A Pattern Drawing Material Recommendation Method Based on Deep Hashing Coding
Through the deep hash coding method, the feature vectors of the flower pattern material are extracted and clustered and matched frequency statistics are performed, and the problem of fewer applications of recommendation systems in the design field is solved, efficient and personalized flower pattern material recommendations are achieved, and designers' work efficiency and creative quality are improved.
Patent Information
- Application Number
- CN202111480789.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-06
- Publication Date
- 2025-06-13
- Estimated Expiration
- 2041-12-06
AI Technical Summary
In the prior art, information usage efficiency is low, resulting in information overload. Recommended systems are mainly used in the field of e-commerce and have fewer applications in the field of design.
The flower pattern drawing material recommendation method based on deep hash encoding is adopted, including deep hash encoding module, material clustering module, material matching statistics module and material matching recommendation module. The feature vector is extracted through deep learning neural network, and the k-means algorithm is used to cluster, and the material matching frequency is counted, and the Euclidean distance and matching times are recommended.
It effectively improves the design efficiency and quality of the stylist, provides creative inspiration, reduces the designer's workload through intelligent recommendations, and improves the accuracy and personalization of the recommendations.
Smart Images
Figure CN114185965B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical fields of neural networks, machine learning, and recommendation, and in particular to a method for recommending flower pattern drawing materials based on deep hash coding. Background Art
[0002] In recent years, with the development of neural networks and deep learning, the field of images has achieved great breakthroughs. Neural networks, especially convolutional neural networks and their variants, can deeply understand and extract the features of images, and apply the obtained feature vectors to various applications, such as object detection, image retrieval and other technologies; machine learning is a multi-disciplinary cross-discipline that has been widely applied and developed in the past few decades. Due to the emergence of neural networks and deep learning, machine learning has once again attracted a boom, and academic activities and industrial applications related to machine learning are unprecedentedly active.
[0003] The emergence and popularization of the Internet have brought a large amount of information to users, but at the same time, it has also led to a decrease in the use efficiency of information, resulting in information overload. Recommendation systems have emerged, mainly used to study user preferences, mainly in the field of e-commerce, and there are relatively few application recommendation systems in the design field. Summary of the Invention
[0004] The purpose of the present invention is to solve the problem that the existing technology reduces the use efficiency of information, resulting in information overload. Recommendation systems have emerged, mainly used to study user preferences, mainly in the field of e-commerce, and there are relatively few application recommendation systems in the design field. A method for recommending flower pattern drawing materials based on deep hash coding is proposed.
[0005] To achieve the above purpose, the present invention adopts the following technical solutions:
[0006] A method for recommending flower pattern drawing materials based on deep hash coding includes a deep hash coding module, a material clustering module, a material matching statistics module, and a material matching recommendation module. The material clustering module includes feature vectors obtained according to deep hash coding. The material clustering module performs clustering with the Euclidean distance as the metric and the k-means algorithm. The material matching statistics module includes grouping the existing materials by category. The material matching statistics module includes the existing drawing information and material grouping to count the matching frequencies between materials of various categories. The material recommendation module includes returning recommended drawing materials according to the counted matching frequencies and relative Euclidean distances.
[0007] Preferably, the hash coding module represents the target image as a string of 684-dimensional binary codes based on the features of each layer extracted by the deep learning neural network, and ensures that similar images have similar binary codes.
[0008] Preferably, based on the 684-dimensional vectors obtained by the hash encoding of each material, and using the Euclidean distance as a metric, the material clustering module clusters a large number of existing materials through the k-means machine learning algorithm. By continuously adjusting the hyperparameter k, the number of materials in each category is relatively uniform.
[0009] Preferably, according to the existing pattern design scheme, the statistical material matching statistical module statistically analyzes the frequency of the matching relationships between materials of various categories based on the obtained material category results, and stores the results in a two-dimensional matrix.
[0010] Preferably, based on the obtained clustering results, the material matching recommendation module classifies the materials selected by the designer user, and for other material categories with the number of times of matching with this category, based on the relative distance, calculates a weight with the number of matching times as the divisor, and returns the recommendation results sorted according to the weight.
[0011] A pattern drawing material recommendation method based on deep hash encoding. Step 1: Deep hash encoding module: 1 Training process: Input all the pictures in the training set into the deep neural network according to the category labels. The convolutional layer processes the feature maps layer by layer, and inputs the final feature map into the fully connected layer to generate the corresponding hash code. During the training process of the neural network, it tends to improve the similarity of the hash codes of pictures of the same type, so as to achieve the clustering effect; 2 Inference process: Input the picture into the network trained in Step 1 to generate the corresponding hash code for use by the subsequent modules.
[0012] Step 2: Material clustering module based on the k-means clustering algorithm: 1 Cluster and classify all materials with 648-dimensional vectors after passing through the deep hash encoding module. For the existing pattern drawing design, finally select the hyperparameter K; 2 Classify all materials through the clustering algorithm obtained by training in Step 1 to obtain their respective categories.
[0013] Step 3: Material matching statistical module: 1 Statistically analyze the material matches in pairs according to the existing pattern design cost design scheme in the database; 2 Based on the categories obtained by the clustering algorithm and the material matching statistics in Step 1, convert them into the matching frequencies between categories to obtain a two-dimensional matrix, recording the number of mutual matches between categories.
[0014] Step 4. Material Matching Recommendation Module: 1. For the materials selected by the designer, first obtain a specific category through the material clustering module; 2. For the specific category, calculate the spatial vector distance from the material to the clustering center of this category. At the same time, select several central points with the largest number of statistical matches in this category, and perform the same spatial vector translation to obtain the corresponding points; 3. Calculate the several vectors with the closest Euclidean distance from each material in this category to the points obtained in Step 2 as candidate recommended materials. The recommended quantity for this category is based on the number of matches between two categories, and the recommended material quantity is weighted and distributed according to the number of matches. Let the recommended quantity be C, and the number of matches for each category be c 1 , c 2 , c 3 ....c n , and the recommended quantity for each is 4. Here, sort according to the distance and the number of matches, and sort the materials to be recommended by weight and then return. Let the weight be weight, the number of matches between two categories be c, and the Euclidean distance of the points after distance calculation be d. weight = c / d, and return the recommended materials in order according to the size of weight.
[0015] Beneficial Effects:
[0016] 1. In the present invention, by using a large number of existing flower pattern design schemes, explore the matching relationship between their materials, so as to make corresponding recommendations for a specific material, provide creative inspiration for flower pattern designers, and improve their design efficiency and quality;
[0017] 2. Learn through the existing flower pattern design finished products of designers, extract the spatial distance relationship between flower pattern materials according to the existing material matching relationship, deep hash coding of materials and k-means clustering, and perform intelligent flower pattern drawing material recommendation, providing creative inspiration for designers, which is more time-saving and labor-saving, thus helping designers create more excellent and more flower patterns. Description of the Drawings
[0018] Figure 1 is a schematic structural diagram of the implementation step process of a flower pattern drawing material recommendation method based on deep hash coding proposed by the present invention;
[0019] Figure 2 is a schematic structural diagram of the material recommendation relationship of a flower pattern drawing material recommendation method based on deep hash coding proposed by the present invention;
[0020] Figure 3 is a schematic structural diagram of the training process of a flower pattern drawing material recommendation method based on deep hash coding proposed by the present invention. Detailed Embodiment
[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.
[0022] Referring to Figures 1-3 , a flower pattern drawing material recommendation method based on deep hash coding, including a deep hash coding module, a material clustering module, a material matching statistics module, and a material matching recommendation module. The material clustering module includes feature vectors obtained according to deep hash coding. The material clustering module uses the Euclidean distance as a metric and the k-means algorithm for clustering. The material matching statistics module includes grouping existing materials by category. The material matching statistics module includes existing drawing information and material grouping to count the matching frequencies between materials of various categories. The material recommendation module includes returning recommended drawing materials according to the counted matching frequencies and relative Euclidean distances.
[0023] In this embodiment, the hash coding module represents the target image as a string of 684-dimensional binary codes based on the features of each layer extracted by the deep learning neural network, and ensures that similar images have similar binary codes. The material clustering module is based on the 684-dimensional vectors obtained by the hash coding of each material, uses the Euclidean distance as a metric, and clusters a large number of existing materials through the k-means machine learning algorithm. By continuously adjusting the hyperparameter k, the number of materials in each category is relatively uniform. The statistical material matching statistics module counts the matching relationship frequencies between materials of various categories according to the obtained material category results based on the existing flower pattern design scheme, and stores the results in a two-dimensional matrix. The material matching recommendation module classifies the materials selected by the designer user according to the obtained clustering results, and for other material categories with the number of times of matching with this category, based on the relative distance, and calculates a weight with the number of times of matching as the divisor, and returns the recommendation result according to the weight ranking.
[0024] In this embodiment, the steps of the deep hash coding module are as follows: Step 1, the training process (as Figure 3 shown), input all the pictures in the training set into the deep neural network according to the category labels. The convolutional layer processes the feature maps layer by layer, and inputs the final feature map into the fully connected layer to generate the corresponding hash code. During the training process of the neural network, it tends to improve the similarity of the hash codes of pictures of the same type, so as to achieve the clustering effect. Step 2, the inference process, input the picture into the network trained in Step 1 to generate the corresponding hash code for use by the subsequent modules.
[0025] Further, the steps of the material clustering module based on the k-means clustering algorithm: Step 1: Cluster and classify all materials with 648-dimensional vectors passing through the deep hashing encoding module. For the existing flower pattern design, finally select the hyperparameter K; Step 2: Classify all materials through the clustering algorithm obtained by training in Step 1 to obtain their respective categories.
[0026] Further, the steps of the material matching statistics module: Step 1: According to the flower pattern design cost design scheme existing in the database, statistically pair up the material matches therein; Step 2: And convert the categories obtained by the clustering algorithm and the material match statistics in Step 1 into the matching frequencies between categories to obtain a two-dimensional matrix, which records the number of mutual matches between various categories.
[0027] Further, the steps of the material matching recommendation module: Step 1: For the materials selected by the designer, first obtain a specific category through the material clustering module; Step 2: For the specific category, calculate the spatial vector distance from the material to the clustering center of this category, and at the same time select several central points with the largest statistical matching quantity in this category, and perform the same spatial vector translation to obtain corresponding points; Step 3: Calculate the several vectors with the closest Euclidean distance from each material in this category to the points obtained in Step 2 as candidate recommended materials. The recommended quantity for this category is based on the number of matches between two categories. This solution is based on weighted distribution of the recommended material quantity according to the number of matches. Let the recommended quantity be C, and the number of matches for each category be c 1 、c 2 、c 3 ....c n , and the recommended quantity for each is Step 4: Here, sort according to the distance and the number of matches, and sort the materials to be recommended by weight and then return. Let the weight be weight, the number of matches between two categories be c, and the Euclidean distance of the point after distance calculation be d, weight = c / d. Return the recommended materials in order according to the size of weight. Figure 2 This is the description of this solution in two dimensions.
[0028] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its inventive concept, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.
Claims
1. A method for recommending flower pattern drawing materials based on deep hash coding, including a deep hash coding module, a material clustering module, a material matching statistics module, and a material matching recommendation module. It is characterized in that: The material clustering module includes feature vectors obtained according to deep hash coding. The material clustering module performs clustering with the Euclidean distance as the metric and the k-means algorithm. The material matching statistics module includes grouping existing materials by category. The material matching statistics module includes existing drawing information and material grouping to count the matching frequencies between materials of various categories. The material matching recommendation module includes returning recommended drawing materials according to the statistically obtained matching frequencies and relative Euclidean distances. The specific steps of this method for recommending flower pattern drawing materials are as follows: Step 1. Deep hash coding module: 1.1 Training process: Input all pictures in the training set into the deep neural network according to the category labels. The convolutional layer processes the feature maps layer by layer, and inputs the final feature map into the fully connected layer to generate corresponding hash codes. During the training process of the neural network, it tends to improve the similarity of the hash codes of pictures of the same type, thus achieving the clustering effect. 1.2 Inference process: Input the picture into the neural network trained in step 1.1 to generate the corresponding hash code for use by subsequent modules. Step 2. Material clustering module based on the k-means clustering algorithm: 2.1 Cluster and classify all materials with 648-dimensional vectors that have passed through the deep hash coding module. For the existing flower pattern design, finally select the hyperparameter k. 2.2 Classify all materials through the clustering algorithm with the hyperparameter k selected in step 2.1 to obtain their respective categories. Step 3. Material matching statistics module: 3.1 According to the existing flower pattern design cost design scheme in the database, statistically pair up the material matches. 3.2 According to the categories obtained by the clustering algorithm and the material match statistics in step 3.1, convert them into the matching frequencies between categories to obtain a two-dimensional matrix, recording the number of mutual matches between various categories. Step 4. Material matching recommendation module: 4.1 For the materials selected by the designer, first obtain a specific category through the material clustering module. 4.2 For the specific category, calculate the spatial vector distance from the material to the clustering center of this category. At the same time, select several central points with the largest number of statistical matches in this category, and perform the same spatial vector translation to obtain the corresponding points. 4.3 Calculate the several vectors closest to the points obtained in Step 4.2 in terms of Euclidean distance for each material in this category as candidate recommended materials. The recommended quantity for this category is based on the number of times of collocation between two categories, and the recommended material quantity is weighted and allocated according to the number of collocation times. Let the recommended quantity be C, and the number of collocation times for each category be c 1 , c 2 , c 3 ....c n , and each recommended quantity is 4.4 Here, sort according to the distance and the number of matches, and sort the materials to be recommended by weight and then return. Let the weight be weight, the number of matches between two categories be c, and the Euclidean distance of the point after distance calculation be d. weight = c / d. Return the recommended materials in order according to the size of weight.
2. A method for recommending flower pattern drawing materials based on deep hash coding according to claim 1. It is characterized in that: The material matching statistics module statistically calculates the matching relationship frequencies between materials of various categories according to the existing flower pattern design scheme and the obtained material category results, and stores the results in a two-dimensional matrix.
3. A method for recommending pattern drawing materials based on deep hash coding according to claim 2, characterized in that: The material matching recommendation module classifies the materials selected by the designer user according to the obtained clustering results, and for other material categories with the number of times of matching with this category, calculates a weight based on the relative distance and using the number of matching times as the divisor, and returns the recommendation result according to the weight sorting.
Citation Information
Patent Citations
Intelligent type setting method for flower patterns of textile printing and dyeing product
CN106569991A