extraction of subsequent time series features, and by performing time annotation and time series hot spot analysis, the feature dimension of clothing hot feature set can be reduced, the efficiency of extracting time series features can be improved, and the hot feature changes in future time periods can be further analyzed, thereby improving the accuracy of the probability analysis of explosive models, and by performing kernel function mapping and linear regression analysis, the algorithm of support vector machine can be used to perform the probability analysis of explosive models of styles according to the trend hot information and clothing styles in the future time period, thereby realizing the probability analysis of explosive models in the sales cycle and improving the accuracy of the probability analysis of explosive models. Therefore, the method for analyzing the probability of explosive models of clothing styles based on big data proposed by the present invention can solve the problem of low accuracy when analyzing the explosive models of clothing.

[0176] like Figure 4 , which is a functional module diagram of a clothing style hot-selling probability analysis device based on big data provided by an embodiment of the present invention.

[0177] The device 100 for analyzing the probability of a clothing style explosion based on big data of the present invention can be installed in an electronic device. According to the functions to be implemented, the device 100 for analyzing the probability of a clothing style explosion based on big data can include a data acquisition module 101, a style extraction module 102, a hot spot clustering module 103, a hot spot analysis module 104 and an explosion analysis module 105. The module of the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0178] In this embodiment, the functions of each module / unit are as follows:

[0179] The data acquisition module 101 is used to collect clothing images of clothing styles to be analyzed and obtain historical hot spot information corresponding to the clothing images;

[0180] The style extraction module 102 is used to perform hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features, wherein the hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features include: performing image standardization and multi-level scale splitting on the clothing image to obtain a local clothing image set; performing pixel enhancement operations on the local clothing image set using the following clothing pixel enhancement algorithm to obtain an enhanced clothing image set: in, The pixel coordinates in the enhanced clothing image set are The gray value of The pixel coordinates in the local clothing image set are The gray value of The pixel coordinates in the clothing pixel enhancement algorithm are The horizontal enhancement weight corresponding to the pixel is The pixel coordinates in the clothing pixel enhancement algorithm are The vertical enhancement weight corresponding to the pixel is is the preset enhancement factor, is the absolute value symbol, , is the image horizontal axis coordinate index, , is the image vertical axis coordinate index; a hierarchical clothing image set is generated according to the enhanced clothing image set and the clothing image; contour extraction and contour fitting are performed on the hierarchical clothing image set to obtain clothing contour features; morphological convolution and geometric feature extraction are performed on the hierarchical clothing image set to obtain clothing cutting features; pattern detection and texture extraction are performed on the hierarchical clothing image set to obtain clothing style features; feature dimension reduction screening and feature fusion are performed on the clothing contour features, the clothing cutting features and the clothing style features to obtain clothing style features;

[0181] The hot spot clustering module 103 is used to extract hot spot features and perform feature clustering on the historical hot spot information in sequence to obtain a clothing hot spot feature set;

[0182] The hot spot analysis module 104 is used to use the historical hot spot information to time-mark the clothing hot spot feature set to obtain a periodic hot spot feature set, and perform a temporal hot spot analysis on the periodic hot spot feature set to obtain an analysis hot spot feature;

[0183] The hot-selling analysis module 105 is used to perform kernel function mapping and linear regression analysis on the clothing style features and the analysis hot spot features to obtain the probability of a style being a hot-selling product.

[0184] In detail, each module described in the device 100 for analyzing the probability of a hit item of a clothing style based on big data in the embodiment of the present invention is used in the same manner as described above. Figure 1 The same technical means are used as the clothing style hit probability analysis method based on big data described in , and can produce the same technical effects, so I will not go into details here.

[0185] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0186] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0187] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0188] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0189] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0190] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application device that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0191] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices stated in the device embodiment can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0192] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for analyzing the probability of popular clothing styles based on big data, characterized in that: The method comprises: Collect clothing images of the clothing styles to be analyzed, and obtain historical hot spot information corresponding to the clothing images; Performing hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features, wherein the performing hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features includes: performing image standardization and multi-level scale splitting on the clothing image to obtain a local clothing image set; performing pixel enhancement operations on the local clothing image set using the following clothing pixel enhancement algorithm to obtain an enhanced clothing image set: Among them, Z (x,y) is the gray value of the pixel coordinate (x, y) in the enhanced clothing image set, G (i,j) is the gray value of the pixel coordinate (i, j) in the local clothing image set, V(i) is the horizontal enhancement weight corresponding to the pixel with the pixel coordinate (i, j) in the clothing pixel enhancement algorithm, V(j) is the vertical enhancement weight corresponding to the pixel with the pixel coordinate (i, j) in the clothing pixel enhancement algorithm, a is the preset enhancement coefficient, || is the absolute value symbol, x, i are the image horizontal axis coordinate indexes, y, j are the image vertical axis coordinate indexes; generate a hierarchical clothing image set according to the enhanced clothing image set and the clothing image; perform contour extraction and contour fitting on the hierarchical clothing image set to obtain clothing contour features; perform morphological convolution and geometric feature extraction on the hierarchical clothing image set to obtain clothing cutting features; perform pattern detection and texture extraction on the hierarchical clothing image set to obtain clothing style features; perform feature dimensionality reduction screening and feature fusion on the clothing contour features, the clothing cutting features and the clothing style features to obtain clothing style features; Extracting hotspot features and clustering the features of the historical hotspot information in turn to obtain a clothing hotspot feature set; The historical hot spot information is used to time-mark the clothing hot spot feature set to obtain a periodic hot spot feature set, and the periodic hot spot feature set is subjected to temporal hot spot analysis to obtain an analysis hot spot feature, wherein the historical hot spot information is used to time-mark the clothing hot spot feature set to obtain a periodic hot spot feature set, including: selecting the features in the clothing hot spot feature set one by one as target hot spot features, taking the clothing hot spot feature class corresponding to the target hot spot feature in the clothing hot spot feature set as the target feature class; taking the cluster center of the target feature class as the target center feature; filtering out the time information corresponding to the target hot spot feature from the historical hot spot information as the target time; and using the target time to time-mark the target center feature. Note, obtain periodic hot spot features; gather the periodic hot spot features of all target hot spot features in the clothing hot spot feature set into a target hot spot feature set, wherein the time series hot spot analysis of the periodic hot spot feature set to obtain analytical hot spot features includes: time series sorting of each periodic hot spot feature in the periodic hot spot feature set to obtain a periodic hot spot feature sequence; performing a stationarity check on the periodic hot spot feature sequence to obtain a stationary check result; performing differential processing on the periodic hot spot feature sequence according to the stationary check result to obtain a stationary hot spot feature sequence; performing autocorrelation time series fitting on the stationary hot spot feature sequence to obtain a fitting periodic model; performing period rolling analysis on the stationary hot spot feature sequence using the fitting periodic model to obtain analytical hot spot features; Kernel function mapping and linear regression analysis are performed on the clothing style features and the analysis hot spot features to obtain the probability of a style becoming a hit.

2. The method for analyzing the probability of popular clothing styles based on big data as claimed in claim 1, characterized in that: The step of performing image standardization and multi-level scale splitting on the clothing image to obtain a local clothing image set includes: Gray-scaling and filtering the clothing image to obtain a denoised clothing image; Performing image enhancement and brightness compensation on the denoised clothing image to obtain a compensated clothing image; Performing image convolution on the compensated clothing image to obtain clothing features; Performing a clothing key point recognition operation on the compensated clothing image according to the clothing features to obtain a clothing key point set; The clothing image is positioned and locally scaled according to the clothing key point set to obtain a local clothing image set.

3. The method for analyzing the probability of popular clothing styles based on big data as claimed in claim 1, characterized in that: The step of extracting and fitting the contour of the hierarchical clothing image set to obtain clothing contour features includes: Performing primary contour filtering on the hierarchical clothing image set to obtain a primary clothing contour set; The pixel change rate of each pixel in the hierarchical clothing image set is calculated using the following pixel change rate algorithm: Among them, L (x,y) is the pixel change rate of the pixel coordinate (x, y) in the hierarchical clothing image set, is the symbol of partial derivative, C (x,y) is the gray value of the pixel coordinate (x, y) in the hierarchical clothing image set, where x is the image horizontal axis coordinate index and y is the image vertical axis coordinate index; Performing contour pixel screening on the hierarchical clothing image set according to the pixel change rate to obtain a secondary clothing contour set; Generating a clothing outline set according to the primary clothing outline set and the secondary clothing outline set; Performing contour splitting on the clothing contour set to obtain a split clothing contour set; Performing polynomial contour fitting on the split garment contour set to obtain a fitted garment contour set; The fitted clothing contour set is subjected to contour splicing to obtain clothing contour features.

4. The method for analyzing the probability of popular clothing styles based on big data as claimed in claim 1, characterized in that: The step of performing feature dimension reduction screening and feature fusion on the clothing outline feature, the clothing cutting feature and the clothing style feature to obtain clothing style features includes: Performing feature standardization and feature matrix operations on the clothing outline features, the clothing cutting features, and the clothing style features to obtain a clothing feature matrix; Performing a covariance operation on the clothing feature matrix to obtain a clothing covariance feature matrix; Performing eigendecomposition on the clothing covariance feature matrix to obtain clothing eigenvalues ​​and clothing eigenvectors; Filtering out dimension-reduced clothing features from the clothing feature vector according to the clothing feature value; Performing a feature matrix operation on the reduced-dimensional clothing features to obtain a reduced-dimensional clothing feature matrix; The reduced-dimensional clothing feature matrix is ​​used to perform feature projection on the clothing feature matrix to obtain clothing style features.

5. The method for analyzing the probability of popular clothing styles based on big data as claimed in claim 1, characterized in that: The hot spot feature extraction and feature clustering are performed on the historical hot spot information in sequence to obtain a clothing hot spot feature set, including: Performing text segmentation and text vectorization operations on the historical hot spot information to obtain a hot spot text feature set; Performing position encoding on the hotspot text feature set to obtain a primary hotspot encoding set; Performing semantic encoding on the primary hotspot code set to obtain a secondary hotspot code set; Extracting semantic features from the secondary hotspot coding set to obtain hotspot semantic features; A feature clustering operation is performed on the hot semantic feature set to obtain a clothing hot cluster set.

6. The method for analyzing the probability of popular clothing styles based on big data as claimed in claim 5, characterized in that: The performing of a feature clustering operation on the hotspot semantic feature set to obtain a hotspot feature set includes: The hotspot semantic feature set is randomly divided into a plurality of hotspot semantic feature groups, and a primary hotspot center feature is randomly selected for each hotspot semantic feature group; The feature correlation distance between each hotspot semantic feature in the hotspot semantic feature set and each of the primary hotspot center features is calculated using the following correlation distance formula: Wherein, S refers to the feature association distance, p refers to the hotspot semantic feature, q refers to the primary hotspot center feature, T is the transposition symbol, cov() is the covariance symbol, cov(p,p) refers to the covariance of the hotspot semantic feature, cov(p,q) refers to the covariance between the hotspot semantic feature and the primary hotspot center feature, cov(q,p) refers to the covariance between the primary hotspot center feature and the hotspot semantic feature, and cov(q,q) refers to the covariance of the primary hotspot center feature; updating the hotspot semantic feature groups into standard hotspot semantic feature groups one by one according to the feature association distances; Calculate the standard hotspot central features of each of the standard hotspot semantic feature groups, and calculate the central feature association distances between the standard hotspot central features and the corresponding primary hotspot central features one by one; Each of the standard hotspot semantic feature groups is iteratively updated into a corresponding clothing hotspot feature class according to all center feature association distances, and all clothing hotspot feature classes are aggregated into a clothing hotspot feature class set.

7. The method for analyzing the probability of popular clothing styles based on big data as claimed in claim 1, characterized in that: The performing of kernel function mapping and linear regression analysis on the clothing style features and the analysis hot spot features to obtain the probability of a style becoming a hit includes: Generate a style hotspot feature according to the clothing style feature and the analysis hotspot feature; Performing kernel function mapping on the style hotspot features to obtain mapped style hotspot features; Linearly combine and bias-weight the mapped style hot spot features to obtain a hot item analysis result; The hot-selling analysis results are regressed and outputted to obtain the probability of the style becoming a hot-selling item.

8. A device for analyzing the probability of popular clothing styles based on big data, characterized in that: The device comprises: A data acquisition module, used to collect clothing images of clothing styles to be analyzed, and obtain historical hot spot information corresponding to the clothing images; The style extraction module is used to perform hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features, wherein the hierarchical feature extraction and style feature fusion operations on the clothing image to obtain clothing style features include: performing image standardization and multi-level scale splitting on the clothing image to obtain a local clothing image set; performing pixel enhancement operations on the local clothing image set using the following clothing pixel enhancement algorithm to obtain an enhanced clothing image set: Among them, Z (x,y) is the gray value of the pixel coordinate (x, y) in the enhanced clothing image set, G (i,j) is the gray value of the pixel coordinate (i, j) in the local clothing image set, V(i) is the horizontal enhancement weight corresponding to the pixel with the pixel coordinate (i, j) in the clothing pixel enhancement algorithm, V(j) is the vertical enhancement weight corresponding to the pixel with the pixel coordinate (i, j) in the clothing pixel enhancement algorithm, a is the preset enhancement coefficient, || is the absolute value symbol, x, i are the image horizontal axis coordinate indexes, y, j are the image vertical axis coordinate indexes; generate a hierarchical clothing image set according to the enhanced clothing image set and the clothing image; perform contour extraction and contour fitting on the hierarchical clothing image set to obtain clothing contour features; perform morphological convolution and geometric feature extraction on the hierarchical clothing image set to obtain clothing cutting features; perform pattern detection and texture extraction on the hierarchical clothing image set to obtain clothing style features; perform feature dimensionality reduction screening and feature fusion on the clothing contour features, the clothing cutting features and the clothing style features to obtain clothing style features; A hotspot clustering module is used to extract hotspot features and perform feature clustering on the historical hotspot information in turn to obtain a clothing hotspot feature set; The hot spot analysis module is used to use the historical hot spot information to time-mark the clothing hot spot feature set to obtain a periodic hot spot feature set, and perform temporal hot spot analysis on the periodic hot spot feature set to obtain an analysis hot spot feature, wherein the use of the historical hot spot information to time-mark the clothing hot spot feature set to obtain a periodic hot spot feature set includes: selecting the features in the clothing hot spot feature set one by one as target hot spot features, and taking the clothing hot spot feature class corresponding to the target hot spot feature in the clothing hot spot feature set as the target feature class; taking the cluster center of the target feature class as the target center feature; filtering out the time information corresponding to the target hot spot feature from the historical hot spot information as the target time; and using the target time to perform temporal hot spot analysis on the target center feature. Time annotation is performed to obtain periodic hot spot features; the periodic hot spot features of all target hot spot features in the clothing hot spot feature set are aggregated into a target hot spot feature set, wherein the periodic hot spot feature set is subjected to time series hot spot analysis to obtain analytical hot spot features, including: time series sorting of each periodic hot spot feature in the periodic hot spot feature set to obtain a periodic hot spot feature sequence; stationarity check of the periodic hot spot feature sequence to obtain a stationary check result; differential processing of the periodic hot spot feature sequence according to the stationary check result to obtain a stationary hot spot feature sequence; autocorrelation time series fitting of the stationary hot spot feature sequence to obtain a fitting periodic model; period rolling analysis of the stationary hot spot feature sequence using the fitting periodic model to obtain analytical hot spot features; The hot-selling analysis module is used to perform kernel function mapping and linear regression analysis on the clothing style features and the analysis hot spot features to obtain the probability of a style being a hot-selling product.

Images