A method, device and related medium for managing Internet of Things cards based on deep learning

Through deep learning technology, the image data of IoT cards is intelligently recognized and text word segmentation processed, and the classification regression tree algorithm is used to predict, which solves the problem of inefficient supervision of IoT cards and realizes the closed-loop process of automated supervision and risk monitoring. , improving regulatory efficiency and industry stability.

CN115880696BActive Publication Date: 2024-11-29E SURFING IOT CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202211709487.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-29
Publication Date
2024-11-29
Estimated Expiration
2042-12-29

AI Technical Summary

Technical Problem

It is difficult for existing technology to efficiently supervise the violations of IoT cards, especially in the face of massive commodities. Manual review is inefficient and expensive, making it difficult to effectively supervise IoT cards sold illegally, affecting industry stability and social security.

Method used

Using deep learning-based methods, the original image data is collected through image intelligent recognition technology, text data is extracted, and text word segmentation is processed using machine learning models. Classification regression tree algorithm is used for classification prediction, realizing automated management and risks of IoT cards. Supervision.

Benefits of technology

The closed-loop process of rapid and automated search, pre-warning, in-process detection, and post-process disposal of IoT cards has been realized, which has significantly improved the efficiency of risk supervision and improved the ability to supervise violations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115880696B_ABST
    Figure CN115880696B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, device and related medium for managing an Internet of Things card based on deep learning, including: collecting original image data; performing image recognition on the original image data through image intelligent recognition technology to extract text data from the original image data, and using the text data as sample data; performing text segmentation processing on the sample data using a machine learning model to obtain target data; using a classification regression tree algorithm to perform classification prediction on the target data; and managing the Internet of Things card according to the results of the classification prediction. The present invention collects original image data, processes the original image data to obtain text data, then performs word segmentation processing on the text data, performs classification prediction based on the results, and manages the Internet of Things card, so as to realize rapid and automated search for such commodities, achieve a closed-loop process of pre-warning, in-process detection, and post-process disposal tracking, and comprehensively improve the efficiency of risk supervision.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer software technology, and in particular to an Internet of Things card management method, device and related media based on deep learning. Background Art

[0002] Since IoT cards are very similar to ordinary mobile phone data cards, there are often "data cards" in the market that are difficult to distinguish between true and false. Most of these cards are sold by agents in the form of data cards in the secondary market. This behavior is a violation of regulations, that is, IoT cards cannot be sold on e-commerce platforms. In addition, this behavior will also have a series of risk problems, such as agents running away and losing contact or imperfect after-sales service, resulting in the inability to use the business normally, and personal property cannot be guaranteed during use due to the lack of formal procedures. In addition to the above-mentioned defects that affect individuals, it will also affect the overall stability of the industry and social security. For example, many criminals will use such anonymous cards to commit telephone fraud.

[0003] Currently, operators mainly use manual review to find illegal IoT card products on e-commerce platforms, but this method is inefficient and costly in the face of massive amounts of goods. Therefore, how to efficiently supervise IoT card behavior is a problem that technicians in this field need to solve. Summary of the invention

[0004] The embodiments of the present invention provide a method, apparatus, computer equipment and storage medium for managing an Internet of Things card based on deep learning, aiming to improve the efficiency of managing an Internet of Things card.

[0005] In a first aspect, an embodiment of the present invention provides an Internet of Things card management method based on deep learning, comprising:

[0006] Collecting raw image data;

[0007] Performing image recognition on the original image data by using an image intelligent recognition technology to extract text data from the original image data, and using the text data as sample data;

[0008] Using a machine learning model to perform text segmentation on the sample data to obtain target data;

[0009] Using a classification and regression tree algorithm to perform classification prediction on the target data;

[0010] The Internet of Things cards are managed according to the results of the classification prediction.

[0011] In a second aspect, an embodiment of the present invention provides an IoT card management device based on deep learning, including:

[0012] An information acquisition unit, used for acquiring original image data;

[0013] An image recognition unit, used to perform image recognition on the original image data by using an image intelligent recognition technology to extract text data from the original image data, and use the text data as sample data;

[0014] A word segmentation processing unit, used to perform text segmentation processing on the sample data using a machine learning model to obtain target data;

[0015] A classification prediction unit, used to perform classification prediction on the target data using a classification regression tree algorithm;

[0016] A result management unit is used to manage the Internet of Things card according to the result of the classification prediction.

[0017] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the deep learning-based Internet of Things card management method as described in the first aspect is implemented.

[0018] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the deep learning-based Internet of Things card management method as described in the first aspect is implemented.

[0019] The embodiment of the present invention discloses a method, device, computer equipment and storage medium for managing an Internet of Things card based on deep learning, including: collecting original image data; performing image recognition on the original image data through image intelligent recognition technology to extract text data from the original image data, and using the text data as sample data; performing text segmentation processing on the sample data using a machine learning model to obtain target data; using a classification regression tree algorithm to perform classification prediction on the target data; and managing the Internet of Things card according to the results of the classification prediction. The embodiment of the present invention collects the original image data of the Internet of Things card, processes the original image data to obtain text data, then performs word segmentation processing on the text data, performs classification prediction based on the results and manages the Internet of Things card, thereby constructing an online detection model for Internet of Things card products, so as to realize rapid and automated search for such products, and achieve a closed-loop process of pre-warning, in-process detection, and post-process disposal tracking, thereby comprehensively improving the efficiency of risk supervision for Internet of Things cards. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0021] Figure 1 A flowchart of an Internet of Things card management method based on deep learning provided by an embodiment of the present invention;

[0022] Figure 2 A schematic diagram of a sub-process of an Internet of Things card management method based on deep learning provided by an embodiment of the present invention;

[0023] Figure 3 A schematic block diagram of an Internet of Things card management device based on deep learning provided by an embodiment of the present invention;

[0024] Figure 4 A sub-schematic block diagram of an Internet of Things card management device based on deep learning provided by an embodiment of the present invention;

[0025] Figure 5 A word cloud diagram of commodity types and keywords in an Internet of Things card management method based on deep learning provided by an embodiment of the present invention;

[0026] Figure 6 A CART classification binary tree recognition flow chart in a deep learning-based Internet of Things card management method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0029] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0030] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0031] See below Figure 1 , Figure 1 A flowchart of an Internet of Things card management method based on deep learning provided in an embodiment of the present invention specifically includes: steps S101 to S105.

[0032] S101, collecting original image data;

[0033] S102, performing image recognition on the original image data by using an image intelligent recognition technology to extract text data from the original image data, and using the text data as sample data;

[0034] S103, performing text segmentation processing on the sample data using a machine learning model to obtain target data;

[0035] S104, using a classification and regression tree algorithm to perform classification prediction on the target data;

[0036] S105: Manage the Internet of Things cards according to the results of the classification prediction.

[0037] In this embodiment, the original image data is first collected, and then image recognition is performed on the original image data, and the text data therein is extracted as sample data based on the image recognition result, and then the sample data is machine-learned through the machine learning model to process the sample data into target data, and then the classification and regression tree algorithm is used to classify and predict the target data, so that the Internet of Things card can be supervised according to the classification prediction result. For example, when the classification prediction result is not an Internet of Things card, this information can be ignored, and when the classification prediction result is an Internet of Things card, it can be further tracked. In addition, it can be understood that the original image data collected in this embodiment can contain relevant information of the Internet of Things card, such as words such as traffic and monthly rent, or it can not contain relevant information of the Internet of Things card. Of course, the method provided in this embodiment can also be understood as a supervision model, and in order to improve the accuracy of the model, the original image data containing relevant information of the Internet of Things card can be used for model learning in the early construction process, and when the model is built, the specified image data can be classified and managed, and at this time, it is not necessary to care whether it contains relevant information of the Internet of Things card.

[0038] This embodiment collects original image data, processes the original image data to obtain text data, and then performs word segmentation on the text data. Classification prediction is performed based on the results and the Internet of Things cards are managed, thereby building an online detection model for Internet of Things card products to achieve rapid and automated search for such products, and to achieve a closed-loop process of pre-warning, in-process detection, and post-process tracking, thereby comprehensively improving risk supervision efficiency.

[0039] In a specific embodiment, when collecting raw image data, data related to telecom IoT traffic cards can be crawled from mainstream popular Internet websites, and the crawled data sources mainly include e-commerce, search engines, social media, and news information. Preferably, a timed collection strategy can also be set to periodically complete the collection of multiple data sources to achieve the update of collected raw data.

[0040] In one embodiment, the step S101 includes:

[0041] According to the following formula, the original image data is deblurred using a deconvolution-based image deblurring algorithm:

[0042] B=I*K+N

[0043] In the formula, B is the blurred original image data, I is the deblurred original image data, K is the blur kernel, N is the image noise, and * represents the convolution operation.

[0044] This embodiment uses an image deblurring algorithm to perform deblurring processing on the original image data, thereby improving the image resolution of the original image data to solve the problem of low resolution of network acquisition.

[0045] In a specific embodiment, the various prior constraints of the image blur kernel are first used to obtain the minimized energy function, and then the alternating minimization method is used to solve the minimized energy function. The energy function E(I,K)=-log(P(I,K|B)) is predefined, and the energy function is converted into the minimum energy function. Where f(N), γ(I), γ(K) are the energy terms of noise N, clear image I and blur kernel K respectively; λ1 and λ2 are weight coefficients.

[0046] Extract the fuzzy constraint conditions of the fuzzy edge area, where the constraint equations are:

[0047]

[0048] Solving this constraint, we can estimate the accurate blur kernel The original clear image is restored through the RL deconvolution algorithm, thereby improving the image resolution.

[0049] In one embodiment, the step S101 further includes:

[0050] According to the following formula, the maximum inter-class variance method is used to perform threshold segmentation on the original image data after deblurring:

[0051] g=ω0*(μ0-μ) 2 +ω1*(μ1-μ) 2

[0052] Where g represents the variance of the foreground image and the background image in the original image data, ω0 represents the ratio of the number of foreground points to the image, μ0 represents the average grayscale of the foreground image, ω1 represents the ratio of the number of background points to the image, μ1 represents the average grayscale of the background image, and μ represents the total average grayscale of the image;

[0053] Traverse the original image data according to the pixel range (0, 255), and select the segmentation threshold corresponding to the maximum variance;

[0054] The original image data is binarized according to the segmentation threshold.

[0055] The present embodiment adopts the maximum inter-class variance method to perform threshold segmentation on the original image data after the deblurring process, and traverses it to select the segmentation threshold when the variance is maximum. Here, when the variance g is maximum, the foreground and background are the most different. At this time, the grayscale T is the maximum threshold. According to the maximum variance, the key text area is identified to realize the binarization of the image. The maximum inter-class variance method is a method for automatically obtaining a threshold value that is suitable for a bimodal situation, also known as the Otsu method, referred to as Otsu. It is based on the grayscale characteristics of the image, and the image is divided into two parts, the background and the target. The larger the inter-class variance between the background and the target, the larger the difference between the two parts constituting the image. When part of the target is misclassified as the background or part of the background is misclassified as the target, the difference between the two parts will be reduced. Therefore, the segmentation that maximizes the inter-class variance means that the probability of misclassification is minimal.

[0056] In one embodiment, if Figure 2 As shown, the step S102 also includes: steps S201 to S204.

[0057] S201, using contourlet transform and shearlet transform to decompose the original image data respectively, and extracting statistical values ​​of the decomposed original image data as a first eigenvector;

[0058] S202, using a nuclear locality preserving projection algorithm to reduce the dimension of the first eigenvector to obtain a second eigenvector;

[0059] S203, calculating the fuzzy connectivity of the second feature vector by using a sample compactness-based fuzzy support vector machine;

[0060] S204: extracting the text data from the original image data using image recognition technology according to the fuzzy connectivity.

[0061] In this embodiment, in order to solve the problem of low accuracy of structural data extraction without network images, the image is decomposed by contourlet transform and shearlet transform respectively, and the statistical values ​​after decomposition are extracted as the first eigenvector, and the first eigenvector is reduced to a second eigenvector by using the nuclear locality preserving projection algorithm. The fuzzy connectivity is then calculated for the reduced second eigenvector by using a fuzzy support vector machine based on sample compactness, and finally the text data is extracted from the original image data according to the fuzzy connectivity.

[0062] Contourlet transform is a two-dimensional image representation method with properties such as multi-resolution, local positioning, multi-directionality, nearest neighbor sampling and anisotropy. Its basis functions are distributed in multiple scales and directions. A small number of coefficients can effectively capture the edge contours in the image, which is the main feature of natural images. The basic idea of ​​Contourlet transform is to first capture edge singular points with a multi-scale decomposition similar to wavelets, and then gather singular points with similar positions into contour segments according to directional information. The Laplacian filter structure (LP) is used to decompose the image multi-resolution to capture singular points. LP decomposition first generates a low-pass sampling approximation of the original signal and a difference image between the original image and the low-pass predicted image. The obtained low-pass image is further decomposed to obtain the low-pass image and difference image of the next layer, and the multi-resolution decomposition of the image is obtained by filtering step by step. The two-dimensional directional filter bank (DFB) is applied to each level of high-frequency components obtained by LP decomposition, and 2 to the power of n directional subbands can be decomposed at any scale. Each time the image is decomposed by the LP subband, the high-pass subband generated is input into the DFB, gradually connecting the singular points into a linear structure, thereby capturing the contours in the image. LP and DFB are combined to form a two-layer filter bank structure, called a pyramid directional filter bank (PDFB). Since this transform approximates the original image with basis functions in the form of contour segments, it is also called discrete Contourlet transform. The advantage of DFB is that it performs better for high-frequency parts, and each layer of LP decomposition decomposes the image into high-frequency parts and low-frequency parts. Contourlet transform combines the common advantages of the two well.

[0063] Shear transformation is one of the linear transformations in space and a primitive transformation of affine transformation. It refers to the process similar to the instability of quadrilaterals, where a square becomes a parallelogram and any side can be elongated.

[0064] The kernel discriminant locality preserving projection algorithm (KDLPP) maps samples to high-dimensional space through the kernel technique, and effectively combines the local manifold structure of samples and the discriminant information of samples in high-dimensional space to construct a new objective function. Its advantage is that on the basis of maintaining the manifold structure of samples, it makes full use of the category information of samples and uses the kernel method to extract the nonlinear characteristics of samples.

[0065] In a specific embodiment, when calculating the fuzzy connectivity of the second feature vector by using the sample compactness-based fuzzy support vector machine, the distance measurement characteristics of the sample categories in the Euclidean space are considered, and the set (E n ,d E ), where E n ={(x1,…x n ):x i∈R}, by defining the fuzzy connectivity Improve the distinguishability of samples, thereby improving the recognition rate of text features.

[0066] In one embodiment, the step S103 includes:

[0067] Performing semantic analysis on the sample data to obtain a corresponding sample character string;

[0068] Get the product characteristics and product types corresponding to the IoT card;

[0069] Calculating the relevance between each character in the sample character string and the product characteristics;

[0070] Screen out characters whose relevance is lower than a preset relevance threshold, and use the remaining characters as target characters;

[0071] Performing data cleaning on the target characters; wherein the data cleaning includes default value processing, data deduplication, and data filtering;

[0072] A keyword word cloud is drawn according to the commodity type, and key characters are extracted from the target characters using the keyword word cloud as the target data.

[0073] In this embodiment, semantic analysis is performed on the sample data to extract the sample character string corresponding to the sample data, which is then associated with the commodity characteristic character string of the IoT card, and characters with low association, such as punctuation marks, modal particles, prepositions, conjunctions, etc., are screened out. The remaining characters are used as target characters, and data cleaning is performed on them. Then, key characters are extracted through the keyword word cloud as target data for subsequent classification prediction.

[0074] In a specific embodiment, if Figure 5 As shown, according to the product type classification (such as IoT card, mobile phone card, IoT device, etc.), the keyword word cloud corresponding to the product type is drawn for subsequent extraction of key characters.

[0075] In one embodiment, the step S104 includes:

[0076] Dividing the target data into first sub-target data D1 and second sub-target data D2;

[0077] The Gini coefficient of the target data is calculated according to the following formula:

[0078]

[0079] In the formula, D represents the target data, GiniIndex represents the Gini coefficient, A represents each target feature in the target data, and a represents the value of each target feature;

[0080] Select the target data with the smallest Gini coefficient as the optimal feature, and the value corresponding to the target data with the smallest Gini coefficient as the optimal switching point;

[0081] According to the following formula, a CART classification binary tree is generated according to the optimal features and the optimal switching points:

[0082] D1=(x, y)∈D|A(x)=a, D2=D-D1

[0083] In the formula, x and y represent the left node and the right node respectively;

[0084] The CART classification binary tree is used to output the classification prediction result of the target data.

[0085] In this embodiment, the Gini coefficient of the target data is calculated to determine the optimal features and the optimal entry point, and a CART classification binary tree is generated based on the optimal features and the optimal entry point, and the CART classification binary tree outputs the classification prediction results of the target data.

[0086] In a specific embodiment, according to Figure 6 The CART classification binary tree recognition process shown outputs the classification prediction results of the target data, identifies the target data based on whether the main title of the product contains non-restrictive traffic keywords, whether the subtitle of the product contains free monthly rent keywords, whether the store type is a specialty store, etc., and divides the target data into different output types.

[0087] In one embodiment, the step S104 further includes:

[0088] Based on 10-fold cross validation, the CART classification binary tree was evaluated according to precision, recall and comprehensive index, where comprehensive index = 2*precision*recall / (precision+recall).

[0089] In the present embodiment, the CART classification binary tree is evaluated based on 10-fold cross validation, and feedback is given according to the application of the recognition result, so as to optimize the CART classification binary tree and improve the classification prediction accuracy. 10-fold cross validation (10-fold cross-validation) is used to test the accuracy of the algorithm and is a commonly used test method. 10-fold cross validation divides the data set into ten parts, and 9 of them are used as training data and 1 part as test data in turn for testing. Each test will result in a corresponding accuracy (or error rate). The average value of the accuracy (or error rate) of the results of 10 times is used as an estimate of the accuracy of the algorithm. Generally, multiple 10-fold cross validations (such as 10 10-fold cross validations) are required, and then the average is calculated as an estimate of the accuracy of the algorithm. The reason why the 10-fold cross validation chooses to divide the data set into 10 parts is that a large number of experiments using a large number of data sets and different learning techniques show that 10 folds are the appropriate choice to obtain the best error estimate.

[0090] Figure 3 A schematic block diagram of an Internet of Things card management device 300 based on deep learning provided by an embodiment of the present invention, the device 300 includes:

[0091] An information acquisition unit 301 is used to acquire original image data;

[0092] An image recognition unit 302 is used to perform image recognition on the original image data by using an image intelligent recognition technology to extract text data from the original image data and use the text data as sample data;

[0093] A word segmentation processing unit 303 is used to perform text segmentation processing on the sample data using a machine learning model to obtain target data;

[0094] A classification prediction unit 304 is used to perform classification prediction on the target data using a classification regression tree algorithm;

[0095] The result management unit 305 is used to manage the Internet of Things cards according to the results of the classification prediction.

[0096] In one embodiment, the information collection unit 301 includes:

[0097] The deblurring processing unit is used to perform deblurring processing on the original image data based on a deconvolution image deblurring algorithm according to the following formula:

[0098] B=I*K+N

[0099] In the formula, B is the blurred original image data, I is the deblurred original image data, K is the blur kernel, N is the image noise, and * represents the convolution operation.

[0100] In one embodiment, the Internet of Things card management device 300 based on deep learning further includes:

[0101] The threshold segmentation unit is used to perform threshold segmentation on the original image data after deblurring processing by using the maximum inter-class variance method according to the following formula:

[0102] g=ω0*(μ0-μ) 2 +ω1*(μ1-μ) 2

[0103] Where g represents the variance of the foreground image and the background image in the original image data, ω0 represents the ratio of the number of foreground points to the image, μ0 represents the average grayscale of the foreground image, ω1 represents the ratio of the number of background points to the image, μ1 represents the average grayscale of the background image, and μ represents the total average grayscale of the image;

[0104] A pixel traversal unit, used to traverse the original image data according to the pixel range (0, 255), and select a segmentation threshold corresponding to the maximum variance;

[0105] The binarization processing unit is used to perform binarization processing on the original image data according to the segmentation threshold.

[0106] In one embodiment, if Figure 4 The image recognition unit 302 includes:

[0107] An image decomposition unit 401 is used to decompose the original image data by using contourlet transform and shearlet transform respectively, and extract the statistical value of the decomposed original image data as a first eigenvector;

[0108] A vector dimension reduction unit 402, configured to reduce the dimension of the first eigenvector by using a nuclear locality preserving projection algorithm to obtain a second eigenvector;

[0109] A fuzzy connectivity calculation unit 403, configured to calculate the fuzzy connectivity of the second feature vector by using a sample compactness-based fuzzy support vector machine;

[0110] The data extraction unit 404 is used to extract the text data from the original image data using image recognition technology according to the fuzzy connectivity.

[0111] In one embodiment, the word segmentation processing unit 303 includes:

[0112] A semantic analysis unit, used to perform semantic analysis on the sample data to obtain a corresponding sample character string;

[0113] A characteristic type acquisition unit, used to acquire commodity characteristics and commodity types corresponding to the IoT card;

[0114] A correlation calculation unit, used to calculate the correlation between each character in the sample character string and the commodity characteristic;

[0115] A character screening unit, used to screen out characters whose relevance is lower than a preset relevance threshold, and use the remaining characters as target characters;

[0116] A data cleaning unit, used for performing data cleaning on the target character; wherein the data cleaning includes default value processing, data deduplication, and data filtering;

[0117] The word cloud drawing unit is used to draw a keyword word cloud according to the commodity type, and use the keyword word cloud to extract key characters from the target characters and use them as the target data.

[0118] In one embodiment, the classification prediction unit 404 includes:

[0119] A data division unit, used for dividing the target data into first sub-target data D1 and second sub-target data D2;

[0120] The coefficient calculation unit is used to calculate the Gini coefficient of the target data according to the following formula:

[0121]

[0122] In the formula, D represents the target data, GiniIndex represents the Gini coefficient, A represents each target feature in the target data, and a represents the value of each target feature;

[0123] A data selection unit, used to select the target data with the smallest Gini coefficient as the optimal feature, and to use the value corresponding to the target data with the smallest Gini coefficient as the optimal switching point;

[0124] The binary tree generation unit is used to generate a CART classification binary tree according to the optimal feature and the optimal switching point according to the following formula:

[0125] D1=(x, y)∈D|A(x)=a, D2=D-D1

[0126] In the formula, x and y represent the left node and the right node respectively;

[0127] A result output unit is used to output the classification prediction result of the target data using the CART classification binary tree.

[0128] In one embodiment, the classification prediction unit 404 further includes:

[0129] The comprehensive evaluation unit is used to evaluate the CART classification binary tree based on 10-fold cross validation according to the precision rate, recall rate and comprehensive index, wherein the comprehensive index = 2*precision rate*recall rate / (precision rate+recall rate).

[0130] Since the embodiments of the apparatus part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the apparatus part, which will not be repeated here.

[0131] The embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed, the steps provided in the above embodiment can be implemented. The storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0132] The embodiment of the present invention also provides a computer device, which may include a memory and a processor, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the steps provided in the above embodiment may be implemented. Of course, the computer device may also include various network interfaces, power supplies and other components.

[0133] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of this application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of this application.

[0134] It should also be noted that, in this specification, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprising a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for managing an Internet of Things card based on deep learning, characterized in that: include: Collecting raw image data; Performing image recognition on the original image data by using an image intelligent recognition technology to extract text data from the original image data, and using the text data as sample data; Using a machine learning model to perform text segmentation on the sample data to obtain target data; Using a classification and regression tree algorithm to perform classification prediction on the target data; Managing the Internet of Things cards according to the results of the classification prediction; The performing image recognition on the original image data by using the image intelligent recognition technology to extract text data from the original image data and using the text data as sample data includes: Decomposing the original image data by using contourlet transform and shearlet transform respectively, and extracting the statistical value of the decomposed original image data as a first eigenvector; Using a nuclear locality preserving projection algorithm to reduce the dimension of the first eigenvector to obtain a second eigenvector; Calculating fuzzy connectivity for the second feature vector by using a sample compactness-based fuzzy support vector machine; According to the fuzzy connectivity, extracting the text data from the original image data using image recognition technology; The method of performing text segmentation processing on the sample data using a machine learning model to obtain target data includes: Performing semantic analysis on the sample data to obtain a corresponding sample character string; Get the product characteristics and product types corresponding to the IoT card; Calculating the relevance between each character in the sample character string and the product characteristics; Screen out characters whose relevance is lower than a preset relevance threshold, and use the remaining characters as target characters; Performing data cleaning on the target characters; wherein the data cleaning includes default value processing, data deduplication, and data filtering; Draw a keyword cloud according to the commodity type, and use the keyword cloud to extract key characters from the target characters as the target data; The method of using a classification and regression tree algorithm to perform classification prediction on the target data includes: Dividing the target data into first sub-target data D1 and second sub-target data D2; The Gini coefficient of the target data is calculated according to the following formula: In the formula, D represents the target data, GiniIndex represents the Gini coefficient, A represents each target feature in the target data, and a represents the value of each target feature; Select the target data with the smallest Gini coefficient as the optimal feature, and the value corresponding to the target data with the smallest Gini coefficient as the optimal switching point; According to the following formula, a CART classification binary tree is generated according to the optimal features and the optimal switching points: D1=(x, y)∈D|A(x)=a, D2=D-D1 In the formula, x and y represent the left node and the right node respectively; The CART classification binary tree is used to output the classification prediction result of the target data.

2. According to the deep learning-based Internet of Things card management method of claim 1, it is characterized in that: The collecting of raw image data comprises: According to the following formula, the original image data is deblurred using a deconvolution-based image deblurring algorithm: B=I*K+N In the formula, B is the blurred original image data, I is the deblurred original image data, K is the blur kernel, N is the image noise, and * represents the convolution operation.

3. The method for managing an Internet of Things card based on deep learning according to claim 2, characterized in that: Also includes: According to the following formula, the maximum inter-class variance method is used to perform threshold segmentation on the original image data after deblurring: g=ω0*(μ0-μ) 2 +ω1*(μ1-μ) 2 Where g represents the variance of the foreground image and the background image in the original image data, ω0 represents the ratio of the number of foreground points to the image, μ0 represents the average grayscale of the foreground image, ω1 represents the ratio of the number of background points to the image, μ1 represents the average grayscale of the background image, and μ represents the total average grayscale of the image; Traverse the original image data according to the pixel range (0, 255), and select the segmentation threshold corresponding to the maximum variance; The original image data is binarized according to the segmentation threshold.

4. According to the deep learning-based Internet of Things card management method of claim 1, it is characterized in that: Also includes: Based on 10-fold cross validation, the CART classification binary tree was evaluated according to precision, recall and comprehensive index, where comprehensive index = 2*precision*recall / (precision+recall).

5. The IoT card management device based on deep learning is characterized by: include: An information acquisition unit, used for acquiring original image data; An image recognition unit, used to perform image recognition on the original image data by using an image intelligent recognition technology to extract text data from the original image data, and use the text data as sample data; A word segmentation processing unit, used to perform text segmentation processing on the sample data using a machine learning model to obtain target data; A classification prediction unit, used to perform classification prediction on the target data using a classification regression tree algorithm; A result management unit, used to manage the Internet of Things card according to the result of the classification prediction; The image recognition unit comprises: An image decomposition unit, used to decompose the original image data by using contourlet transform and shearlet transform, respectively, and extract statistical values ​​of the decomposed original image data as a first eigenvector; A vector dimension reduction unit, used for reducing the dimension of the first eigenvector by using a nuclear locality preserving projection algorithm to obtain a second eigenvector; A fuzzy connectivity calculation unit, configured to calculate the fuzzy connectivity of the second feature vector by using a sample compactness-based fuzzy support vector machine; A data extraction unit, configured to extract the text data from the original image data using image recognition technology according to the fuzzy connectivity; The word segmentation processing unit comprises: A semantic analysis unit, used to perform semantic analysis on the sample data to obtain a corresponding sample character string; A characteristic type acquisition unit, used to acquire commodity characteristics and commodity types corresponding to the IoT card; A correlation calculation unit, used to calculate the correlation between each character in the sample character string and the commodity characteristic; A character screening unit, used to screen out characters whose relevance is lower than a preset relevance threshold, and use the remaining characters as target characters; A data cleaning unit, used for performing data cleaning on the target character; wherein the data cleaning includes default value processing, data deduplication, and data filtering; A word cloud drawing unit, used for drawing a keyword word cloud according to the commodity type, and extracting key characters from the target characters using the keyword word cloud as the target data; The classification prediction unit comprises: A data division unit, used for dividing the target data into first sub-target data D1 and second sub-target data D2; The coefficient calculation unit is used to calculate the Gini coefficient of the target data according to the following formula: In the formula, D represents the target data, GiniIndex represents the Gini coefficient, A represents each target feature in the target data, and a represents the value of each target feature; A data selection unit, used to select the target data with the smallest Gini coefficient as the optimal feature, and to use the value corresponding to the target data with the smallest Gini coefficient as the optimal switching point; The binary tree generation unit is used to generate a CART classification binary tree according to the optimal feature and the optimal switching point according to the following formula: D1=(x, y)∈D|A(x)=a, D2=D-D1 In the formula, x and y represent the left node and the right node respectively; A result output unit is used to output the classification prediction result of the target data using the CART classification binary tree.

6. A computer device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the deep learning-based Internet of Things card management method as described in any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for managing an Internet of Things card based on deep learning as described in any one of claims 1 to 4 is implemented.

Citation Information

Patent Citations

  • Abnormal object recognition method and device and storage medium

    CN109522304A