CN112598486B - Marketing accurate screening push system based on big data and intelligent internet of things - Google Patents
Marketing accurate screening push system based on big data and intelligent internet of things Download PDFInfo
- Publication number
- CN112598486B CN112598486B CN202110019917.6A CN202110019917A CN112598486B CN 112598486 B CN112598486 B CN 112598486B CN 202110019917 A CN202110019917 A CN 202110019917A CN 112598486 B CN112598486 B CN 112598486B
- Authority
- CN
- China
- Prior art keywords
- customer
- module
- heat
- market
- information
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- 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
- G06F16/90—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
- G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention relates to the technical fields of intelligent marketing, big data and intelligent Internet of things, in particular to a marketing accurate screening pushing system based on the big data and the intelligent Internet of things. The system comprises: the system comprises a market perception module, a market operation big data module, a data calculation and analysis module and a commodity pushing module; customer information and sales volume data in a market are obtained through a market perception module and a market operation big data module; the data calculation and analysis module processes the customer information and sales volume data to obtain a store area feature matrix; and the commodity pushing module analyzes the customer information and the store regional characteristic matrix and pushes the commodity to the customer. According to the invention, commodity pushing is carried out on the customers in the online entity economic system through the intelligent Internet of things and big data analysis, so that the customer consumption capability is promoted.
Description
Technical Field
The invention relates to the technical fields of intelligent marketing, big data and intelligent Internet of things, in particular to a marketing accurate screening pushing system based on the big data and the intelligent Internet of things.
Background
With the development of scientific technology, new technologies, new products and new services of the big data market are continuously emerging, and big data is becoming a current hot topic. However, the rapid development of online economy causes lag and operation drop of offline entity economy development, but entity stores are also indispensable in daily life, so that we pay attention to the co-development of entity economy while developing online economy. Meanwhile, shopping on the internet has a plurality of defects that pictures are inconsistent with objects, quality is unqualified and the like. How to promote consumer ability through big data analysis in a online lower entity economic system is a problem to be solved at present.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a marketing accurate screening pushing system based on big data and intelligent Internet of things, and the adopted technical scheme is as follows:
the invention provides a marketing accurate screening and pushing system based on big data and intelligent Internet of things, which comprises: the system comprises a market perception module, a market operation big data module, a data calculation and analysis module and a commodity pushing module;
the market sensing module is used for acquiring customer information through monitoring equipment in the market; the customer information includes customer location information;
the store operation big data module is used for acquiring operation data of the store, wherein the operation data comprises sales volume data of stores in the store;
the data calculation and analysis module is used for projectively transforming the customer position information to a preset market plane overlook image, and obtaining a projected heat map of the market area through projection points; the projected heat map comprises discrete data points representing heat; establishing a Thiessen polygon through the discrete data points; calculating the vertex heat value of each vertex in the Thiessen polygon through an interpolation algorithm; obtaining the degree of influence of the vertex on the heat of any position by calculating the distance between the vertex heat value and any position in the heat map so as to obtain the heat of any position; acquiring a heat level characteristic diagram of the market according to the heat of the arbitrary position; converting the sales data into a sales proportion matrix; combining the sales volume proportion matrix and the heat level characteristic diagram into a shop area characteristic matrix;
the commodity pushing module is used for combining the customer information and the store area feature matrix analysis and pushing commodities to the customer through display equipment.
Further, the market sensing module further comprises a customer behavior detection module;
the customer behavior detection module is used for judging whether the customer enters a fitting room or not through a pedestrian re-identification technology, and recording the times of the customer entering the fitting room; identifying the tag brought into the clothes hanger of the fitting room by utilizing a wireless video technology to obtain a fitting list taken by the customer; the fitting list includes the number of identified garments and garment attributes; the number of times the customer enters the fitting room and the fitting list are used as customer behavior information; the customer information includes the customer behavior information.
Further, the market perception module further comprises a customer bounding box acquisition module and a bounding box analysis module;
the customer bounding box acquisition module is used for outputting a target bounding box of the customer through a pre-trained target detection network;
the bounding box analysis module is used for analyzing the target bounding box through a pre-trained convolutional neural network to obtain customer physique information; the customer information includes the customer physique information; obtaining customer clothing collocation information through a pre-trained example segmentation network; the customer information includes the customer garment collocation information.
Further, the data calculation and analysis module further comprises a heat map acquisition module;
the heat map acquisition module is used for carrying out four-point method estimation on the corresponding coordinates of the image ground label acquired by the monitoring equipment and the market plane overlook image, projecting the bottom edge center of the target bounding box into the market plane overlook image through a homography matrix, and acquiring the projection points; generating a thermodynamic diagram of a two-dimensional gaussian distribution based on the projection points in the mall planar top view image; and obtaining the projection heat map of the market area according to time sequence statistics.
Further, the data calculation and analysis module further comprises an information filtering module;
the information filtering module is used for filtering the heat map through a maximum lattice point sampling method; and processing the projected heat map through a preset sliding window, wherein each processing only keeps the maximum value in the sliding window, and other values are zeroed.
Further, the data calculation and analysis module further comprises a region heat acquisition module;
the regional heat acquisition module is used for classifying the plan view of the mall to obtain regional images; calculating the area of any region in the region image, and calculating the region heat according to the quantity, the area and the discrete data point value of the Thiessen polygons contained in the region:
wherein, areaH i Representing the heat degree of the ith region in the region image, S i,j Representing the area of the jth Thiessen polygon in the ith region, S i Representing the area of the ith region in the region image, H i,j Discrete data point values representing the jth Thiessen polygon contained in the ith region, and n representing n regions divided in the region image.
Further, the data calculation and analysis module further comprises a cluster analysis module;
and the cluster analysis module is used for generating a market heat map after obtaining heat at any position, dividing the market heat map into heat levels based on pixel values through a clustering algorithm, and generating the heat level characteristic map.
Further, the commodity pushing module further comprises a commodity pushing neural network module;
the commodity pushing neural network module is used for analyzing the customer information and the store regional characteristic matrix input into the network through the trained commodity pushing neural network and outputting the pushed commodity.
Further, the commodity pushing module further comprises an offline commodity pushing module;
the off-line commodity pushing module is used for counting the heat and the stay time of the area in the mall where the track information is located by analyzing the track information of the customer in the mall, and pushing the commodity through a mobile phone after the customer leaves the mall.
The invention has the following beneficial effects:
1. according to the embodiment of the invention, the information of the customers and the information in the market are constructed through the visual perception in the market and the big data perceived by the intelligent Internet of things. Through analysis and training of the information, the system can analyze the current information in real time and accurately push commodities to customers.
2. The commodity pushing neural network in the embodiment of the invention only needs to acquire the labeling data based on the historical sales volume data, the visual perception data and the Internet of things perception data, the training method is simple, the data is easy to acquire, and the label data does not need to be marked manually.
3. According to the embodiment of the invention, the offline pushing module analyzes the moving track information of the customer in the mall, and the offline commodity pushing can be carried out on the customer according to the track information when the customer needs to push the commodity after leaving the mall, so that the pushing efficiency is improved.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions and advantages of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are only some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a block diagram of a marketing accurate screening and pushing system based on big data and intelligent Internet of things according to an embodiment of the present invention;
fig. 2 is a block diagram of a commodity pushing neural network according to an embodiment of the present invention.
Detailed Description
In order to further explain the technical means and effects adopted by the invention to achieve the preset aim, the following is a detailed description of a specific implementation, structure, characteristics and effects of the marketing accurate screening pushing system based on big data and intelligent internet of things according to the invention by combining the accompanying drawings and the preferred embodiment. In the following description, different "one embodiment" or "another embodiment" means that the embodiments are not necessarily the same. Furthermore, the particular features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The invention provides a specific scheme of a marketing accurate screening pushing system based on big data and intelligent Internet of things, which is specifically described below with reference to the accompanying drawings.
Referring to fig. 1, a block diagram of a marketing accurate screening and pushing system based on big data and intelligent internet of things according to an embodiment of the present invention is shown, where the system includes: the system comprises a market perception module 101, a market operation big data module 102, a data calculation and analysis module 103 and a commodity pushing module 104.
The market sensing module 101 is configured to obtain customer information through monitoring devices inside the market. The monitoring equipment comprises a monitoring camera, a sensor and the like. The customer information includes customer physique information, customer behavior information, and customer clothing collocation information.
In the embodiment of the invention, the customer physique information comprises the sex, age, height and weight grade of the customer; the customer behavior information comprises the times that the customer enters the fitting room and a fitting list; the customer clothing collocation information includes the color, style, fabric and style of clothing worn by the customer.
Preferably, the market sensing module 101 in the embodiment of the present invention includes a customer bounding box acquisition module and a bounding box analysis module.
The customer bounding box acquisition module is used for outputting the target bounding box of the customer through a pre-trained target detection network. The training process of the target detection network specifically comprises the following steps:
1) And taking the image acquired by monitoring equipment in the mall as a training image. And labeling the length and width information of the bounding box at the center point position of the customer in the training image. And (5) checking the customer center position through Gaussian convolution to carry out convolution, so as to obtain a customer center heat map. There is (x, y, w, h) information for each point location in the customer's central heat map, where x, y represents the coordinates of the customer's point, and w, h is the length and width of the bounding box.
2) And (3) normalizing the training image and the label data, and changing the picture matrix into a floating point number between [0,1] so that the model can be better converged. And inputting the normalized data into a network.
3) The network structure adopts an encoding-decoding structure. The target detection encoder performs feature extraction on the training image and outputs a feature map. The object detection decoder performs up-sampling and feature extraction on the feature map and outputs a customer center heat map and a bounding box.
4) The network uses a weighted sum of the center point prediction loss and bounding box size loss. The mathematical formula for the center point loss is as follows:
where α and β are hyper-parameters, and N is the number of center points of bounding boxes in the image, set by human experience. Gamma ray xy For the value of xy coordinates, y in the predicted customer center heat map xy Is the value of the xy coordinates in the real data heat map (Ground Truth Heatmap).
The mathematical formula for bounding box size loss is as follows:
wherein N is the center point of the bounding box in the imageQuantity S Pk S is the length and width of the predicted bounding box k Is the length and width of the bounding box of the real data.
The total loss function is:
Total Loss＝CenterLoss+δ*SizeLoss
where δ is the weight. In the embodiment of the invention, delta is 0.1.
5) And carrying out post-processing on the obtained customer center heat map and the bounding box, and searching for peak points to obtain specific target bounding box information and center point coordinate information as customer position information. Post-processing methods include maximum suppression, softargmax, and the like.
The bounding box analysis module is used for analyzing the target bounding box through a pre-trained convolutional neural network to obtain the customer physique information, and the convolutional neural network can be analyzed through a residual error network (ResNet). And analyzing the image in the target bounding box through a pre-trained example segmentation network to obtain a segmentation image of each piece of clothing on the customer. Multiplying the segmented image with the original image to obtain an original image of the clothing instance without being affected by the doped background, classifying by a convolutional neural network to obtain the wearing color, style, fabric and style of the customer, and matching the clothing instance with the customer
The mall awareness module 101 also includes a customer behavior detection module. The customer behavior detection module is used for judging whether a customer enters the fitting room or not through a pedestrian re-identification technology and recording the times of the customer entering the fitting room. And identifying the tag on the clothes hanging tag in the fitting room by utilizing a wireless radio frequency technology to obtain a fitting list taken by a customer. The fitting list includes the number of identified garments and garment attributes.
The store operation big data module 102 is used for counting through long-term real-time data statistics and accumulated big data based on systems such as store video analysis, cash register system analysis and the like. And acquiring operation data of the mall, wherein the operation data comprises sales volume data of shops in the mall.
The data calculation and analysis module 103 projects and transforms the customer information onto a pre-established market plane top view image, and obtains a heat map of the market area through projection points.
Preferably, the data calculation analysis module 103 includes a heat map acquisition module. The heat map acquisition module is used for estimating coordinates corresponding to the image ground label acquired by the monitoring equipment and the market plane overlook image by a four-point method, and performing projection transformation on the center point of the bottom edge of the target bounding box through the homography matrix. A thermodynamic diagram of a two-dimensional gaussian distribution is generated based on projected points in a market plane top view image. The gaussian kernel size is set to 3*3 in the embodiment of the present invention. And adding the thermodynamic diagrams between frames according to pixels, and obtaining a projected thermodynamic diagram of the market area through time sequence statistics, wherein the pixel value of each pixel point in the projected thermodynamic diagram is the heat of the position. In the embodiment of the invention, the cycle time set by the time sequence statistics is one hour.
Preferably, the data calculation and analysis module 103 further comprises an information filtering module. The information filtering module is used for filtering the projection heat map by adopting a lattice sampling method. And processing the projected heat map through the set sliding window, wherein only the maximum value in the window is reserved in each sliding window, and other values are subjected to zero resetting processing. The information filtering module can filter most data, and the subsequent calculated amount is reduced. In the embodiment of the invention, the window size of the sliding window is 5*5, and the stride is 5.
The data calculation analysis module 103 builds a Thiessen polygon by projecting discrete data points representing heat within the heat map. The theoretical region of influence for each discrete data point can be determined by constructing a Thiessen polygon. The step of the Thiessen polygon is:
1) The discrete data points automatically construct a triangulation network, i.e., a triangulation (Delaunay) triangulation network. For the discrete data points and the triangle numbers formed, record which three discrete data points each triangle is made up of.
2) The numbers of all triangles adjacent to each discrete data point are found and recorded. I.e. all triangles with one and the same vertex are found in the constructed triangle mesh.
3) Triangles adjacent to each discrete data point are ordered in either a clockwise or counterclockwise direction for the next connection to generate a Thiessen polygon. Let the discrete data point be o. Finding out a triangle taking o as a vertex, and setting the triangle as A; taking the other vertex except o of the triangle A and setting the other vertex as a, and finding out the other vertex, namely f; the next triangle must be of-sided, i.e., triangle F; the other vertex of triangle F is e, then the next triangle is on the side of oe; this is repeated until the oa edge is returned.
4) Calculating the circle center of the circumscribed circle of each triangle, and recording.
5) And connecting the circle centers of the circumscribed circles of the adjacent triangles according to the adjacent triangles of each discrete data point to obtain the Thiessen polygon. For Thiessen polygons at the triangle mesh edge, the perpendicular bisector can be made to intersect the drawing profile, and the Thiessen polygons are formed together with the drawing profile.
The data calculation and analysis module 103 calculates vertex heat values of the vertices in the Thiessen polygon by a spatial interpolation algorithm. The method specifically comprises the following steps:
a) The vertex heat value of the common vertex belonging to at least two Thiessen polygons in the Thiessen polygon vertices is first calculated. The distance from each Thiessen polygon vertex (x 1, y 1) to a discrete data point (x, y) within an adjacent Thiessen polygon is calculated. Calculating a distance weight for each vertex, the weights being represented by the inverse of the distance:
wherein, the liquid crystal display device comprises a liquid crystal display device,represents the distance from the ith vertex to the discrete data point, n represents the total number of vertices, lambda, of the total number of vertices of the Thiessen polygon in which the discrete data point is located i The distance weight for each vertex.
And calculating the final Thiessen polygon vertex hotness through the obtained weight:
wherein, the liquid crystal display device comprises a liquid crystal display device,the heat value for the Thiessen polygon vertex is calculated. F (X) i ,Y i ) Representing discrete data point values within the ith neighboring Thiessen polygon.
b) When the Thiessen polygon vertex belongs to only one Thiessen polygon, namely only one discrete data point affects the vertex, the vertex is positioned at the boundary of the image, and the method for obtaining the vertex heat comprises the following steps:
firstly, obtaining the average heat attenuation ratio of the polygon:
where k is the number of vertices in the polygon that have been spatially interpolated,for the first vertex heat value, d, after spatial interpolation l The distance from the first vertex, which has been spatially interpolated, to the polygon discrete data point. Bx is the average heat decay ratio of the ith tesen polygon. H i I.e., the discrete data point heat value of the ith Thiessen polygon.
Then, the vertex heat value which is not interpolated in the polygon is calculated:
wherein d O Is the distance from the vertex prior to spatial interpolation in the ith polygon to the polygon discrete data point.The heat value of the vertex which is not interpolated in the O-th polygon.
After the data calculation and analysis module 103 obtains the heat value of each vertex of the Thiessen polygon, the distance from any point in the Thiessen polygon to the vertex of the Thiessen polygon is calculated. And a step of calculating the heat degree of the point by using the weight calculation formula and the heat degree calculation formula in the step a).
The influence range of the heat points is not considered by directly interpolating the heat values of all points through IDW, and the vertexes of the Thiessen polygons are commonly owned by a plurality of Thiessen polygons, so that errors exist in direct interpolation, and the precision is further influenced. The heat value of the pixels in the region is distributed to each pixel according to the proportion occupied by the reciprocal of the distance (or the square of the reciprocal), and the heat value of the adjacent region is not considered, so that the heat of points in the polygon region, which are farther from the discrete data points, is lower, and the rule of heat statistics is not met. Meanwhile, the heat value is reflected by Gaussian distribution of the projection positions of the people and time sequence superposition statistics, and the discrete data points after the maximum grid point sampling can cause larger heat value difference among formed Thiessen polygons. Therefore, the embodiment of the invention adopts double IDW interpolation to obtain the heat value of the Thiessen polygon vertexes, and then the heat value of any position in each Thiessen polygon area is obtained through the vertex heat value interpolation. The reliability and the accuracy of the calculation result are ensured.
Preferably, the data calculation and analysis module 103 further includes a region heat acquisition module. The regional heat acquisition module is used for classifying the plan view of the mall to obtain regional images. Calculating the area of any region in the region image, and calculating the region heat according to the quantity, the area and the discrete data point value of the Thiessen polygons contained in the region:
wherein, areaH i Representing the heat degree of the ith region in the region image, S i,j Representing the overlapping area of the jth Thiessen polygon contained in the ith region, S i Represents the area of the ith region in the region image, H i,j The discrete data point heat value of the jth Thiessen polygon contained in the ith region is represented, and n represents that n regions are divided in the region image.
The data computation analysis module 103 also includes a cluster analysis module. And the cluster analysis module is used for generating a mall heat map after obtaining the heat of any point. And processing pixels with pixel values larger than 0 in the market heat map by a Kmeans clustering algorithm. In the embodiment of the invention, the number of clustering centers of Kmeans is 3, which represents the division of low heat, medium heat and high heat based on pixel values, and the numbers 1, 2 and 3 are used for representing corresponding heat levels respectively. And generating a heat level characteristic diagram.
The data calculation and analysis module 103 constructs sales data obtained by the mall operation big data module 102 into a sales sequence of the clothing region. A sales ratio sequence for each clothing region was obtained using a Softmax function. A clothing region sales volume ratio matrix is formed, which has the same size as the heat level feature map. And combining the clothing region sales volume proportion matrix and the heat grade characteristic diagram to obtain a shop region characteristic matrix.
The commodity pushing module 104 is used for analyzing the combination of the customer information and the store area feature matrix, and pushing commodities to customers through the display device.
Preferably, the merchandise push module 104 includes a merchandise push neural network module. Referring to fig. 2, a block diagram of a commodity pushing neural network according to an embodiment of the present invention is shown. The commodity pushing neural network module is used for analyzing customer physical information, customer clothing collocation information and a shop regional characteristic matrix which are input into the network through the pre-trained commodity pushing neural network, and outputting the pushed commodity.
The commodity pushing neural network training process specifically comprises the following steps:
1) The data is first integrated.
A one-dimensional matrix 201 of the constitution of the customer is constructed, and the sex, age, weight level and height of the customer are obtained from the constitution information of the customer. The customer constitution information is formed into a one-dimensional matrix, the matrix is [1,4],4 represents the customer constitution information, and 1 represents a matrix.
A one-dimensional matrix 203 of customer behaviors is constructed, the number of times of the customer's trial entry and the fitting room, the number of times of the changing, the properties of the changing clothes and the position of the customer are obtained through the customer behavior information, a one-dimensional matrix is generated, the shapes of the matrices are [1,4],4 respectively represent the customer behavior information, and 1 represents a matrix. The customer location (X, Y coordinates) refers to the customer projected coordinates in the planar top view image of the store.
A customer clothing matching matrix 202 is constructed, colors, styles, fabrics and styles of customer clothing are obtained through the customer clothing matching information, a one-dimensional matrix is generated, the shape of the matrix is [1,4,5],4 represents four descriptions of the clothing matching information, 5 represents a clothing type index value, and 1 represents a matrix.
In the embodiment of the present invention, the structure of the matrix is illustrated by the customer clothing matching matrix, and the colors in the customer clothing matching matrix 202 represent the maximum color proportion of the clothing, including red (1), yellow (2), black (3), white (4), blue (5), brown (6), etc., and the numbers in brackets represent the index values of different colors; the styles comprise profession (1), fashion (2), elegance (3), printing (4), leisure (5), evening (6) and the like, and the numbers in brackets represent index values of the same style; the fabric represents the maximum fabric proportion of the garment, including cotton (1), hemp (2), silk (3), chemical fiber (4) and the like, and the numbers in brackets represent index values of different fabrics; styles include sweet (1), daily (2), european (3), england (4), etc., with numerals in brackets representing index values of different styles and index value 0 when the customer is not wearing a cap scarf. The customer clothing matching matrix 202 is shown in Table 1, and the matrix values are index values:
table 1 clothing match matrix
2) When the customer constitution one-dimensional matrix 201, the customer behavior one-dimensional matrix 203 and the customer clothing collocation matrix 202 are input into the network, normalization processing is performed, so that the convergence of the network is quickened.
The customer constitution one-dimensional matrix 201 inputs the first full-connected network 205 in the shape of [ B,4], B is the batch size of the network input, 4 is the 4 attributes of the customer, and finally the first full-connected network 205 outputs a 64-dimensional high-dimensional feature vector to the first embedding layer 209.
The one-dimensional matrix 203 of customer behavior is input to the first fully-connected network 205 in the form of [ C,4], C being the batch size of the network input, 4 being four features of the customer behavior, and finally the second fully-connected network 207 outputs a 64-dimensional high-dimensional feature vector to the second embedded layer 210.
The customer clothing matching matrix 202 is input into the one-dimensional convolutional neural network 206 for feature extraction and downsampling to obtain a feature map, and then the feature map is input into the fully-connected layer through a flattening (flat) operation, and mapped into a 64-dimensional high-dimensional feature vector to the third embedded layer 211.
The store area feature matrix 204 is input into the two-dimensional convolutional neural network 208 for feature extraction and downsampling to obtain a feature map, and then the feature map is input into the fully-connected layer through a flattening (flat) operation, and mapped into a 64-dimensional high-dimensional feature vector to the fourth embedded layer 212.
The first fully connected network 205 and the second fully connected network 207 act as a map, with a final number of neurons of N, N representing the output dimension. Each fully connected network design should be at two or more layers to ensure that the sequence tensors can be fully mapped to the feature space. In the embodiment of the invention, N is 64.
The first, second, third and fourth embedded layers 209, 210, 211 and 212 perform a dot multiplication operation to obtain a 64-dimensional vector, which is input into the fifth embedded layer 213. The fifth embedding layer 213 fuses the behavior information, attribute information, clothing collocation and heat information of each area inside the store of the customer, so that similarity measurement can be better performed on the customer, and clothing recommendation can be more accurately realized.
3) The network training tag data is information data obtained when each customer purchases, and is obtained based on online video perception, intelligent internet of things perception and cashing statistics. The information data at each purchase by the customer is one category. The information data of each time the customer purchases comprises customer attributes, clothing collocation and behavior information data.
4) The network training method uses an AM-softmax loss function for classification training, removes the last classification layer of the third fully connected network 214 from the trained network, and selects the last hidden layer output as a feature of data. The two data features are computed using cosine similarity. The classification layer outputs the probability of each category, and finally adopts a Softmax classification function.
Preferably, the last layer of the sorting layer neuron number of the third fully-connected network 214 is modified and trained based on the previous network, and the network update can be realized based on the commodity update, so that the neural network is dynamically changed based on clothing, the accuracy of the network is further improved, and the online learning function is realized.
The commodity pushing neural network working process specifically comprises the following steps: after the customer enters the store, the second fully-connected network 207 is closed, the third embedded layer 211 is set to be 1, finally, input information of the first embedded layer 209 and the second embedded layer 210 is acquired through the first fully-connected network 205 and the one-dimensional convolutional neural network 206, and finally, the input information is input into the third fully-connected network 214 after dot multiplication to evaluate the similarity of purchase information data of the historical customer, and the first clothing optimization recommendation is performed. And (3) saving the data information of the first embedded layer 209 and the second embedded layer 210, closing the first fully-connected network 205 and the one-dimensional convolutional neural network 206, performing real-time reasoning of the second fully-connected network 207 by sensing the behavior characteristics of the customers through monitoring equipment, intelligent Internet of things and other equipment, acquiring the data of the third embedded layer 211 with dynamic change to perform similarity evaluation of the purchase information data of historical customers, and performing subsequent dynamic clothing recommendation based on the behaviors of the customers. For the fourth embedded layer 212, since the internal heat map of the store is time series statistical, it is also once every statistical period, then the fourth embedded layer 212 is saved and the two-dimensional convolutional neural network 208 is turned off.
The merchandise pushing module 104 performs similarity assessment of the customer information data through the merchandise pushing module, and then may recommend historical purchase garments of similar customer information data of Top-K to the customer. In the embodiment of the invention, K is taken as 3. Meanwhile, the pushing commodity result is displayed through an LED screen or other display devices, and the characteristics of the commodity can be explained through broadcasting through voice devices, so that the customer is guided to purchase.
Preferably, the merchandise pushing module 104 further includes an offline merchandise pushing module. The off-line commodity pushing module is used for counting the heat information H of the customer in each clothing area by analyzing the moving track information of the customer in the mall after leaving the mall i ：
H i ＝t i *AreaH i
Wherein H is i Indicating the heat level of the customer in the ith clothing area, t i Indicating the customer's residence time in the ith clothing area, areaH i Is the heat value of the ith garment region.
And then selecting commodities of the customer in the area of the store clothing area heat Top-K to push. The customer's heat at various locations within the store Top-K heat clothing area may be selected for more accurate merchandise pushing.
The off-line commodity pushing module installs an APP capable of identifying a store Bluetooth beacon on the mobile phone and pushes commodities on the mobile phone.
In summary, the embodiment of the invention constructs the customer characteristic information through the visual perception in the market and the big data perceived by the intelligent internet of things, carries out real-time commodity pushing on the customers through the commodity pushing network in the commodity pushing module, and carries out commodity pushing on the customers through the mobile phone by analyzing the moving track and the stay time of the customers in the market through the offline pushing module after the customers leave the market, so that the purchasing strength of the customers is improved.
It should be noted that: the sequence of the embodiments of the present invention is only for description, and does not represent the advantages and disadvantages of the embodiments. And the foregoing description has been directed to specific embodiments of this specification. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.
Claims (9)
1. Marketing accurate screening push system based on big data and intelligent thing networking, its characterized in that, the system includes: the system comprises a market perception module, a market operation big data module, a data calculation and analysis module and a commodity pushing module;
the market sensing module is used for acquiring customer information through monitoring equipment in the market; the customer information includes customer location information;
the store operation big data module is used for acquiring operation data of the store, wherein the operation data comprises sales volume data of stores in the store;
the data calculation and analysis module is used for projectively transforming the customer position information to a preset market plane overlook image, and obtaining a projected heat map of the market area through projection points; the projected heat map comprises discrete data points representing heat; establishing a Thiessen polygon through the discrete data points; calculating the vertex heat value of each vertex in the Thiessen polygon through an interpolation algorithm; obtaining the degree of influence of the vertex on the heat of any position by calculating the distance between the vertex heat value and any position in the heat map so as to obtain the heat of any position; acquiring a heat level characteristic diagram of the market according to the heat of the arbitrary position; converting the sales data into a sales proportion matrix; combining the sales volume proportion matrix and the heat level characteristic diagram into a shop area characteristic matrix;
the commodity pushing module is used for combining the customer information and the store area feature matrix analysis and pushing commodities to the customer through display equipment.
2. The marketing accurate screening pushing system based on the big data and the intelligent internet of things of claim 1, wherein the market perception module further comprises a customer behavior detection module;
the customer behavior detection module is used for judging whether the customer enters a fitting room or not through a pedestrian re-identification technology, and recording the times of the customer entering the fitting room; identifying the tag brought into the clothes hanger of the fitting room by utilizing a wireless video technology to obtain a fitting list taken by the customer; the fitting list includes the number of identified garments and garment attributes; the number of times the customer enters the fitting room and the fitting list are used as customer behavior information; the customer information includes the customer behavior information.
3. The marketing accurate screening pushing system based on the big data and the intelligent internet of things according to claim 1, wherein the market perception module further comprises a customer bounding box acquisition module and a bounding box analysis module;
the customer bounding box acquisition module is used for outputting a target bounding box of the customer through a pre-trained target detection network;
the bounding box analysis module is used for analyzing the target bounding box through a pre-trained convolutional neural network to obtain customer physique information; the customer information includes the customer physique information; obtaining customer clothing collocation information through a pre-trained example segmentation network; the customer information includes the customer garment collocation information.
4. The marketing accurate screening pushing system based on the big data and the intelligent internet of things according to claim 3, wherein the data calculation and analysis module further comprises a heat map acquisition module;
the heat map acquisition module is used for carrying out four-point method estimation on the corresponding coordinates of the image ground label acquired by the monitoring equipment and the market plane overlook image, projecting the bottom edge center of the target bounding box into the market plane overlook image through a homography matrix, and acquiring the projection points; generating a thermodynamic diagram of a two-dimensional gaussian distribution based on the projection points in the mall planar top view image; and obtaining the projection heat map of the market area according to time sequence statistics.
5. The marketing accurate screening pushing system based on the big data and the intelligent Internet of things according to claim 1, wherein the data calculation and analysis module further comprises an information filtering module;
the information filtering module is used for filtering the heat map through a maximum lattice point sampling method; and processing the projected heat map through a preset sliding window, wherein each processing only keeps the maximum value in the sliding window, and other values are zeroed.
6. The marketing accurate screening pushing system based on the big data and the intelligent internet of things according to claim 1, wherein the data calculation and analysis module further comprises a regional heat acquisition module;
the regional heat acquisition module is used for classifying the plan view of the mall to obtain regional images; calculating the area of any region in the region image, and calculating the region heat according to the quantity, the area and the discrete data point value of the Thiessen polygons contained in the region:
wherein, areaH i Representing the heat degree of the ith region in the region image, S i,j Represents the jth in the ith said regionThe area of the Thiessen polygon, S i Representing the area of the ith region in the region image, H i,j Discrete data point values representing the jth Thiessen polygon contained in the ith region, and n representing n regions divided in the region image.
7. The marketing accurate screening pushing system based on the big data and the intelligent internet of things according to claim 1, wherein the data calculation and analysis module further comprises a cluster analysis module;
and the cluster analysis module is used for generating a market heat map after obtaining heat at any position, dividing the market heat map into heat levels based on pixel values through a clustering algorithm, and generating the heat level characteristic map.
8. The marketing accurate screening pushing system based on the big data and the intelligent internet of things of claim 1, wherein the commodity pushing module further comprises a commodity pushing neural network module;
the commodity pushing neural network module is used for analyzing the customer information and the store regional characteristic matrix input into the network through the trained commodity pushing neural network and outputting the pushed commodity.
9. The marketing accurate screening pushing system based on the big data and the intelligent internet of things of claim 1, wherein the commodity pushing module further comprises an offline commodity pushing module;
the off-line commodity pushing module is used for counting the heat and the stay time of the area in the mall where the track information is located by analyzing the track information of the customer in the mall, and pushing the commodity through a mobile phone after the customer leaves the mall.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110019917.6A CN112598486B (en) | 2021-01-07 | 2021-01-07 | Marketing accurate screening push system based on big data and intelligent internet of things |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110019917.6A CN112598486B (en) | 2021-01-07 | 2021-01-07 | Marketing accurate screening push system based on big data and intelligent internet of things |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112598486A CN112598486A (en) | 2021-04-02 |
| CN112598486B true CN112598486B (en) | 2023-08-11 |
Family
ID=75207887
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110019917.6A Active CN112598486B (en) | 2021-01-07 | 2021-01-07 | Marketing accurate screening push system based on big data and intelligent internet of things |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112598486B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113344603A (en) * | 2021-04-16 | 2021-09-03 | 华蔚集团(广东)有限公司 | Intelligent marketing system for clients served after class |
| CN114240552A (en) * | 2021-12-15 | 2022-03-25 | 平安科技（深圳）有限公司 | Product recommendation method, device, equipment and medium based on deep clustering algorithm |
| CN115564472A (en) * | 2022-09-22 | 2023-01-03 | 内蒙古宏联网络科技有限责任公司 | Equipment interactive information service system and method based on big data |
| CN116629982B (en) * | 2023-07-24 | 2023-09-29 | 北京一凌宸飞科技有限公司 | Network marketing method based on big data analysis |
| CN116958908B (en) * | 2023-09-19 | 2023-12-15 | 坤前计算机有限公司 | Monitoring data processing method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016025321A1 (en) * | 2014-08-13 | 2016-02-18 | OneCloud Labs, Inc. | Replication of virtualized infrastructure within distributed computing environments |
| WO2017054041A1 (en) * | 2015-09-28 | 2017-04-06 | Beehaviour.Net Pty Ltd | Method, system and computer program for recording online browsing behaviour |
| CN109189949A (en) * | 2018-08-03 | 2019-01-11 | 江苏省城市规划设计研究院 | A kind of population distribution calculation method |
| CN111988734A (en) * | 2020-08-28 | 2020-11-24 | 北京思特奇信息技术股份有限公司 | Regional marketing method and system based on thermodynamic diagram |
| CN113344603A (en) * | 2021-04-16 | 2021-09-03 | 华蔚集团(广东)有限公司 | Intelligent marketing system for clients served after class |
| CN113506144A (en) * | 2021-09-03 | 2021-10-15 | 海门市创睿机械有限公司 | Clothing sales prediction method and system based on artificial intelligence and big data |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150339952A1 (en) * | 2014-05-24 | 2015-11-26 | Nirit Glazer | Method and system for using location services to teach concepts |
-
2021
- 2021-01-07 CN CN202110019917.6A patent/CN112598486B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016025321A1 (en) * | 2014-08-13 | 2016-02-18 | OneCloud Labs, Inc. | Replication of virtualized infrastructure within distributed computing environments |
| WO2017054041A1 (en) * | 2015-09-28 | 2017-04-06 | Beehaviour.Net Pty Ltd | Method, system and computer program for recording online browsing behaviour |
| CN109189949A (en) * | 2018-08-03 | 2019-01-11 | 江苏省城市规划设计研究院 | A kind of population distribution calculation method |
| CN111988734A (en) * | 2020-08-28 | 2020-11-24 | 北京思特奇信息技术股份有限公司 | Regional marketing method and system based on thermodynamic diagram |
| CN113344603A (en) * | 2021-04-16 | 2021-09-03 | 华蔚集团(广东)有限公司 | Intelligent marketing system for clients served after class |
| CN113506144A (en) * | 2021-09-03 | 2021-10-15 | 海门市创睿机械有限公司 | Clothing sales prediction method and system based on artificial intelligence and big data |
Non-Patent Citations (1)
| Title |
|---|
| 卷烟营销网建动态监管服务体系研究;贺东;《全国流通经济》;16-17 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112598486A (en) | 2021-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112598486B (en) | Marketing accurate screening push system based on big data and intelligent internet of things | |
| CN108549891B (en) | Multi-scale diffusion well-marked target detection method based on background Yu target priori | |
| US20200272902A1 (en) | Pedestrian attribute identification and positioning method and convolutional neural network system | |
| CA2734143C (en) | Method and apparatus for estimating body shape | |
| CN111784845B (en) | Virtual try-on method and device based on artificial intelligence, server and storage medium | |
| US11847800B2 (en) | Systems and methods for analysis of images of apparel in a clothing subscription platform | |
| CN108537166A (en) | The method and apparatus for determining shelf pageview and analyzing pageview | |
| Kobayashi et al. | Three-way auto-correlation approach to motion recognition | |
| CN106951870A (en) | The notable event intelligent detecting prewarning method of monitor video that active vision notes | |
| CN113240626A (en) | Neural network-based method for detecting and classifying concave-convex flaws of glass cover plate | |
| CN106023249A (en) | Moving object detection method based on local binary similarity pattern | |
| Jia et al. | Fabric defect inspection based on lattice segmentation and lattice templates | |
| CN111310531B (en) | Image classification method, device, computer equipment and storage medium | |
| CN112766145B (en) | Method and device for identifying dynamic facial expressions of artificial neural network | |
| Senyukova | Segmentation of blurred objects by classification of isolabel contours | |
| CN104718560A (en) | Image monitoring apparatus for estimating size of singleton, and method therefor | |
| CN116611875A (en) | Recommendation method and system for predicting new products based on customer characteristics and behaviors | |
| CN112232334B (en) | Intelligent commodity selling identification and detection method | |
| Chen et al. | Slender Flexible Object Segmentation Based on Object Correlation Module and Loss Function Optimization | |
| CN109034887A (en) | Method of adjustment, the device and system of item price | |
| CN115661703A (en) | Method for extracting shop signboard information based on deep learning | |
| CN115131274A (en) | High-resolution remote sensing image change detection method based on multi-temporal joint decomposition | |
| Zhang et al. | Semisupervised Change Detection Based on Bihierarchical Feature Aggregation and Extraction Network | |
| Sun et al. | Research on commodity image classification based on local oppugnant color vector angle pattern | |
| CN112242007B (en) | Virtual fitting method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |