Intelligent cabinet control system based on big data and control method thereof

Abstract

The application relates to the field of intelligent cabinets, and particularly discloses an intelligent cabinet control system based on big data and a control method thereof. The intelligent cabinet control system uses a convolutional neural network model based on artificial intelligence technology to mine deep feature correlation information from type labels of clothes stored in each cabinet of the intelligent cabinet and personal basic data and personal preference data, to make personalized recommendations for people's dressing, and then open the corresponding cabinet of the intelligent cabinet based on the recommended result, to meet the needs of people.

Description

Technical Field

[0001] This invention relates to the field of smart cabinets, and more specifically, to a smart cabinet control system and control method based on big data. Background Technology

[0002] As society continues to develop and the pace of life becomes faster, people spend a lot of time on what to wear every day, and are still not satisfied with the final result. Therefore, if clothing recommendations could be made based on people's personal characteristics and preferences, it would not only save people a lot of time, but also make the clothing choices meet their needs.

[0003] Based on this, smart lockers with clothing recommendation functions have emerged. However, existing smart lockers can only recommend fixed clothing based on basic user data, lacking personalized and targeted recommendations. This results in clothing combinations that fail to meet user needs. Therefore, an optimized smart locker clothing recommendation system is desired. Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a smart locker control system and method based on big data. This system uses a convolutional neural network model based on artificial intelligence technology to mine deep-level feature correlation information from the type tags of clothing stored in each compartment of the smart locker, as well as personal basic data and personal preference data. This information is used to provide personalized recommendations for people's clothing choices, and then, based on the recommendations, the corresponding compartment of the smart locker is opened to meet people's needs.

[0005] According to one aspect of this application, a smart cabinet control system based on big data is provided, comprising:

[0006] The tag data acquisition module is used to obtain the type tags of the clothing stored in each compartment of the smart cabinet;

[0007] The tag data encoding module is used to perform one-hot encoding on the type tags of clothing stored in each compartment of the smart cabinet to obtain the tag vector corresponding to each compartment.

[0008] The outfit combination data construction module is used to arrange the four tag vectors corresponding to each of the four compartments of the smart cabinet, which are combinations of the type tags of tops, bottoms, underwear and shoes, in a two-dimensional arrangement to obtain multiple tag matrices;

[0009] The outfit combination data encoding module is used to pass each of the multiple label matrices through a first convolutional neural network, which acts as a feature extractor, to obtain multiple outfit label feature matrices.

[0010] The clustering module is used to cluster the multiple outfit tag feature matrices to obtain cluster centers;

[0011] An evaluation module is used to calculate the distance between each of the multiple outfit tag feature matrices and the cluster center to obtain a cluster feature vector;

[0012] The correction module is used to perform weighted correction on each label matrix in the plurality of label matrices based on the clustering feature vector to obtain a plurality of weighted outfit label feature matrices;

[0013] The personal data encoding module is used to pass the acquired personal basic data and personal preference data through a context encoder containing an embedding layer to obtain multiple personal clothing feature vectors, and then pass the multiple personal clothing feature vectors into a feature matrix through a second convolutional neural network as a feature extractor to obtain a clothing demand latent feature matrix.

[0014] The recalibration module is used to calibrate the implicit feature matrix of the dressing requirements and each weighted feature matrix of the multiple weighted dressing label feature matrices to obtain the calibrated implicit feature vector of dressing requirements and multiple calibrated weighted dressing label feature vectors.

[0015] The measurement module is used to calculate the distance between each of the corrected weighted outfit label feature vectors in the corrected outfit requirement latent feature vector and the multiple corrected weighted outfit label feature vectors to obtain multiple distance values.

[0016] The recommendation result determination module is used to select the four cabinets corresponding to the smallest distance value among the multiple distance values ​​as the outfit recommendation results; and

[0017] The control result generation module is used to activate the corresponding four compartments of the smart cabinet based on the outfit recommendation results.

[0018] According to another aspect of this application, a control method for a smart cabinet control system based on big data includes:

[0019] Obtain the type tags of the clothing stored in each compartment of the smart locker;

[0020] The type tags of the clothing stored in each compartment of the smart cabinet are uniquely encoded to obtain the tag vector corresponding to each compartment.

[0021] The four tag vectors corresponding to each of the four compartments of the smart cabinet, which are composed of the type tags of tops, bottoms, underwear and shoes, are arranged in two dimensions to obtain multiple tag matrices.

[0022] Each of the multiple label matrices is passed through a first convolutional neural network, which acts as a feature extractor, to obtain multiple outfit label feature matrices.

[0023] Cluster the multiple outfit tag feature matrices to obtain cluster centers;

[0024] Calculate the distance between each of the multiple outfit tag feature matrices and the cluster center to obtain the cluster feature vector;

[0025] Based on the clustering feature vector, each label matrix in the plurality of label matrices is weighted and corrected to obtain a plurality of weighted outfit label feature matrices;

[0026] The acquired basic personal data and personal preference data are passed through a context encoder containing an embedding layer to obtain multiple personal outfit feature vectors. The multiple personal outfit feature vectors are arranged into a feature matrix and then passed through a second convolutional neural network as a feature extractor to obtain an outfit requirement latent feature matrix.

[0027] The implicit feature matrix of dressing requirements and each weighted feature matrix of the multiple weighted dressing label feature matrices are respectively corrected to obtain the corrected implicit feature vector of dressing requirements and multiple corrected weighted dressing label feature vectors.

[0028] Calculate the distance between each of the corrected weighted outfit label feature vectors in the corrected outfit requirement latent feature vector and the multiple corrected weighted outfit label feature vectors to obtain multiple distance values;

[0029] The four cabinets corresponding to the smallest of the multiple distance values ​​are used as the outfit recommendation results; and

[0030] Based on the outfit recommendations, the corresponding four compartments of the smart cabinet are activated.

[0031] Compared with existing technologies, the smart locker control system and control method based on big data provided in this application use a convolutional neural network model based on artificial intelligence technology to mine deep feature correlation information from the type tags of clothing stored in each compartment of the smart locker, as well as personal basic data and personal preference data, in order to make personalized recommendations for people's clothing. Based on the recommendation results, the corresponding compartment of the smart locker is opened to meet people's needs. Attached Figure Description

[0032] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0033] Figure 1 This is an application scenario diagram of a big data-based intelligent cabinet control system according to an embodiment of this application.

[0034] Figure 2 This is a block diagram of a big data-based intelligent cabinet control system according to an embodiment of this application.

[0035] Figure 3 This is a block diagram of the calibration module in a big data-based intelligent cabinet control system according to an embodiment of this application.

[0036] Figure 4 This is a flowchart of a control method for a big data-based intelligent cabinet control system according to an embodiment of this application.

[0037] Figure 5 This is a schematic diagram of the architecture of the control method of the big data-based intelligent cabinet control system according to an embodiment of this application. Detailed Implementation

[0038] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0039] Scene Overview

[0040] As mentioned earlier, with the continuous development of society, the pace of life is getting faster and faster. However, people spend a lot of time on clothing every day, and are still not satisfied in the end. Therefore, if clothing recommendations can be made based on people's personal characteristics and preferences, it can not only save people a lot of time, but also make the clothing meet people's needs.

[0041] Based on this, smart lockers with clothing recommendation functions have emerged. However, existing smart lockers can only recommend fixed clothing based on basic user data, lacking personalized and targeted recommendations. This results in clothing combinations that fail to meet user needs. Therefore, an optimized smart locker clothing recommendation system is desired.

[0042] Accordingly, considering that intelligent and targeted recommendations based on people's personal characteristics are possible, it is necessary to identify the tag features of the clothing in the smart cabinet. The smart cabinet in this application has multiple compartments, each storing different types of clothing. For example, the first compartment holds jeans, the second holds casual shirts, and the third holds polo shirts, etc. Furthermore, feature mining is performed on people's basic information (e.g., height, weight, 3D measurements, and photos) and needs information (e.g., clothing style, preferences) to provide intelligent recommendations. The recommendation scheme is a combination of cabinet doors.

[0043] Specifically, in the technical solution of this application, the first step is to obtain the type tags of the clothing stored in each compartment of the smart cabinet. It should be understood that since the type tags of the clothing stored in each compartment of the smart cabinet are text data, and the data processed by the classifier by default is continuous and ordered, it cannot be used for subsequent classifier processing. Therefore, it is necessary to perform one-hot encoding on the type tags of the clothing stored in each compartment of the smart cabinet to obtain the tag vector corresponding to each compartment. It should be understood that the one-hot encoding, also known as effective encoding, uses an N-bit register to encode N states, each state having its own independent register bit, and at any given time, only one bit is valid.

[0044] Then, to recommend outfits that better meet people's needs, it is necessary to consider four different aspects of outfits: tops, bottoms, underwear, and shoes. When making recommendations for these four aspects, it is important to consider the implicit correlations between different combinations, such as different preferred color and style combinations. Therefore, the four label vectors corresponding to each of the four compartments of the smart cabinet (tops, bottoms, underwear, and shoes) are arranged in a two-dimensional pattern to obtain multiple label matrices. These multiple label matrices are then input into a first convolutional neural network model, which excels in identifying implicit correlations, to perform feature mining. This extracts the implicit correlation information of different combinations within each label matrix, thereby obtaining multiple outfit label feature matrices.

[0045] Considering that different recommendation schemes have different types within the multiple tag matrices, to differentiate these schemes and more accurately recommend outfits based on user preferences, the technical solution of this application further clusters the multiple outfit tag feature matrices to obtain cluster centers. Accordingly, in a specific example, the positional mean of the multiple outfit tag feature matrices can be calculated to obtain the cluster centers. Then, the distance between each outfit tag feature matrix and the cluster centers is calculated to obtain cluster feature vectors. In a specific example, the cosine distance between each outfit tag feature matrix and the cluster centers can be calculated to obtain the cluster feature vectors. Thus, the multiple tag matrices can be corrected; that is, based on the cluster feature vectors, each tag matrix in the multiple tag matrices is weighted and corrected to obtain multiple weighted outfit tag feature matrices.

[0046] Furthermore, in the technical solution of this application, in order to make accurate intelligent recommendations based on an individual's basic information and preferences, it is also necessary to obtain basic personal data (e.g., height, weight, 3D measurements, and photos) and personal preference data (e.g., clothing style, preferences, etc.). Then, considering that the basic personal data and personal preference data have correlated feature information, a context encoder with an embedding layer is used to encode the basic personal data and personal preference data to extract global high-dimensional semantic features that are more suitable for representing the essential characteristics of an individual's clothing, thereby obtaining multiple personal clothing feature vectors. These multiple personal clothing feature vectors are then arranged into a feature matrix to integrate the implicit essential characteristics of an individual's clothing, and then passed through a second convolutional neural network as a feature extractor to obtain a clothing requirement implicit feature matrix.

[0047] It should be understood that the implicit feature matrix for the aforementioned dressing requirements, for example denoted as M1, and multiple weighted dressing label feature matrices, for example denoted as M... 2t Since its row vectors correspond to the semantic directions of the label vectors and feature vectors, while the column directions correspond to the arrangement direction of the vectors, if we simply calculate M1 and M... 2t The distance between them means that the intra-sample distribution of eigenvalues ​​at various positions of the matrix in the semantic direction will affect the accuracy of the distance representation of the inter-sample feature distribution. Therefore, in the technical solution of this application, it is preferable that for matrices M1 and M... 2t Explicit generalization of semantic reasoning information is represented as follows:

[0048]

[0049] in, This represents the eigenvalues ​​at each position in the implicit feature matrix M1 representing the clothing requirements. M represents the multiple weighted outfit label feature matrices. 2t The feature values ​​at each position in the t-th weighted outfit label feature matrix, where V1 represents the latent feature vector of the corrected outfit requirement, V 2t Let represent the t-th corrected weighted outfit label feature vector among the plurality of corrected weighted outfit label feature vectors, and Σ j (·) represents summing the eigenvalues ​​of each row of a matrix to obtain a column vector. This indicates differentiation based on position. In this way, we can obtain the corrected latent feature vector of outfit requirements and multiple corrected weighted feature vectors of outfit labels.

[0050] Here, the explicit generalization of semantic reasoning information decouples the semantic groups between samples based on column direction by explicitly generalizing the semantic concepts corresponding to the feature values ​​along the column. In this way, the information plasticity of the high-dimensional manifold corresponding to the feature in the high-dimensional semantic space along the sample direction is obtained through information reasoning of the feature semantics, thereby improving the calculation accuracy of the spatial distance between matrices at the semantic expression level.

[0051] Then, the distances between the corrected outfit requirement latent feature vectors and the multiple corrected weighted outfit label feature vectors are calculated separately, for example, using Euclidean distance, to obtain multiple distance values. Calculating the Euclidean distance here can better highlight the consistency between the outfit requirement and the outfit label. Subsequently, the four cabinet compartments corresponding to the smallest of the multiple distance values ​​can be used as the outfit recommendation results, and based on the outfit recommendation results, the corresponding four cabinet compartments of the smart cabinet are activated.

[0052] Based on this, this application proposes a smart locker control system based on big data, comprising: a tag data acquisition module for acquiring the type tags of clothing stored in each compartment of the smart locker; a tag data encoding module for performing one-hot encoding on the type tags of clothing stored in each compartment of the smart locker to obtain tag vectors corresponding to each compartment; a combination data construction module for arranging the four tag vectors corresponding to each of the four compartments of the smart locker, where the type tags are combinations of tops, bottoms, underwear, and shoes, in a two-dimensional arrangement to obtain multiple tag matrices; a combination data encoding module for passing each tag matrix in the multiple tag matrices through a first convolutional neural network as a feature extractor to obtain multiple combination tag feature matrices; a clustering module for clustering the multiple combination tag feature matrices to obtain cluster centers; an evaluation module for calculating the distance between each combination tag feature matrix in the multiple combination tag feature matrices and the cluster centers to obtain cluster feature vectors; and a correction module for correcting each combination tag matrix in the multiple tag matrices based on the cluster feature vectors. The system employs a weighted correction module to obtain multiple weighted outfit label feature matrices; a personal data encoding module to process acquired personal basic data and personal preference data through a context encoder containing an embedding layer to obtain multiple personal outfit feature vectors, and then arranges these multiple personal outfit feature vectors into a feature matrix, which is then processed by a second convolutional neural network as a feature extractor to obtain an outfit requirement latent feature matrix; a recorrection module to correct each weighted outfit label feature matrix in the outfit requirement latent feature matrix and the multiple weighted outfit label feature matrices to obtain corrected outfit requirement latent feature vectors and multiple corrected weighted outfit label feature vectors; a measurement module to calculate the distances between each corrected weighted outfit label feature vector in the corrected outfit requirement latent feature vector and the multiple corrected weighted outfit label feature vectors to obtain multiple distance values; a recommendation result determination module to select the four cabinets corresponding to the smallest of the multiple distance values ​​as outfit recommendation results; and a control result generation module to open the corresponding four cabinets of the smart cabinet based on the outfit recommendation results.

[0053] Figure 1 The illustration shows an application scenario of a big data-based intelligent cabinet control system according to an embodiment of this application. For example... Figure 1 As shown, in this application scenario, firstly, a tag-based identification sensor (e.g., such as...) Figure 1 The label detector T shown in the image obtains the smart cabinet (e.g., such as...). Figure 1 The clothing stored in each compartment of cabinet B as shown (e.g., ... Figure 1 The type label of H shown, and from the smartphone (e.g., such as) connected to the smart cabinet. Figure 1The system retrieves basic personal data and personal preference data from the P shown in the diagram. Then, it inputs the type tags of the clothing stored in each compartment, along with the basic personal data and personal preference data, into a server deployed with a big data-based smart locker control algorithm (e.g., such as...). Figure 1 The cloud server S shown in the diagram is capable of processing the type tags of clothing stored in each cabinet, as well as the individual's basic data and personal preference data, using a big data-based smart cabinet control algorithm to generate outfit recommendations. Then, based on the outfit recommendations, the corresponding four cabinets of the smart cabinet—namely, the cabinets for tops, bottoms, underwear, and shoes—are activated.

[0054] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0055] Exemplary System

[0056] Figure 2 The diagram illustrates a block diagram of a big data-based intelligent cabinet control system according to an embodiment of this application. Figure 2As shown, the smart locker control system 200 based on big data according to an embodiment of this application includes: a tag data acquisition module 210, used to acquire the type tags of clothing stored in each compartment of the smart locker; a tag data encoding module 220, used to perform one-hot encoding on the type tags of clothing stored in each compartment of the smart locker to obtain tag vectors corresponding to each compartment; and a combination data construction module 230, used to arrange the four tag vectors corresponding to each of the four compartments of the smart locker, where the type tags are combined into tops, bottoms, underwear, and shoes, in a two-dimensional arrangement to obtain... The system comprises: a label matrix; an outfit combination data encoding module 240, used to pass each label matrix in the multiple label matrices through a first convolutional neural network as a feature extractor to obtain multiple outfit label feature matrices; a clustering module 250, used to cluster the multiple outfit label feature matrices to obtain cluster centers; an evaluation module 260, used to calculate the distance between each outfit label feature matrix in the multiple outfit label feature matrices and the cluster centers to obtain cluster feature vectors; and a correction module 270, used to correct each of the multiple label matrices based on the cluster feature vectors. The label matrix is ​​weighted and corrected to obtain multiple weighted outfit label feature matrices; the personal data encoding module 280 is used to pass the acquired personal basic data and personal preference data through a context encoder containing an embedding layer to obtain multiple personal outfit feature vectors, and then arrange the multiple personal outfit feature vectors into a feature matrix and pass it through a second convolutional neural network as a feature extractor to obtain an outfit requirement latent feature matrix; the recorrection module 290 is used to respectively correct each weighted outfit label feature matrix in the outfit requirement latent feature matrix and the multiple weighted outfit label feature matrices. The system performs line correction to obtain a corrected implicit feature vector of clothing requirements and multiple corrected weighted feature vectors of clothing labels; a measurement module 300 is used to calculate the distance between each of the corrected weighted feature vectors of clothing requirements and multiple corrected weighted feature vectors of clothing labels to obtain multiple distance values; a recommendation result determination module 310 is used to take the four cabinets corresponding to the smallest of the multiple distance values ​​as the clothing recommendation result; and a control result generation module 320 is used to open the corresponding four cabinets of the smart cabinet based on the clothing recommendation result.

[0057] Specifically, in this embodiment, the tag data acquisition module 210 and the tag data encoding module 220 are used to acquire the type tags of clothing stored in each compartment of the smart cabinet, and perform one-hot encoding on the type tags of clothing stored in each compartment of the smart cabinet to obtain a tag vector corresponding to each compartment. As mentioned above, to make intelligent targeted recommendations based on people's personal characteristics, it is necessary to identify the tag features of the clothing in the smart cabinet. In the technical solution of this application, the smart cabinet has multiple compartments, and different compartments store different types of clothing. For example, the first compartment stores jeans, the second compartment stores casual shirts, and the third compartment stores polo shirts, etc. Furthermore, feature mining is also performed on people's basic information (e.g., height, weight, 3D and photos, etc.) and demand information (e.g., clothing style, preferences, etc.) to make intelligent recommendations. The recommendation scheme is a combination of cabinet doors.

[0058] Specifically, in the technical solution of this application, the first step is to obtain the type tags of the clothing stored in each compartment of the smart cabinet. It should be understood that since the type tags of the clothing stored in each compartment of the smart cabinet are text data, and the data processed by the classifier by default is continuous and ordered, it cannot be used for subsequent classifier processing. Therefore, it is necessary to perform one-hot encoding on the type tags of the clothing stored in each compartment of the smart cabinet to obtain the tag vector corresponding to each compartment. It should be understood that the one-hot encoding, also known as effective encoding, uses an N-bit register to encode N states, each state having its own independent register bit, and at any given time, only one bit is valid.

[0059] Specifically, in this embodiment, the outfit combination data construction module 230 and the outfit combination data encoding module 240 are used to arrange the four label vectors corresponding to each of the four compartments of the smart cabinet, where the type label combination is tops, bottoms, underwear, and shoes, in a two-dimensional arrangement to obtain multiple label matrices. Each of the multiple label matrices is then passed through a first convolutional neural network, which acts as a feature extractor, to obtain multiple outfit label feature matrices. That is, in the technical solution of this application, to recommend outfits that better meet people's needs, it is necessary to consider four different aspects of outfits: tops, bottoms, underwear, and shoes. When recommending outfits in these four aspects, it is considered that there are implicit correlations between different combinations, such as different color combinations and style combinations that people prefer. Therefore, in the summary of the technical solution of this application, the four label vectors corresponding to each of the four compartments of the smart cabinet, where the type label combination is tops, bottoms, underwear, and shoes, are arranged in a two-dimensional arrangement to obtain multiple label matrices. The multiple label matrices are then input into a first convolutional neural network model that performs well in terms of implicit association features, which is used as a feature extractor, to perform feature mining, so as to extract the implicit association feature information of different combinations in each of the multiple label matrices, thereby obtaining multiple outfit label feature matrices.

[0060] More specifically, in this embodiment of the application, the outfit combination data encoding module is further configured to: each layer of the first convolutional neural network performs the following during the forward propagation of the layer: convolution processing on the input data to obtain a convolutional feature map; mean pooling based on the local channel dimension on the convolutional feature map to obtain a pooled feature map; and non-linear activation on the pooled feature map to obtain an activation feature map; wherein, the output of the last layer of the first convolutional neural network is the plurality of outfit label feature matrices, and the input of the first layer of the first convolutional neural network is each of the label matrices in the plurality of label matrices.

[0061] Specifically, in this embodiment, the clustering module 250, the evaluation module 260, and the correction module 270 are used to cluster the plurality of clothing label feature matrices to obtain cluster centers, calculate the distance between each clothing label feature matrix and the cluster center to obtain a cluster feature vector, and then perform weighted correction on each label matrix in the plurality of label matrices based on the cluster feature vector to obtain a plurality of weighted clothing label feature matrices. It should be understood that, considering that different recommendation schemes have different types in the plurality of label matrices, in order to distinguish these schemes and make more accurate clothing recommendations based on people's preferences, the technical solution of this application further clusters the plurality of clothing label feature matrices to obtain cluster centers. Accordingly, in a specific example, the positional mean of the plurality of clothing label feature matrices can be calculated to obtain the cluster center. Then, the distance between each clothing label feature matrix and the cluster center is calculated to obtain the cluster feature vector. In a specific example, the cosine distance between each of the multiple clothing label feature matrices and the cluster center can be calculated to obtain the clustering feature vector. This allows for the correction of the multiple label matrices; that is, based on the clustering feature vector, each label matrix in the multiple label matrices is weighted and corrected to obtain multiple weighted clothing label feature matrices.

[0062] More specifically, in this embodiment, the correction module includes: first, inputting the clustering feature vector into a Sigmoid function to map the feature values ​​at each position in the clustering feature vector to a probability interval of 0 to 1 to obtain a probabilistic clustering feature vector; then, using the feature values ​​at each position in the probabilistic clustering feature vector as weights to weight each of the plurality of label matrices to obtain the plurality of weighted outfit label feature matrices.

[0063] Figure 3 The diagram illustrates a block diagram of a calibration module in a big data-based intelligent cabinet control system according to an embodiment of this application. Figure 3 As shown, the correction module 270 includes: a weighting unit 271, used to input the clustering feature vector into a Sigmoid function to map the feature values ​​at each position in the clustering feature vector to a probability interval of 0 to 1 to obtain a probabilistic clustering feature vector; and a weighting unit 272, used to use the feature values ​​at each position in the probabilistic clustering feature vector as weights to weight each of the multiple label matrices to obtain the multiple weighted outfit label feature matrices.

[0064] Specifically, in this embodiment, the personal data encoding module 280 is used to pass the acquired basic personal data and personal preference data through a context encoder containing an embedding layer to obtain multiple personal outfit feature vectors, and then arrange the multiple personal outfit feature vectors into a feature matrix and pass it through a second convolutional neural network as a feature extractor to obtain an outfit requirement latent feature matrix. That is, further, in the technical solution of this application, in order to make accurate intelligent recommendations based on personal basic information and preference information, it is also necessary to acquire personal basic data (e.g., height, weight, 3D, and photos) and personal preference data (e.g., clothing style, preferences, etc.). Then, considering that the personal basic data and personal preference data of individuals have correlated feature information, a context encoder containing an embedding layer is further used to encode the personal basic data and personal preference data to extract global high-dimensional semantic features of the personal basic data and personal preference data to better represent the essential features of an individual's outfit, thereby obtaining multiple personal outfit feature vectors. In this way, the multiple personal outfit feature vectors are arranged into a feature matrix to integrate the implicit essential features of the individual's outfits, and then a second convolutional neural network, which acts as a feature extractor, is used to obtain the implicit feature matrix of outfit requirements.

[0065] Specifically, in this embodiment, the recalibration module 290 and the measurement module 300 are used to calibrate the implicit feature matrix of the dressing requirements and each weighted feature matrix of the plurality of weighted dressing label feature matrices to obtain calibrated implicit feature vectors of dressing requirements and plurality of calibrated weighted dressing label feature vectors. It should be understood that the implicit feature matrix of dressing requirements, for example denoted as M1, and the plurality of weighted feature matrices, for example denoted as M... 2t Since its row vectors correspond to the semantic directions of the label vectors and feature vectors, while the column directions correspond to the arrangement direction of the vectors, if we simply calculate M1 and M... 2t The distance between them means that the intra-sample distribution of eigenvalues ​​at various positions of the matrix in the semantic direction will affect the accuracy of the distance representation of the inter-sample feature distribution. Therefore, in the technical solution of this application, it is preferable that for the matrices M1 and M... 2t Perform explicit generalization of semantic reasoning information.

[0066] More specifically, in the embodiments of this application, the recalibration module is further configured to: calibrate each weighted clothing label feature matrix in the implicit feature matrix of the dressing requirements and the plurality of weighted clothing label feature matrices respectively using the following formulas to obtain the calibrated implicit feature vector of the dressing requirements and the plurality of calibrated weighted clothing label feature vectors;

[0067] The formula is as follows:

[0068]

[0069]

[0070] in, This represents the eigenvalues ​​at each position in the implicit feature matrix M1 representing the clothing requirements. M represents the multiple weighted outfit label feature matrices. 2t The feature values ​​at each position in the t-th weighted outfit label feature matrix, where V1 represents the latent feature vector of the corrected outfit requirement, V 2t Let represent the t-th corrected weighted outfit label feature vector among the plurality of corrected weighted outfit label feature vectors, and ∑ j (·) represents summing the eigenvalues ​​of each row of a matrix to obtain a column vector. This indicates differentiation based on position. In this way, the corrected implicit feature vector of clothing requirements and multiple corrected weighted feature vectors of clothing labels can be obtained. It should be understood that the explicit generalization of semantic reasoning information described above achieves information decoupling between sample semantic groups based on column direction by explicitly generalizing the semantic concepts corresponding to feature values ​​along the rows. This allows for the acquisition of the information plasticity of the high-dimensional manifold corresponding to the feature in the high-dimensional semantic space along the sample direction through information-based reasoning of feature semantics, thereby improving the accuracy of the calculation of spatial distance between matrices at the semantic expression level.

[0071] Specifically, in this embodiment, the measurement module 300, the recommendation result determination module 310, and the control result generation module 320 are used to calculate the distances between the corrected outfit requirement implicit feature vector and each of the multiple corrected weighted outfit label feature vectors to obtain multiple distance values. The four cabinets corresponding to the smallest of these multiple distance values ​​are then used as the outfit recommendation results. Based on these outfit recommendation results, the corresponding four cabinets of the smart cabinet are then activated. That is, in this technical solution, the distances between the corrected outfit requirement implicit feature vector and each of the multiple corrected weighted outfit label feature vectors are calculated, for example, using Euclidean distance, to obtain multiple distance values. Calculating the Euclidean distance here can better highlight the consistency between the outfit requirement and the outfit label. Therefore, the four cabinets corresponding to the smallest of the multiple distance values ​​can be used as the outfit recommendation results, and based on these results, the corresponding four cabinets of the smart cabinet are activated.

[0072] More specifically, in the embodiments of this application, the measurement module is further configured to: calculate the distance between each of the corrected weighted outfit label feature vectors in the corrected outfit requirement implicit feature vector and the plurality of corrected weighted outfit label feature vectors respectively using the following formulas to obtain the plurality of distance values;

[0073] The formula is as follows:

[0074]

[0075] Where V1 represents the feature value at each position in the latent feature vector of the corrected dressing requirements, V 2t D represents the feature value at each position of the t-th corrected weighted outfit label feature vector among the plurality of corrected weighted outfit label feature vectors. t This refers to the plurality of distance values.

[0076] In summary, the big data-based smart locker control system 200 described in the embodiments of this application is explained. It uses a convolutional neural network model based on artificial intelligence technology to mine deep feature association information from the type tags of clothing stored in each compartment of the smart locker, as well as personal basic data and personal preference data, in order to make personalized recommendations for people's clothing. Then, based on the recommendation results, the corresponding compartment of the smart locker is opened to meet people's needs.

[0077] As described above, the big data-based smart cabinet control system 200 according to the embodiments of this application can be implemented in various terminal devices, such as servers with big data-based smart cabinet control algorithms. In one example, the big data-based smart cabinet control system 200 according to the embodiments of this application can be integrated into the terminal device as a software module and / or a hardware module. For example, the big data-based smart cabinet control system 200 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the big data-based smart cabinet control system 200 can also be one of many hardware modules of the terminal device.

[0078] Alternatively, in another example, the big data-based smart cabinet control system 200 and the terminal device can also be separate devices, and the big data-based smart cabinet control system 200 can be connected to the terminal device via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0079] Exemplary methods

[0080] Figure 4 The diagram illustrates a flowchart of the control method for a smart cabinet control system based on big data. (For example...) Figure 4As shown, the control method of the smart cabinet control system based on big data according to the embodiment of this application includes the following steps: S110, obtaining the type tags of clothing stored in each compartment of the smart cabinet; S120, performing one-hot encoding on the type tags of clothing stored in each compartment of the smart cabinet to obtain tag vectors corresponding to each compartment; S130, arranging the four tag vectors corresponding to each of the four compartments of the smart cabinet, where the type tags are combinations of tops, bottoms, underwear, and shoes, in a two-dimensional arrangement to obtain multiple tag matrices; S140, passing each tag matrix in the multiple tag matrices through a first convolutional neural network as a feature extractor to obtain multiple outfit tag feature matrices; S150, clustering the multiple outfit tag feature matrices to obtain cluster centers; S160, calculating the distance between each outfit tag feature matrix in the multiple outfit tag feature matrices and the cluster centers to obtain cluster feature vectors; S170, based on the cluster feature vectors, performing... S180: Weighted correction is performed to obtain multiple weighted outfit label feature matrices; S190: The acquired personal basic data and personal preference data are passed through a context encoder containing an embedding layer to obtain multiple personal outfit feature vectors, and the multiple personal outfit feature vectors are arranged into a feature matrix and then passed through a second convolutional neural network as a feature extractor to obtain an outfit requirement latent feature matrix; S200: The outfit requirement latent feature matrix and each weighted outfit label feature matrix in the multiple weighted outfit label feature matrices are corrected to obtain corrected outfit requirement latent feature vectors and multiple corrected weighted outfit label feature vectors; S210: The distance between each corrected weighted outfit label feature vector in the corrected outfit requirement latent feature vector and the multiple corrected weighted outfit label feature vectors is calculated to obtain multiple distance values; S220: The four cabinets corresponding to the smallest of the multiple distance values ​​are taken as outfit recommendation results; and S220: Based on the outfit recommendation results, the corresponding four cabinets of the smart cabinet are opened.

[0081] Figure 5 The diagram illustrates the architecture of a control method for a big data-based intelligent cabinet control system according to an embodiment of this application. Figure 5 As shown, in the network architecture of the control method for the intelligent cabinet control system based on big data, firstly, the obtained type label (e.g., such as...) is... Figure 5 The combination of P1 shown is the four tag vectors corresponding to each of the four compartments of the smart cabinet for tops, bottoms, underwear, and shoes (e.g., such as...). Figure 5 The V1 shown is arranged in two dimensions to obtain multiple label matrices (e.g., as shown in the figure). Figure 5M1 as shown in the diagram); then, each of the multiple label matrices is passed through a first convolutional neural network (e.g., as shown in the diagram) which serves as a feature extractor. Figure 5 The CNN1 shown is used to obtain multiple clothing label feature matrices (e.g., such as...). Figure 5 (as shown in MF1); then, cluster the multiple outfit tag feature matrices to obtain cluster centers (e.g., such as...). Figure 5 (as shown in CC); then, the distance between each clothing tag feature matrix in the plurality of clothing tag feature matrices and the cluster center is calculated to obtain the cluster feature vector (e.g., as shown in CC); Figure 5 The VF1 shown in the diagram is then used; based on the clustering feature vector, each label matrix in the plurality of label matrices is weighted and corrected to obtain a plurality of weighted outfit label feature matrices (e.g., as shown in the diagram). Figure 5 (as shown in MF2); then, the acquired basic personal data and personal preference data (e.g., such as...) Figure 5 The P2 shown is transmitted through a context encoder containing an embedding layer (e.g., as shown in the image). Figure 5 The E shown is used to obtain multiple personal outfit feature vectors (e.g., such as...). Figure 5 The VF2 shown in the figure is used to arrange the multiple personal outfit feature vectors into a feature matrix (e.g., as shown in the figure). Figure 5 The MF3 shown is then passed through a second convolutional neural network (e.g., as shown) which acts as a feature extractor. Figure 5 The CNN2 shown is used to obtain the latent feature matrix of clothing needs (e.g., such as...). Figure 5 The MF4 shown in the diagram is then used to correct the implicit feature matrix of the dressing requirements and each weighted feature matrix of the multiple weighted dressing label feature matrices to obtain the corrected implicit feature vector of the dressing requirements (e.g., as shown in the diagram). Figure 5 The VF3 shown in the figure) and multiple corrected weighted outfit label feature vectors (e.g., such as Figure 5 The VF4 shown in the diagram is used; next, the distance between each of the corrected weighted outfit label feature vectors in the corrected outfit requirement latent feature vector and the plurality of corrected weighted outfit label feature vectors is calculated to obtain multiple distance values ​​(e.g., such as...). Figure 5 (as shown in the DV); then, the four cabinets corresponding to the smallest of the multiple distance values ​​are used as the outfit recommendation results (e.g., such as...). Figure 5 As shown in the diagram (R); and finally, based on the outfit recommendation results, open the corresponding four compartments of the smart cabinet.

[0082] In summary, the control method of the smart locker control system based on big data described in the embodiments of this application is explained. It uses a convolutional neural network model based on artificial intelligence technology to mine deep feature association information from the type tags of clothing stored in each compartment of the smart locker, as well as personal basic data and personal preference data, in order to make personalized recommendations for people's clothing. Then, based on the recommendation results, the corresponding compartment of the smart locker is opened to meet people's needs.

[0083] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0084] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0085] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0086] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0087] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A big data based intelligent cabinet control system, characterized in that, The method comprises the following steps: a label data acquisition module is configured to acquire type labels of clothes stored in each compartment of a smart cabinet; a label data encoding module is configured to encode the type labels of clothes stored in each compartment of the smart cabinet to obtain label vectors corresponding to each compartment; a dressing combination data construction module is configured to arrange four label vectors corresponding to each four-compartment of the smart cabinet in a two-dimensional manner to obtain a plurality of label matrices, wherein the four label vectors correspond to the combination of the type labels of clothes as tops, bottoms, underwears and shoes; a dressing combination data encoding module is configured to extract features from each label matrix in the plurality of label matrices by a first convolutional neural network to obtain a plurality of dressing label feature matrices; a clustering module is configured to cluster the plurality of dressing label feature matrices to obtain a cluster center; an evaluation module is configured to calculate the distance between each dressing label feature matrix in the plurality of dressing label feature matrices and the cluster center to obtain a cluster feature vector; a correction module is configured to weight and correct each label matrix in the plurality of label matrices based on the cluster feature vector to obtain a plurality of weighted dressing label feature matrices; a personal data encoding module is configured to extract personal basic data and personal preference data, and then encode the extracted data by a context encoder comprising an embedding layer to obtain a plurality of personal dressing feature vectors, and then arrange the plurality of personal dressing feature vectors into a feature matrix and extract features from the feature matrix by a second convolutional neural network to obtain a dressing demand implicit feature matrix; a re-correction module is configured to correct the dressing demand implicit feature matrix and each weighted dressing label feature matrix in the plurality of weighted dressing label feature matrices to obtain a corrected dressing demand implicit feature vector and a plurality of corrected weighted dressing label feature vectors; a metric module is configured to calculate the distance between the corrected dressing demand implicit feature vector and each corrected weighted dressing label feature vector in the plurality of corrected weighted dressing label feature vectors to obtain a plurality of distance values; a recommendation result determination module is configured to determine the four compartments corresponding to the smallest distance value in the plurality of distance values as a dressing recommendation result; and a control result generation module is configured to open the corresponding four compartments of the smart cabinet based on the dressing recommendation result.

2. The big data based intelligent cabinet control system of claim 1, wherein, The dressing combination data encoding module is further configured to perform the following operations in the forward propagation of each layer of the first convolutional neural network: perform convolution processing on input data to obtain a convolution feature map; perform mean pooling based on the local channel dimension on the convolution feature map to obtain a pooling feature map; and perform nonlinear activation on the pooling feature map to obtain an activated feature map. The output of the last layer of the first convolutional neural network is the plurality of dressing label feature matrices, and the input of the first layer of the first convolutional neural network is each label matrix in the plurality of label matrices. The clustering module is further configured to calculate the position-wise mean of the plurality of dressing label feature matrices to obtain the cluster center.

3. The big data based intelligent cabinet control system of claim 2, wherein, ​ 4. The big data based intelligent cabinet control system of claim 3, wherein, The evaluation module is configured to calculate cosine distances between each of the plurality of dressing label feature matrices and the cluster center to obtain a cluster feature vector.

5. The big data based intelligent cabinet control system of claim 4, wherein, The correction module comprises: a weighting unit configured to input the cluster feature vector into a Sigmoid function to map feature values at each position in the cluster feature vector into a probability interval of 0 to 1 to obtain a probability cluster feature vector; and a weighting unit configured to weight each of the plurality of label matrices respectively with the feature values at the corresponding positions in the probability cluster feature vector as weights to obtain the plurality of weighted dressing label feature matrices.

6. The big data based intelligent cabinet control system of claim 5, wherein, The re-correction module is further configured to correct the dressing demand implicit feature matrix and each of the plurality of weighted dressing label feature matrices respectively according to the following formula to obtain a corrected dressing demand implicit feature vector and a plurality of corrected weighted dressing label feature vectors. The formula is: wherein, represents the eigenvalue of each position in the clothing demand implied feature matrix, represents the eigenvalue of each position in the tth weighted clothing label feature matrix in the plurality of weighted clothing label feature matrices, V1represents the corrected clothing demand implied feature vector, V 2t represents the tth corrected weighted clothing label feature vector in the plurality of corrected weighted clothing label feature vectors, and Σ j (·) represents the sum of eigenvalues of each row of a matrix, represents the difference by position.

7. The big data based intelligent cabinet control system of claim 6, wherein, The measurement module is further configured to calculate distances between the corrected dressing demand implicit feature vector and each of the plurality of corrected weighted dressing label feature vectors respectively according to the following formula to obtain a plurality of distance values. The formula is: wherein V1 represents the eigenvalue of each position in the corrected fitting demand implicit feature vector, V 2t represents the eigenvalue of each position in the tth corrected weighted fitting label feature vector in the plurality of corrected weighted fitting label feature vectors, D t represents the plurality of distance values. 8.A control method of a big data-based intelligent cabinet control system, characterized by, comprises: obtaining type labels of clothes stored in each of the compartments of the intelligent cabinet; one-hot encoding the type labels of the clothes stored in each of the compartments of the intelligent cabinet to obtain a label vector corresponding to each of the compartments; performing two-dimensional arrangement on four label vectors corresponding to every four compartments of the intelligent cabinet, in which the combination of the type labels is upper garment, lower garment, underwear and shoes, to obtain a plurality of label matrices; extracting features from each of the plurality of label matrices through a first convolutional neural network as a feature extractor to obtain a plurality of dressing label feature matrices; performing clustering on the plurality of dressing label feature matrices to obtain a cluster center; calculating distances between each of the plurality of dressing label feature matrices and the cluster center to obtain a cluster feature vector; weighting and correcting each of the plurality of label matrices based on the cluster feature vector to obtain a plurality of weighted dressing label feature matrices; obtaining personal basic data and personal preference data through a context encoder comprising an embedding layer to obtain a plurality of personal dressing feature vectors, and performing feature extraction on the plurality of personal dressing feature vectors arranged as a feature matrix through a second convolutional neural network as a feature extractor to obtain a dressing demand implicit feature matrix; correcting the dressing demand implicit feature matrix and each of the plurality of weighted dressing label feature matrices respectively to obtain a corrected dressing demand implicit feature vector and a plurality of corrected weighted dressing label feature vectors; calculating distances between the corrected dressing demand implicit feature vector and each of the plurality of corrected weighted dressing label feature vectors respectively to obtain a plurality of distance values; and The four cabinet compartments corresponding to the minimum distance value in the plurality of distance values are taken as the dressing recommendation result; And Based on the dressing recommendation result, the corresponding four cabinet compartments of the smart cabinet are opened. 9.The control method of the big data-based intelligent cabinet control system according to claim 8, wherein, Respectively passing each label matrix in the plurality of label matrices through a first convolutional neural network as a feature extractor to obtain a plurality of dressing label feature matrices, including that each layer of the first convolutional neural network respectively performs: Convolution processing on input data to obtain a convolution feature map; Local channel dimension-based mean pooling on the convolution feature map to obtain a pooling feature map; and Nonlinear activation on the pooling feature map to obtain an activated feature map; Wherein, the output of the last layer of the first convolutional neural network is the plurality of dressing label feature matrices, and the input of the first layer of the first convolutional neural network is each label matrix in the plurality of label matrices. 10.The control method of the big data-based intelligent cabinet control system according to claim 9, wherein, Based on the clustering feature vector, weighting correction is performed on each label matrix in the plurality of label matrices to obtain a plurality of weighted dressing label feature matrices, including: Inputting the clustering feature vector into a Sigmoid function to map the feature values at each position in the clustering feature vector to a probability interval of 0 to 1 to obtain a probabilistic clustering feature vector; and Respectively weighting each label matrix in the plurality of label matrices with the feature values at each position in the probabilistic clustering feature vector as weights to obtain the plurality of weighted dressing label feature matrices.

Images