CN114187495A - Garment fashion trend prediction method based on images - Google Patents
Garment fashion trend prediction method based on images Download PDFInfo
- Publication number
- CN114187495A CN114187495A CN202210127383.3A CN202210127383A CN114187495A CN 114187495 A CN114187495 A CN 114187495A CN 202210127383 A CN202210127383 A CN 202210127383A CN 114187495 A CN114187495 A CN 114187495A
- Authority
- CN
- China
- Prior art keywords
- image
- clothing
- foreground
- fashion trend
- trend prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000000605 extraction Methods 0.000 claims abstract description 34
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000011176 pooling Methods 0.000 claims description 48
- 230000004927 fusion Effects 0.000 claims description 32
- 230000006870 function Effects 0.000 claims description 30
- 230000011218 segmentation Effects 0.000 claims description 24
- 238000000354 decomposition reaction Methods 0.000 claims description 19
- 239000000203 mixture Substances 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 9
- 230000003044 adaptive effect Effects 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 4
- 101100001677 Emericella variicolor andL gene Proteins 0.000 claims description 3
- 230000004931 aggregating effect Effects 0.000 claims description 3
- 230000004069 differentiation Effects 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000007620 mathematical function Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 4
- 238000004364 calculation method Methods 0.000 abstract description 2
- 230000000694 effects Effects 0.000 abstract description 2
- 206010063385 Intellectualisation Diseases 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 241001580935 Aglossa pinguinalis Species 0.000 description 1
- 101100481876 Danio rerio pbk gene Proteins 0.000 description 1
- 101100481878 Mus musculus Pbk gene Proteins 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Human Resources & Organizations (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a garment fashion trend prediction method based on images, which comprises garment image data acquisition, image foreground extraction, garment image feature extraction and garment fashion trend prediction. Firstly, collecting a clothing image data set, and preprocessing clothing image data; then obtaining a foreground image, and extracting clothing features based on a multi-convolution kernel deep neural network; and finally, a garment fashion trend prediction method based on deep learning is adopted, and the garment image characteristics are used as the input of the model to obtain the current garment fashion trend. The method can greatly reduce the calculation cost and the system complexity, promote the intellectualization of fashion trend prediction in the fashion field, and improve the effect and the quality of fashion prediction.
Description
Technical Field
The invention belongs to the technical field of intelligent clothes, and particularly relates to a clothes fashion trend prediction method based on images.
Background
At present, in the field of online clothing, designers usually design new clothes through learning experience, each time a piece of clothes is designed, a great deal of time and energy are consumed, the designers cannot possibly design clothes of each required style, popular clothes in various areas in the future cannot be easily predicted, and a plurality of designers familiar with the areas are usually required to participate. Therefore, in the field of clothing, intelligent prediction of future development trends of clothing has a potential and huge application scene.
Chinese patent publication No. CN110705755A discloses "a garment popularity trend prediction method and apparatus based on deep learning", which collects popular garment pictures and information from e-commerce website over the year, performs feature extraction and integration, and outputs a garment ranking scheme with garment popularity topk according to model results, but the scheme is not accurate for garment popularity trend prediction and needs further optimization.
Disclosure of Invention
In view of the above defects or improvement needs of the prior art, the present invention provides an image-based garment popularity trend prediction method, which aims to predict garment popularity trend by a deep learning method by collecting garment pictures of current garment shopping websites on the internet, and is reliable and real-time.
To achieve the above object, according to one aspect of the present invention, there is provided an image-based clothing fashion trend prediction method, including the steps of:
step 1, firstly, a clothing image data set is collected, and the clothing image data is preprocessed;
step2, extracting the clothing image foreground by using a foreground extraction model based on image multi-scale decomposition;
step 3, extracting and fusing clothing image features based on the multi-convolution kernel deep convolution neural network to obtain a final clothing image feature map;
step4, constructing a clothing fashion trend prediction model, and taking the final clothing image characteristics as the input of the clothing fashion trend prediction model to obtain the current clothing fashion trend;
the clothing fashion trend prediction module comprises: an adaptive weighted pooling layer, a full link layer, and a softmax layer.
Further, in step 1, clothing images of all shopping websites are collected through a web crawler and a manual collection mode, wherein the shopping websites comprise amazon online shopping malls, tianmao malls, tabby and jingdong malls.
Further, the garment image data preprocessing comprises: and adjusting the size of the image by a bilinear interpolation method, and then carrying out image scale normalization and image standardization.
Further, the specific implementation manner of step2 is as follows;
step21, carrying out multi-scale decomposition on the image by using total variation to obtain a series of smooth images;
step22, representing the foreground color distribution of the given smooth image as a Gaussian mixture model, and optimizing the number of Gaussian functions of the Gaussian mixture model by using a histogram shape analysis method;
and Step23, designing an iteration termination condition according to the segmentation results of different smooth images, so that the foreground is extracted from the decomposition scale of the smooth images.
Further, the specific implementation manner of optimizing the gaussian mixture model by using a histogram shape analysis method in Step22 is as follows;
the gaussian distribution of the mth zone is expressed as follows,
in the formula, G represents a Gaussian function,μ m sum ΣmRespectively the mean vector and covariance matrix of the color distribution of the region,u(i) Representing smooth imagesuTo (1)iPixel value is taken when calculating; det is a mathematical function used for solving a determinant of a square matrix;
using each peak of histogram to represent brightness distribution of image region, smoothing with median filter to obtain a smoothed histogram
The histogram has 256 values, respectively
The method comprises the steps of dividing an image into N regions by utilizing the troughs of a histogram, and calculating the number of the regions and pixels in a foreground F and a background B by combining a segmentation curveu(i) The probability of belonging to the foreground or background is:
in the formula,
the result of the segmentation is represented, wherein,x n =1 represents the foreground of the image,x n =0 is for the background, and the background,L F andL B respectively represent pixels areu(i) The likelihood of the foreground and the background,ω F andω B parameters representing the foreground and the background respectively,n F which represents the number of foreground pixels,n B representing the number of background pixels; the optimized gaussian mixture model is expressed as:
wherein,U(x,w,u) As a parameterwLower partxFor smooth imageuEvaluating the segmentation result of (1);wrepresenting the front and background color distribution parameters.
Further, combining CrabCut and a color distribution model of a smooth image in Step23, converting foreground extraction into joint optimization of segmentation and decomposition scales, and extracting energy functional of foregroundx*Comprises the following steps:
in the formula,αandβrepresenting a weight; first itemM(u,u 0 ) Is a multi-scale decomposition of the image,u 0 which represents the original image or images of the original image,uis a smooth image; the second term is the foreground extraction of the smoothed image,S(x,w,u) Expressed as:
in the formula,U(x,w,u) As a parameterwLower partxFor smooth imageuEvaluating the segmentation result of (1);wrepresenting the color distribution parameters of the front and the background,
the result of the segmentation is represented, wherein,x n =1 represents the foreground of the image,x n =0 represents background;V(x, u) Defined as the penalty of placing the segmentation curve on the foreground boundary as follows:
in the formula,A i is the firstiA set of adjacent pixels of the one pixel,jis composed ofA i A pixel of (1);dis(∙) representing the Euclidean distance of the pixel pairs; [ ∙]Is an indicator function;γandβrepresenting a weight; u (∙) represents a smoothed image.
Further, the multi-convolution kernel deep convolution neural network comprises a multi-convolution kernel feature extraction module and a multi-convolution kernel feature fusion module, and the specific implementation mode is as follows;
(31) extracting clothing image features including style, color system and style by using a multi-convolution kernel feature extraction module;
firstly, performing convolution operation and activation function operation twice on an input image, extracting image features to generate a clothing image feature map with the dimension of 224 multiplied by 64, performing maximum pooling operation on the extracted feature image on the basis, and converting the feature map dimension to 112 multiplied by 64; then, performing convolution operation and activation function operation, performing maximum pooling operation on the feature maps extracted after corresponding operation, and generating a feature map with dimensions of 56 × 56 × 128 as the input of the multi-convolution kernel fusion module;
(32) fusing the extracted features by using a multi-convolution kernel feature fusion module to obtain a final clothing image feature map
The multi-convolution kernel feature fusion module comprises: intra-module feature information fusion and inter-module feature information fusion;
wherein, the intra-module feature information fusion comprises three branches, the three branches respectively use convolution kernel with the sizes of 3 multiplied by 3, 5 multiplied by 5 and 7 multiplied by 7 to further extract the features output by the multi-convolution kernel feature extraction module, the three branches use a parallel mode to extract the features, the inter-module feature information fusion respectively carries out 3 multiplied by 3 convolution operation after aggregating the three branches in the intra-module feature information fusion, then the three branches are input into the intra-module feature information fusion part to aggregate the clothing feature information again, finally the extracted feature information is fused and aggregated with the input feature information again through 1 multiplied by 1 convolution operation, thereby achieving the purpose of fusing the extracted feature information, in the above convolution operations, all convolution operations except the 1 × 1 convolution operation are followed by the Relu activation function.
Further, the specific processing procedure of the adaptive weighting pooling layer comprises;
inputting: feature map to be pooled, pooling window sizenLoss functionJLearning rateβ；
Step 41: for each pooling layer, selecting the number of importance parameters according to the size of the pooling window of the layer, whereinnCharacteristic valueα i Pooling window random initializationnAn importance parameterk i ，i=1,2,...,n；
Step 42: sorting the characteristic values in each pooling window from large to small to obtain
；
Step 43: performing softmax normalization on the initialized importance parameters to obtain weight parameters;
step 44: multiplying the weight parameter by the corresponding characteristic value in each pooling window, and accumulating to obtain a pooling result:
step 45: initialized weight parameterw i And continuously iterating and optimizing through gradient reduction along with the progress of back propagation in the training process until convergence:
wherein,α i in order to be a characteristic value of the image,k i the parameters are randomly initialized for the pooling window,w i is a weight parameter, z is a weight parameter multiplied by the corresponding characteristic value in each pooling window and then accumulated to obtain a pooling result,∂representing partial differentiation.
Further, the loss function of the whole garment fashion trend prediction model and the loss function in the self-adaptive weighting pooling layerJLikewise, a cross entropy loss function is employed:
wherein,xclothing image features representing input modelThe figure is a figure of merit,pandqrespectively representing the classification real value and the clothing classification predicted value of the clothing classification.
Further, the processing procedure of the softmax layer is as follows;
wherein,Z i is as followsiThe output value of each node, C, is the number of output nodes, i.e. the number of categories of the final classification result.
According to another aspect of the invention, an image-based garment fashion trend prediction system is provided, which comprises the following modules:
the clothing image data acquisition module is used for collecting a clothing image data set and preprocessing clothing image data;
the image foreground extraction module is used for extracting the clothing image foreground by using a foreground extraction model based on image multi-scale decomposition;
the clothing image feature extraction module is used for extracting and fusing clothing image features based on the multi-convolution kernel deep convolution neural network to obtain a final clothing image feature map;
the garment fashion trend prediction module is used for constructing a garment fashion trend prediction model, and the final garment image characteristics are used as the input of the garment fashion trend prediction model to obtain the current garment fashion trend;
the garment fashion trend prediction model comprises: an adaptive weighted pooling layer, a full link layer, and a softmax layer.
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
(1) the garment fashion trend prediction method based on the images, provided by the invention, is used for predicting the garment fashion trend by utilizing the garment images, wherein the garment fashion trend comprises garment style, color system and style, and the garment fashion trend is predicted by a deep learning method, so that the garment fashion trend prediction method is reliable and has real-time performance;
(2) compared with the prior art, the garment popularity trend prediction method based on the images can greatly reduce the calculation cost, reduce the system complexity and improve the popularity prediction effect and quality.
Drawings
FIG. 1 is a schematic flow chart of a system for predicting fashion trends of clothing based on images according to an embodiment of the present invention;
fig. 2 is a structural diagram of a deep convolutional neural network with multiple convolutional kernels according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Fig. 1 is a schematic flow diagram of a garment fashion trend prediction system based on an image according to an embodiment, and the system includes 4 parts, namely a garment image data acquisition module, an image foreground extraction module, a garment image feature extraction module, and a garment fashion trend prediction module, and the specific processing procedures of the modules are as follows:
the clothing image data acquisition module is used for collecting a clothing image data set and preprocessing clothing image data;
the image foreground extraction module is used for extracting the clothing image foreground by using a foreground extraction model based on image multi-scale decomposition;
the clothing image feature extraction module is used for extracting and fusing clothing image features based on the multi-convolution kernel deep convolution neural network to obtain a final clothing image feature map;
the garment fashion trend prediction module is used for constructing a garment fashion trend prediction model, and the final garment image characteristics are used as the input of the garment fashion trend prediction model to obtain the current garment fashion trend;
the garment fashion trend prediction model comprises: an adaptive weighted pooling layer, a full link layer, and a softmax layer.
Corresponding to the system, the invention also provides a garment fashion trend prediction method based on the image, which comprises the following steps:
(1) firstly, collecting a clothing image data set, and preprocessing clothing image data;
in this embodiment, the clothing images of the large shopping websites are collected by a web crawler and a manual collection method, and the clothing images are subjected to image scale normalization and image standardization, wherein the shopping websites include amazon-online shopping mall (amazon.com), makitar mall (tmall.com), Taobao.com and Jingdong mall (jd.com).
Wherein, clothing image data preprocessing includes: and adjusting the size of the image by a bilinear interpolation method, and then carrying out image scale normalization and image standardization. In an embodiment, the image size is 224 × 224 × 3.
(2) Extracting an image foreground for obtaining a foreground image;
and extracting the clothing image foreground by using a foreground extraction model based on image multi-scale decomposition.
Step 21: carrying out multi-scale decomposition on the image by using total variation to obtain a series of smooth images, wherein the decomposition protects the edges of the image, smoothes the texture and compresses the distribution range of the colors of the image region;
step 22: expressing the foreground color distribution of the given smooth image as a Gaussian mixture model, and optimizing the number of Gaussian functions of the Gaussian mixture model by using a histogram shape analysis method;
and for each smooth image, accurately modeling the color distribution of the image by adopting a histogram shape analysis method.
Wherein the histogram shape analysis method optimizes a gaussian mixture model. Assuming that the color distribution of each region in the smoothed image is compact, the color distribution of the region can be expressed as a gaussian function, taking the mth region as an example:
in the formula, G represents a Gaussian function,μ m sum ΣmRespectively the mean vector and covariance matrix of the color distribution of the region,u(i) Representing smooth imagesuTo (1)iPixel value is taken when calculating; det is a mathematical function used for solving a determinant of a square matrix;
using each peak of histogram to represent brightness distribution of image region, smoothing with median filter to obtain a smoothed histogram
The histogram has 256 values, respectively
The method comprises the steps of dividing an image into N regions by utilizing the troughs of a histogram, and calculating the number of the regions and pixels in a foreground F and a background B by combining a segmentation curveu(i) The probability of belonging to the foreground or background is:
in the formula,
the result of the segmentation is represented, wherein,x n =1 represents the foreground of the image,x n =0 is for the background, and the background,L F andL B respectively represent pixels areu(i) The likelihood of the foreground and the background,ω F andω B parameters representing the foreground and the background respectively,n F which represents the number of foreground pixels,n B representing the number of background pixels; the optimized gaussian mixture model is expressed as:
wherein,U(x,w,u) As a parameterwLower partxFor smooth imageuEvaluating the segmentation result of (1);wrepresenting the front and background color distribution parameters.
Step 23: and designing an iteration termination condition according to the segmentation results of different smooth images so as to extract the foreground from the decomposition scale of the smooth image.
And combining CrabCut and a color distribution model of the smooth image, and converting foreground extraction into joint optimization of segmentation and decomposition scale. An original image with N pixels
Divided by an initial rectangular frame into a background region B and a foreground region F with a small number of background pixels, RGB representing red green blue respectively,u R which represents a set of red pixels, is shown,u G a set of green pixels is represented as,u B representing a set of blue pixels that are to be addressed,
representing an original imageu 0 The set of all pixels in. Energy functional with its foreground extractedx*Comprises the following steps:
in the formula,αandβrepresenting a weight; first itemM(u,u 0 ) Is a multi-scale decomposition of the image,u 0 which represents the original image or images of the original image,uis a smooth image; the second term is the foreground extraction of the smoothed image,S(x,w,u) Expressed as:
in the formula,U(x,w,u) As a parameterwLower partxFor smooth imageuEvaluating the segmentation result of (1);wrepresenting the color distribution parameters of the front and the background,
the result of the segmentation is represented, wherein,x n =1 represents the foreground of the image,x n =0 represents background;V(x, u) Defined as the penalty of placing the segmentation curve on the foreground boundary as follows:
in the formula,A i is the firstiA set of adjacent pixels of the one pixel,jis composed ofA i A pixel of (1);dis(∙) representing the Euclidean distance of the pixel pairs; [ ∙]Is an indicator function;γandβrepresenting a weight; u (∙) represents a smoothed image.
In a particular embodiment of the present invention,γtaking 50, to ensure that the energy of the above formula is greater at low gradients and less at high gradients,
where 〈 ∙ 〉 represents the mean value.
(3) Clothing image feature extraction and fusion are carried out on the basis of a multi-convolution kernel deep convolution neural network to obtain a final clothing image feature map, wherein the multi-convolution kernel deep convolution neural network comprises a multi-convolution kernel feature extraction module and a multi-convolution kernel feature fusion module;
(31) and extracting the clothing image features including style, color system and style by using a multi-convolution kernel feature extraction module.
In a specific embodiment, firstly, the convolution operation and the activation function operation are performed twice on the input image, the image features are extracted, a clothing image feature map with the dimension of 224 × 224 × 64 is generated, and on the basis, the maximum pooling operation is performed on the extracted feature image, and the feature map dimension is converted into 112 × 112 × 64. Then, performing convolution operation and activation function operation, and performing maximum pooling operation on the feature maps extracted after corresponding operation to generate feature maps with dimensions of 56 × 56 × 128 as input of the multi-convolution kernel fusion module.
(32) And fusing the extracted features by using a multi-convolution kernel feature fusion module to obtain a final clothing image feature map.
As shown in fig. 2, it is a network structure diagram of a multi-convolution kernel feature fusion module provided in the embodiment, where the multi-convolution kernel feature fusion module includes: intra-module feature information fusion and inter-module feature information fusion.
Wherein, the intra-module feature information fusion comprises three branches, the three branches respectively use convolution kernel with the sizes of 3 multiplied by 3, 5 multiplied by 5 and 7 multiplied by 7 to further extract the features output by the multi-convolution kernel feature extraction module, the three branches use a parallel mode to extract the features, the inter-module feature information fusion respectively carries out 3 multiplied by 3 convolution operation after aggregating the three branches in the intra-module feature information fusion, then the three branches are input into the intra-module feature information fusion part to aggregate the clothing feature information again, finally the extracted feature information is fused and aggregated with the input feature information again through 1 multiplied by 1 convolution operation, thereby achieving the purpose of fusing the extracted feature information, in the above convolution operations, all convolution operations except the 1 × 1 convolution operation are followed by the Relu activation function.
(4) And constructing a garment fashion trend prediction model, and taking the final garment image characteristics as the input of the garment fashion trend prediction model to obtain the current garment fashion trend.
Wherein the clothing fashion trend prediction module comprises: an adaptive weighted pooling layer, a full link layer, and softmax.
The processing procedure of the self-adaptive weighting pooling layer specifically comprises the following steps:
inputting: feature map to be pooled, pooling window sizenLoss functionJLearning rateβ；
Step 41: for each pooling layer, selecting the number of importance parameters according to the size of the pooling window of the layer, whereinnCharacteristic valueα i Pooling window random initializationnAn importance parameterk i ，i=1,2,...,n；
Step42：Sorting the characteristic values in each pooling window from large to small to obtain
；
Step 43: performing softmax normalization on the initialized importance parameters to obtain weight parameters;
step 44: multiplying the weight parameter by the corresponding characteristic value in each pooling window, and accumulating to obtain a pooling result:
step 45: initialized weight parameterw i And continuously iterating and optimizing through gradient reduction along with the progress of back propagation in the training process until convergence:
wherein,α i in order to be a characteristic value of the image,k i the parameters are randomly initialized for the pooling window,w i is a weight parameter, z is a weight parameter multiplied by the corresponding characteristic value in each pooling window and then accumulated to obtain a pooling result,∂representing partial differentiation.
Wherein, the loss function of the whole garment fashion trend prediction model and the loss function in the self-adaptive weighting pooling layerJLikewise, a cross entropy loss function is employed:
wherein,xa clothing image feature map representing the input model,pandqclassification true value and clothing classification respectively representing clothing classificationAnd (4) predicting the class. The numerical value obtained by calculating the cross entropy loss function does not necessarily meet the condition and significance of probability distribution, so that the data is processed into a probability distribution form by finally activating the function by softmax, and the requirement of a multi-classification task of a clothing image algorithm is met.
Wherein,Z i is as followsiThe output value of each node, C, is the number of output nodes, i.e. the number of categories of the final classification result.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. An image-based garment fashion trend prediction method is characterized by comprising the following steps:
step 1, firstly, a clothing image data set is collected, and the clothing image data is preprocessed;
step2, extracting the clothing image foreground by using a foreground extraction model based on image multi-scale decomposition;
step 3, extracting and fusing clothing image features based on the multi-convolution kernel deep convolution neural network to obtain a final clothing image feature map;
step4, constructing a clothing fashion trend prediction model, and taking the final clothing image characteristics as the input of the clothing fashion trend prediction model to obtain the current clothing fashion trend;
the clothing fashion trend prediction module comprises: an adaptive weighted pooling layer, a full link layer, and a softmax layer.
2. The image-based garment fashion trend prediction method of claim 1, characterized in that: in the step 1, clothing images of all large shopping websites are collected through a web crawler and a manual collection mode, wherein the shopping websites comprise an Amazon-online shopping mall, a Temple mall, a Taobao net and a Jingdong mall;
the garment image data preprocessing comprises the following steps: and adjusting the size of the image by a bilinear interpolation method, and then carrying out image scale normalization and image standardization.
3. The image-based garment fashion trend prediction method of claim 1, characterized in that: the specific implementation manner of the step2 is as follows;
step21, carrying out multi-scale decomposition on the image by using total variation to obtain a series of smooth images;
step22, representing the foreground color distribution of the given smooth image as a Gaussian mixture model, and optimizing the number of Gaussian functions of the Gaussian mixture model by using a histogram shape analysis method;
and Step23, designing an iteration termination condition according to the segmentation results of different smooth images, so that the foreground is extracted from the decomposition scale of the smooth images.
4. The image-based garment fashion trend prediction method of claim 3, characterized in that: the concrete implementation mode of optimizing the Gaussian mixture model by using a histogram shape analysis method in Step22 is as follows;
the gaussian distribution of the mth zone is expressed as follows,
in the formula, G represents a Gaussian function,μ m sum ΣmRespectively the mean vector and covariance matrix of the color distribution of the region,u(i) Representing smooth imagesuTo (1)iPixel value is taken when calculating; det is a mathematical function used for solving a determinant of a square matrix;
using each peak of histogram to represent brightness distribution of image region, and smoothing by median filteringThe smoothed histogram is
The histogram has 256 values, respectively
The method comprises the steps of dividing an image into N regions by utilizing the troughs of a histogram, and calculating the number of the regions and pixels in a foreground F and a background B by combining a segmentation curveu(i) The probability of belonging to the foreground or background is:
in the formula,
the result of the segmentation is represented, wherein,x n =1 represents the foreground of the image,x n =0 is for the background, and the background,L F andL B respectively represent pixels areu(i) The likelihood of the foreground and the background,ω F andω B parameters representing the foreground and the background respectively,n F which represents the number of foreground pixels,n B representing the number of background pixels; the optimized gaussian mixture model is expressed as:
wherein,U(x,w,u) As a parameterwLower partxFor smooth imageuEvaluating the segmentation result of (1);wrepresenting the front and background color distribution parameters.
5. The image-based garment fashion trend prediction method of claim 4, characterized in that: color distribution model combining CrabCut and smooth image in Step23Type, converting foreground extraction into joint optimization of segmentation and decomposition scale, energy functional of foreground extractionx*Comprises the following steps:
in the formula,αandβrepresenting a weight; first itemM(u,u 0 ) Is a multi-scale decomposition of the image,u 0 which represents the original image or images of the original image,uis a smooth image; the second term is the foreground extraction of the smoothed image,S(x,w,u) Expressed as:
in the formula,U(x,w,u) As a parameterwLower partxFor smooth imageuEvaluating the segmentation result of (1);wrepresenting the color distribution parameters of the front and the background,
the result of the segmentation is represented, wherein,x n =1 represents the foreground of the image,x n =0 represents background;V(x,u) Defined as the penalty of placing the segmentation curve on the foreground boundary as follows:
in the formula,A i is the firstiA set of adjacent pixels of the one pixel,jis composed ofA i A pixel of (1);dis(∙) representing the Euclidean distance of the pixel pairs; [ ∙]Is an indicator function;γandβrepresenting a weight; u (∙) represents a smoothed image.
6. The image-based garment fashion trend prediction method of claim 1, characterized in that: the multi-convolution kernel deep convolution neural network comprises a multi-convolution kernel feature extraction module and a multi-convolution kernel feature fusion module, and the specific implementation mode is as follows;
(31) extracting clothing image features including style, color system and style by using a multi-convolution kernel feature extraction module;
firstly, performing convolution operation and activation function operation twice on an input image, extracting image features to generate a clothing image feature map with the dimension of 224 multiplied by 64, performing maximum pooling operation on the extracted feature image on the basis, and converting the feature map dimension to 112 multiplied by 64; then, performing convolution operation and activation function operation, performing maximum pooling operation on the feature maps extracted after corresponding operation, and generating a feature map with dimensions of 56 × 56 × 128 as the input of the multi-convolution kernel fusion module;
(32) fusing the extracted features by using a multi-convolution kernel feature fusion module to obtain a final clothing image feature map
The multi-convolution kernel feature fusion module comprises: intra-module feature information fusion and inter-module feature information fusion;
wherein, the intra-module feature information fusion comprises three branches, the three branches respectively use convolution kernel with the sizes of 3 multiplied by 3, 5 multiplied by 5 and 7 multiplied by 7 to further extract the features output by the multi-convolution kernel feature extraction module, the three branches use a parallel mode to extract the features, the inter-module feature information fusion respectively carries out 3 multiplied by 3 convolution operation after aggregating the three branches in the intra-module feature information fusion, then the three branches are input into the intra-module feature information fusion part to aggregate the clothing feature information again, finally the extracted feature information is fused and aggregated with the input feature information again through 1 multiplied by 1 convolution operation, thereby achieving the purpose of fusing the extracted feature information, in the above convolution operations, all convolution operations except the 1 × 1 convolution operation are followed by the Relu activation function.
7. The image-based garment fashion trend prediction method of claim 1, characterized in that: the specific processing procedure of the self-adaptive weighting pooling layer comprises the following steps of;
inputting: feature map to be pooled, pooling window sizenLoss functionJLearning rateβ；
Step 41: for each pooling layer, selecting the number of importance parameters according to the size of the pooling window of the layer, whereinnCharacteristic valueα i Pooling window random initializationnAn importance parameterk i ，i=1,2,...,n；
Step 42: sorting the characteristic values in each pooling window from large to small to obtain
；
Step 43: performing softmax normalization on the initialized importance parameters to obtain weight parameters;
step 44: multiplying the weight parameter by the corresponding characteristic value in each pooling window, and accumulating to obtain a pooling result:
step 45: initialized weight parameterw i And continuously iterating and optimizing through gradient reduction along with the progress of back propagation in the training process until convergence:
wherein,α i in order to be a characteristic value of the image,k i the parameters are randomly initialized for the pooling window,w i is a weight parameter, and z is a weight parameter corresponding to the feature in each pooling windowThe values are multiplied and then accumulated to obtain a pooling result,∂representing partial differentiation.
8. The image-based garment fashion trend prediction method of claim 7, characterized in that: loss function of whole garment fashion trend prediction model and loss function in adaptive weighted pooling layerJLikewise, a cross entropy loss function is employed:
wherein,xa clothing image feature map representing the input model,pandqrespectively representing the classification real value and the clothing classification predicted value of the clothing classification.
9. The image-based garment fashion trend prediction method of claim 1, characterized in that: the processing procedure of the softmax layer is as follows;
wherein,Z i is as followsiThe output value of each node, C, is the number of output nodes, i.e. the number of categories of the final classification result.
10. An image-based garment fashion trend prediction system is characterized by comprising the following modules:
the clothing image data acquisition module is used for collecting a clothing image data set and preprocessing clothing image data;
the image foreground extraction module is used for extracting the clothing image foreground by using a foreground extraction model based on image multi-scale decomposition;
the clothing image feature extraction module is used for extracting and fusing clothing image features based on the multi-convolution kernel deep convolution neural network to obtain a final clothing image feature map;
the garment fashion trend prediction module is used for constructing a garment fashion trend prediction model, and the final garment image characteristics are used as the input of the garment fashion trend prediction model to obtain the current garment fashion trend;
the garment fashion trend prediction model comprises: an adaptive weighted pooling layer, a full link layer, and a softmax layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210127383.3A CN114187495A (en) | 2022-02-11 | 2022-02-11 | Garment fashion trend prediction method based on images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210127383.3A CN114187495A (en) | 2022-02-11 | 2022-02-11 | Garment fashion trend prediction method based on images |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114187495A true CN114187495A (en) | 2022-03-15 |
Family
ID=80545831
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210127383.3A Pending CN114187495A (en) | 2022-02-11 | 2022-02-11 | Garment fashion trend prediction method based on images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114187495A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116188792A (en) * | 2023-02-23 | 2023-05-30 | 四川大学 | Quantitative analysis method and system for whole blood cell scatter diagram |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103718166A (en) * | 2011-08-02 | 2014-04-09 | 索尼公司 | Information processing apparatus, information processing method, and computer program product |
| CN108960499A (en) * | 2018-06-27 | 2018-12-07 | 东华大学 | A kind of Fashion trend predicting system merging vision and non-vision feature |
| CN110705755A (en) * | 2019-09-07 | 2020-01-17 | 创新奇智(广州)科技有限公司 | Garment fashion trend prediction method and device based on deep learning |
| CN112434210A (en) * | 2020-12-14 | 2021-03-02 | 武汉纺织大学 | Clothing fashion trend prediction system and method |
| CN112819510A (en) * | 2021-01-21 | 2021-05-18 | 江阴逐日信息科技有限公司 | Fashion trend prediction method, system and equipment based on clothing multi-attribute recognition |
| CN113159826A (en) * | 2020-12-28 | 2021-07-23 | 武汉纺织大学 | Garment fashion element prediction system and method based on deep learning |
-
2022
- 2022-02-11 CN CN202210127383.3A patent/CN114187495A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103718166A (en) * | 2011-08-02 | 2014-04-09 | 索尼公司 | Information processing apparatus, information processing method, and computer program product |
| CN108960499A (en) * | 2018-06-27 | 2018-12-07 | 东华大学 | A kind of Fashion trend predicting system merging vision and non-vision feature |
| CN110705755A (en) * | 2019-09-07 | 2020-01-17 | 创新奇智(广州)科技有限公司 | Garment fashion trend prediction method and device based on deep learning |
| CN112434210A (en) * | 2020-12-14 | 2021-03-02 | 武汉纺织大学 | Clothing fashion trend prediction system and method |
| CN113159826A (en) * | 2020-12-28 | 2021-07-23 | 武汉纺织大学 | Garment fashion element prediction system and method based on deep learning |
| CN112819510A (en) * | 2021-01-21 | 2021-05-18 | 江阴逐日信息科技有限公司 | Fashion trend prediction method, system and equipment based on clothing multi-attribute recognition |
Non-Patent Citations (3)
| Title |
|---|
| 王斌 等: ""基于图像多尺度分解的前景提取"", 《四川大学学报（自然科学版）》 * |
| 赵长乐: ""基于卷积神经网络的服装图像分类与去噪研究"", 《万方数据》 * |
| 龚柯: ""基于卷积神经网络的服装分类算法研究"", 《中国优秀硕士学位论文全文数据库 (工程科技Ⅰ辑)》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116188792A (en) * | 2023-02-23 | 2023-05-30 | 四川大学 | Quantitative analysis method and system for whole blood cell scatter diagram |
| CN116188792B (en) * | 2023-02-23 | 2023-10-20 | 四川大学 | Quantitative analysis method and system for whole blood cell scatter diagram |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111476713B (en) | Intelligent weather image identification method and system based on multi-depth convolution neural network fusion | |
| CN104182772B (en) | A kind of gesture identification method based on deep learning | |
| CN108573276A (en) | A kind of change detecting method based on high-resolution remote sensing image | |
| CN111899172A (en) | Vehicle target detection method oriented to remote sensing application scene | |
| CN107133943A (en) | A kind of visible detection method of stockbridge damper defects detection | |
| CN108510012A (en) | A kind of target rapid detection method based on Analysis On Multi-scale Features figure | |
| CN112396160A (en) | Transaction fraud detection method and system based on graph neural network | |
| CN109558806A (en) | The detection method and system of high score Remote Sensing Imagery Change | |
| CN110728179A (en) | Pig face identification method adopting multi-path convolutional neural network | |
| CN115249332B (en) | Hyperspectral image classification method and device based on space spectrum double-branch convolution network | |
| CN111178121B (en) | Pest image positioning and identifying method based on spatial feature and depth feature enhancement technology | |
| CN103390279A (en) | Target prospect collaborative segmentation method combining significant detection and discriminant study | |
| CN111639719A (en) | Footprint image retrieval method based on space-time motion and feature fusion | |
| CN107506792B (en) | Semi-supervised salient object detection method | |
| CN104063686A (en) | System and method for performing interactive diagnosis on crop leaf segment disease images | |
| CN113420794B (en) | Binaryzation Faster R-CNN citrus disease and pest identification method based on deep learning | |
| CN114863173B (en) | Self-mutual-attention hyperspectral image classification method for land resource audit | |
| CN114663769B (en) | Fruit identification method based on YOLO v5 | |
| CN117876890B (en) | Multi-source remote sensing image classification method based on multi-level feature fusion | |
| Li et al. | SGML: A symmetric graph metric learning framework for efficient hyperspectral image classification | |
| CN114187495A (en) | Garment fashion trend prediction method based on images | |
| CN112967296B (en) | Point cloud dynamic region graph convolution method, classification method and segmentation method | |
| CN104933410B (en) | A kind of high spectrum image spectral domain and spatial domain joint classification method | |
| CN117853596A (en) | Unmanned aerial vehicle remote sensing mapping method and system | |
| CN108765384B (en) | Significance detection method for joint manifold sequencing and improved convex hull |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220315 |
|
| RJ01 | Rejection of invention patent application after publication |