CN111797322A - Intelligent recommendation method for layered pattern design chart of textiles with affiliated style - Google Patents
Intelligent recommendation method for layered pattern design chart of textiles with affiliated style Download PDFInfo
- Publication number
- CN111797322A CN111797322A CN202010659465.3A CN202010659465A CN111797322A CN 111797322 A CN111797322 A CN 111797322A CN 202010659465 A CN202010659465 A CN 202010659465A CN 111797322 A CN111797322 A CN 111797322A
- Authority
- CN
- China
- Prior art keywords
- style
- client
- recommendation method
- pattern design
- intelligent recommendation
- 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 34
- 239000004753 textile Substances 0.000 title claims abstract description 19
- 238000013528 artificial neural network Methods 0.000 claims description 31
- 239000013598 vector Substances 0.000 claims description 13
- 239000000463 material Substances 0.000 claims description 5
- 230000000295 complement effect Effects 0.000 claims description 2
- 239000004744 fabric Substances 0.000 claims 2
- 238000013527 convolutional neural network Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 abstract description 9
- 238000004519 manufacturing process Methods 0.000 abstract description 2
- 239000010410 layer Substances 0.000 description 19
- 238000013135 deep learning Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 241001465754 Metazoa Species 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The invention discloses an intelligent recommendation method for a layered pattern design chart of textiles of the style, which comprises the following steps: image coding features → customer portrait → style features → KMedoid method clustering → recommended patterns. According to the intelligent recommendation method, the coding style of a client is obtained through a client figure, a pattern subset to be recommended which accords with the type is found through the type of the client style, finally, the style code of the client is used for finding the most suitable pattern diagram in the pattern subset to be recommended to recommend, layered textile patterns can be recommended for the client in an individualized mode, the requirements that the styles are similar but not too similar in the textile pattern recommendation are met, the recommendation speed is based on the mass pattern diagrams and the production is met, the situation that the patterns are similar in appearance and belong to the client style but are recommended to the client is avoided, and the recommendation speed is guaranteed.
Description
Technical Field
The invention relates to the technical field of neural network and image similarity recommendation, in particular to an intelligent recommendation method for a hierarchical pattern design chart of textiles of the same style.
Background
In recent years, deep learning has begun to be widely used in the direction of recommendation systems. Based on its advantages such as automatic feature selection, sequence modeling, flexibility and its non-linear transformation, many deep learning-based recommendation systems are proposed, the most popular method is based on multi-layer perceptron recommendation, such as neural collaborative filtering, and the traditional collaborative filtering method is extended by using neural network. The multi-layer perceptron is often used as an underlying network in more advanced networks, such as encoding and decoding networks. The encoder, decoder framework attempts to use partial user/item vectors to reconstruct the item-to-user relationship, either for ranking prediction or encoding into semantic features. In the application of the multilayer perceptron, a network with VGG and ResNet convolution is commonly used for image feature extraction. However, the prior art is not practiced in textile type patterns, and due to the particularity of the patterns: the patterns and patterns are complex, and the method has the problem that the existing network cannot well extract the features because the patterns and patterns are mixed in various fields such as animals and plants, cartoons, realistic photography and the like. In the invention, the network part adopts hierarchical input, adds an LSTM (long-short term memory) network as a variant of a deep learning middle-circulation neural network, has long-term memory capacity on a time sequence, and can better code three-layer images.
Because the number of the textile type patterns is large, each pattern is screened based on the customer style, and time and resources are consumed, the method uses KMedoid to perform unsupervised clustering on all the customer styles to optimize the process, and unsupervised learning is used because the customer styles have no labels.
Disclosure of Invention
The invention aims to provide an intelligent recommendation method for a layered pattern design chart of textiles with a style.
The intelligent recommendation method for the textile hierarchical pattern design chart with the affiliated style comprises the following steps of:
s1, based on the neural network 1, all the hierarchical psd graphs to be recommended in the library obtain respective image coding features.
S2, selecting the latest k layered flower type graphs from the purchase history of the customer to obtain a customer portrait P = { v1, v 2.
And S3, obtaining the client image P based on the encoder of the neural network 2 and S2, and generating style characteristics S corresponding to the client.
S4, repeating S3 on all customers to obtain style characteristics of all customers, and clustering the style characteristics by using a KMedoid method to obtain a classification set S; and then 2 indexes are generated, wherein the first is the category of each client style S in the classification set S, and the second is the style category in S matched with all the graphs to be recommended obtained through a neural network 2 decoder.
Preferably, the customer defaults to having a purchase history, and if the customer has no purchase history, the customer uses a popular style for random recommendation.
Preferably, vi is a 1x 128-dimensional image coding vector obtained by passing the ith layered input image through the neural network 1, and if i is less than k, the 1x 128-dimensional vector with the numerical value of 0 is used for complement.
Preferably, the hierarchical psd graph has the shading as one layer, all the materials are combined into one layer, the technique is one layer, and the three graphs are respectively subjected to a pre-trained VGG16 network.
Preferably, the input image feature dimension of the client is 1x128xk, and the input image feature dimension is 1x128x256 output by using 256 convolution kernels with a step size of 1x3 through two convolution layers.
Preferably, the method for clustering the style characteristics s of all customers by the KMedoid method is adopted, the initial centroid point is randomly selected and divided into n categories, and the centroids obtained by clustering are used as style labels.
Compared with the prior art, the invention has the beneficial effects that: according to the intelligent recommendation method, the coding style of a client is obtained through a client figure, a pattern subset to be recommended which accords with the type is found through the type of the client style, finally, the style code of the client is used for finding the most suitable pattern diagram in the pattern subset to be recommended to recommend, layered textile patterns can be recommended for the client in an individualized mode, the requirements that the styles are similar but not too similar in the textile pattern recommendation are met, the recommendation speed is based on the mass pattern diagrams and the production is met, the situation that the patterns are similar in appearance and belong to the client style but are recommended to the client is avoided, and the recommendation speed is guaranteed.
Drawings
FIG. 1 is a network structure diagram of a neural network 1 of the present invention;
FIG. 2 is a block diagram of an encoder of the neural network 2 of the present invention;
fig. 3 is a block diagram of a decoder of the neural network 2 of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The intelligent recommendation method for the textile hierarchical pattern design chart with the belonged style comprises two neural networks of CNN and LSTM, and comprises the following specific steps:
s1, based on the neural network 1, obtaining respective image coding characteristics of all hierarchical psd graphs to be recommended in the library;
(1) inputting a layered psd diagram, wherein the shading is used as a layer, and if the shading does not exist, the client can use a pure color diagram as a base; the adopted technical effect is one layer; all materials are combined into one layer. If the image input has only two layers, the technique does not exist, and the invention can supplement a pure white layer by default.
(2) The input hierarchical psd graph is changed from step 1 to three graphs: material, technique, shading. And (3) respectively obtaining corresponding material characteristics, technical characteristics and shading characteristics through the three graphs by virtue of a pre-trained VGG16 network.
(3) And (3) inputting the three features obtained in the step (2) into an encoder of the neural network 1, respectively, passing through 2 LSTM networks of 2 layers and one VGG16, and outputting a vector of 1x128 dimensions as a final image coding feature. The network of the VGG16 in the step (2) is consistent with the network structure of the network of the VGG16 in the step (2), and the models are trained respectively, so that the weights are different, the former is used for extracting complete pattern type information, the latter is used for extracting feature information of a single layer, and the second VGG16 network comprises 2 layers of long-term and short-term memory models.
(4) And circulating the steps, and obtaining respective image coding characteristics of all the layered psd graphs in the pattern library.
S2, selecting the latest k layers of pattern from the purchase history of the customer to obtain a customer portrait P = { v1, v 2.. and vn }, wherein vi is a 1x 128-dimensional image coding vector obtained by the ith layer of input image through a neural network 1, and if i is smaller than k, complementing the 1x 128-dimensional vector with the numerical value of 0;
(1) and applying the step 1 to each image to obtain k 1x128 image coding vectors, namely obtaining the 1x128xk dimension image vector of a single client.
S3, obtaining a client image P based on an encoder (figure 2) of the neural network 2 and in the step 2, and generating style characteristics S corresponding to the client;
(1) the characteristic dimension of the input image of the client is 1x128xk, 256 convolution kernels of 1x3 are used through two convolution layers, the step length is 1, and output of 1x128x256 dimensions is obtained.
(2) And (3) enabling the result of the step (1) to pass through a maximum pooling layer to obtain a vector with dimensions of 1x32x 256.
(3) And (3) enabling the result in the step (2) to pass through two full-connection networks to obtain a style vector with 1x128 dimensions.
S4, repeating S3 on all customers to obtain style characteristics of all customers, and clustering the style characteristics by using a KMedoid method to obtain a classification set S; secondly, generating two indexes, wherein the first is the category of each client style S in the classification set S, and the second is the style category in the S matched with all the graphs to be recommended obtained through a neural network 2 decoder;
(1) and clustering the style characteristics s of all the customers by a KMedoid method, randomly selecting initial centroid points, dividing the initial centroid points into n categories, and using the centroids obtained by clustering as style labels.
(2) And (3) obtaining style clusters from the step (1) and obtaining style category labels corresponding to the customers.
(3) Combining the image codes of all in-library hierarchical graphs obtained in the step (S1) with the style clustering centroids in the step (1) one by one, inputting the combined image codes into a decoder of the neural network 2, connecting the input style centroids S with the coding variable v of the input pattern to obtain a 1x 256-dimensional vector, outputting a value between 0 and 1 through a three-layer full-connection network by using softmax as an activation function, setting a threshold value t for the output, and judging that the style is matched with the pattern graph if the threshold value is greater than t, otherwise, not matching. All library images thus build an index of the matched styles.
S5, obtaining a pattern subset to be recommended by using the style code of the client and the style class corresponding to the style code, and finding the most suitable graph to recommend to the client based on the decoder of the neural network 2;
(1) and finding out the pattern graph to be recommended belonging to the style based on the style category of the client.
(2) And (3) combining the patterns to be recommended obtained in the step (1) with the client style codes one by one, recording the finally output numerical values of the network through a decoder of the neural network 2, and sequencing the numerical values, wherein the pattern graph with the high numerical value is preferentially recommended.
The working principle is as follows: the method is based on a neural network and a KMedoid machine learning model, wherein CNN (convolutional network) and LSTM (time-loop network) are involved, the neural network 1 carries out image coding on a hierarchical input graph, the neural network 2 consists of an encoder and a decoder, the encoder obtains client style coding based on client figures, the decoder judges whether the pattern graph should be recommended to a client based on the client coding obtained by the decoder and the image coding obtained by the neural network 1, the graph of the client is coded by the neural network 1 to form the client figures, the encoding style of the client is obtained by the neural network 2 encoder, the KMedoid is used for clustering the styles of all the clients, a subset of patterns to be recommended which accord with the category is found by the category of the styles, and finally, the most suitable graph is found in the subset of the patterns to be recommended by combining the neural network 2 decoder and the style coding of the client for recommendation, the following are important considerations: the customer does not need very same patterns but needs a series of pattern drawings with the same style, improves the timeliness of the patterns and the recommendation search speed in mass patterns, and meets the requirements of similar style and not too much shape in the recommendation of the patterns of textiles.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (6)
1. An intelligent recommendation method for a layered pattern design chart of textiles with a style comprises two neural networks of CNN and LSTM, and is characterized in that: the intelligent recommendation method for the layered pattern design chart of the textile class with the style comprises the following specific steps:
s1, based on the neural network 1, obtaining respective image coding characteristics of all hierarchical psd graphs to be recommended in the library;
s2, selecting the latest k layered flower pattern drawings from the purchase history of the customer to obtain a customer portrait P = { v1, v 2.., vn }; s3, obtaining a client image P based on the encoder of the neural network 2 and the S2, and generating style characteristics S corresponding to the client;
s4, repeating S3 on all customers to obtain style characteristics of all customers, and clustering the style characteristics by using a KMedoid method to obtain a classification set S; and then 2 indexes are generated, wherein the first is the category of each client style S in the classification set S, and the second is the style category in S matched with all the graphs to be recommended obtained through a neural network 2 decoder.
2. The intelligent recommendation method for the layered pattern design drawing of the textile goods with the styles as claimed in claim 1, wherein the customer defaults to having a purchase history, and if the customer has no purchase history, the customer uses a popular style for random recommendation.
3. The intelligent recommendation method for the hierarchical pattern design chart of the textile class in the style of claim 1, wherein vi is a 1x 128-dimensional image coding vector obtained by the input chart of the ith hierarchical layer through a neural network 1, and if i is less than k, the 1x 128-dimensional vector with the numerical value of 0 is used for complement.
4. The intelligent recommendation method for the layered pattern design drawing of the fabric class with the style as claimed in claim 1, wherein the layered psd drawing is a layer with the ground tint as one layer, all materials are combined into one layer, the technique is one layer, and the three drawings are respectively subjected to a pre-trained VGG16 network.
5. The intelligent recommendation method for the layered pattern design chart of the fabric products in the style of claim 1, wherein the characteristic dimension of the input image of the customer is 1x128xk, and the output of 1x128x256 dimension is obtained by passing through two convolution layers and using 256 convolution kernels of 1x3 and the step size is 1.
6. The intelligent recommendation method for the layered pattern design chart of the textile class with the style as claimed in claim 1, wherein the clustering is performed on the style features s of all customers by a KMedoid method, the initial centroid points are randomly selected and divided into n categories, and the clustered centroids are used as style labels.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010659465.3A CN111797322A (en) | 2020-07-10 | 2020-07-10 | Intelligent recommendation method for layered pattern design chart of textiles with affiliated style |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010659465.3A CN111797322A (en) | 2020-07-10 | 2020-07-10 | Intelligent recommendation method for layered pattern design chart of textiles with affiliated style |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111797322A true CN111797322A (en) | 2020-10-20 |
Family
ID=72810654
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010659465.3A Pending CN111797322A (en) | 2020-07-10 | 2020-07-10 | Intelligent recommendation method for layered pattern design chart of textiles with affiliated style |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111797322A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107423442A (en) * | 2017-08-07 | 2017-12-01 | 火烈鸟网络(广州)股份有限公司 | Method and system, storage medium and computer equipment are recommended in application based on user's portrait behavioural analysis |
| CN107862022A (en) * | 2017-10-31 | 2018-03-30 | 中国科学院自动化研究所 | Cultural resource commending system |
| CN109543840A (en) * | 2018-11-09 | 2019-03-29 | 北京理工大学 | A kind of Dynamic recommendation design method based on multidimensional classification intensified learning |
| CN111125528A (en) * | 2019-12-24 | 2020-05-08 | 三角兽（北京）科技有限公司 | Information recommendation method and device |
-
2020
- 2020-07-10 CN CN202010659465.3A patent/CN111797322A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107423442A (en) * | 2017-08-07 | 2017-12-01 | 火烈鸟网络(广州)股份有限公司 | Method and system, storage medium and computer equipment are recommended in application based on user's portrait behavioural analysis |
| CN107862022A (en) * | 2017-10-31 | 2018-03-30 | 中国科学院自动化研究所 | Cultural resource commending system |
| CN109543840A (en) * | 2018-11-09 | 2019-03-29 | 北京理工大学 | A kind of Dynamic recommendation design method based on multidimensional classification intensified learning |
| CN111125528A (en) * | 2019-12-24 | 2020-05-08 | 三角兽（北京）科技有限公司 | Information recommendation method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Kim et al. | Efficient facial expression recognition algorithm based on hierarchical deep neural network structure | |
| CN109829537B (en) | Deep learning GAN network children's garment based style transfer method and equipment | |
| Dai et al. | Pdan: Pyramid dilated attention network for action detection | |
| Xiang et al. | Fabric image retrieval system using hierarchical search based on deep convolutional neural network | |
| CN107194371B (en) | User concentration degree identification method and system based on hierarchical convolutional neural network | |
| CN104850890B (en) | Instance-based learning and the convolutional neural networks parameter regulation means of Sadowsky distributions | |
| CN106504064A (en) | Clothes classification based on depth convolutional neural networks recommends method and system with collocation | |
| KR101992986B1 (en) | A recommending learning methods of apparel materials using image retrieval | |
| CN111652049A (en) | Face image processing model training method and device, electronic equipment and storage medium | |
| CN110909754A (en) | Attribute generation countermeasure network and matching clothing generation method based on same | |
| CN110765960B (en) | Pedestrian re-identification method for adaptive multi-task deep learning | |
| Pérez-Delgado | The color quantization problem solved by swarm-based operations | |
| CN110674774A (en) | Improved deep learning facial expression recognition method and system | |
| KR20230082668A (en) | Computer Architecture for Creating Footwear Digital Assets | |
| Deverall et al. | Using generative adversarial networks to design shoes: the preliminary steps | |
| CN116012094A (en) | Depth map matching network and fashion collocation recommendation method based on same | |
| CN109508640A (en) | A kind of crowd's sentiment analysis method, apparatus and storage medium | |
| Fang et al. | Pedestrian attributes recognition in surveillance scenarios with hierarchical multi-task CNN models | |
| CN111797322A (en) | Intelligent recommendation method for layered pattern design chart of textiles with affiliated style | |
| Jia et al. | Sequential dual deep learning with shape and texture features for sketch recognition | |
| Yang | Clothing design style recommendation using decision tree algorithm combined with deep learning | |
| Soderlund et al. | Adversarial image generation using evolution and deep learning | |
| CN113763498B (en) | Industrial manufacturing-oriented portrait simple drawing region self-adaptive color matching method and system | |
| CN114565119A (en) | Fashion compatibility prediction method based on multi-mode graph convolution network | |
| Waltner et al. | Hibster: Hierarchical boosted deep metric learning for image retrieval |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201020 |