A verification code character processing method and device and related equipment

Abstract

The embodiment of the application discloses a kind of verification code character processing method, device and related equipment, the method includes: by web crawler strategy and the business link of target business object that is grabbed from business platform, determine the target verification code picture to be verified;Target verification code picture is input to target character recognition model, and target character recognition model is identified in target verification code picture Target verification code character;Target verification code character is simulated to input to character input area, in response to the simulated submission operation for target verification code character, character verification is carried out to target verification code character, and character verification result is obtained;If character verification result indicates that verification succeeds, obtain the subject information of the operation subject to which target business object belongs from the operation voucher picture bound with business link, from operation voucher picture acquisition.Using the present application, the real-time and accuracy of character extraction can be improved, and then the acquisition efficiency of subject information can be improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus and related equipment for processing verification code characters. Background Technology

[0002] Currently, in some business scenarios (e.g., entity information retrieval scenarios), if a user (e.g., user A) needs to retrieve a certain entity information in this business scenario through a user terminal, the verification code can be automatically solved by a manual verification code solving platform on the business display interface currently displayed on the user terminal. Then, if the verification is successful, the user terminal can be allowed to load the entity information that user A needs to query.

[0003] For example, in the process of automatically solving CAPTCHAs using third-party human CAPTCHA solving platforms, it often relies on external personnel hired by the third party to assist in the manual recognition and input of CAPTCHAs. This allows for automated CAPTCHA solving in the current business scenario, enabling the collection of the necessary information from user A within a certain amount of human interaction time. It should be understood that during manual CAPTCHA recognition, if the CAPTCHA characters are distorted or pasted, the subjectivity of human recognition may lead to misidentification, thus reducing the accuracy of CAPTCHA recognition. Furthermore, when user A needs to retrieve a large amount of information in batches, it inevitably increases the time required for human interaction with the third party, thereby reducing the efficiency with which user A obtains this information. Summary of the Invention

[0004] This application provides a method, apparatus, and related equipment for processing verification code characters, which can improve the real-time performance and accuracy of character extraction, thereby improving the efficiency of obtaining main information.

[0005] One embodiment of this application provides a method for processing verification code characters, including:

[0006] The business links of target business objects on the business platform are crawled using web crawling strategies, and the target CAPTCHA image to be verified is determined based on the business links and the web crawling strategy.

[0007] Obtain the target character recognition model associated with the business platform, input the target verification code image into the target character recognition model, and have the target character recognition model recognize the target verification code image to obtain the target verification code characters in the target verification code image;

[0008] The target verification code characters are simulated and input into the character input area corresponding to the target verification code image. In response to the simulated submission operation of the target verification code characters in the character input area, the target verification code characters are validated and the character validation result is obtained.

[0009] If the character verification result indicates successful verification, the operation certificate image bound to the business link is obtained, and the main information of the operation entity to which the target business object belongs is collected from the operation certificate image.

[0010] One embodiment of this application provides a verification code character processing device, including:

[0011] The target image determination module uses web crawling strategies to crawl business links of target business objects on the business platform, and determines the target CAPTCHA image to be verified based on the business links and web crawling strategies.

[0012] The target character recognition module obtains the target character recognition model associated with the business platform, inputs the target verification code image into the target character recognition model, and the target character recognition model recognizes the target verification code image to obtain the target verification code character corresponding to the target verification code image;

[0013] The character submission module is used to simulate inputting the target verification code characters into the character input area corresponding to the target verification code image. In response to the simulated submission operation of the target verification code characters in the character input area, the module performs character verification on the target verification code characters and obtains the character verification result.

[0014] The main information collection module is used to obtain the operation voucher image bound to the business link if the character verification result indicates successful verification, and to collect the main information of the operation entity to which the target business object belongs from the operation voucher image.

[0015] The target image determination module includes:

[0016] The link crawling unit is used to obtain the web crawling strategy associated with the business platform, and to crawl the link information of N business objects from the business platform through the web crawling strategy; N is a positive integer.

[0017] The link determination unit is used to obtain the target business object from N business objects and use the link information of the target business object as the business link of the target business object.

[0018] The link parsing unit is used to parse business links, obtain the operation links of the operation entities to which the target business object belongs, and output the operation display interface of the operation entity based on the operation links; the operation display interface includes a query interface for querying the operation voucher images of the operation entity.

[0019] The verification interface output unit is used to respond to the simulated trigger operation of the query interface, output the simulated verification interface corresponding to the operation display interface, and obtain the target verification code image to be verified from the simulated verification interface through a web crawler strategy.

[0020] The verification interface output unit includes:

[0021] The interface trigger subunit is used to respond to simulated trigger operations for the query interface and switch the operation display interface to the simulated verification interface; the simulated verification interface includes a target display area; the target display area is used to display the target verification code image associated with the operation voucher image;

[0022] The first query subunit is used to query the verification image address that matches the first key element in the first image collection rule in the data structure tree corresponding to the simulated verification interface, based on the first image collection rule in the web crawler strategy.

[0023] The verification image extraction subunit is used to extract the target verification code image from the target display area based on the verification image address.

[0024] The target character recognition module includes:

[0025] The target image acquisition unit is used to acquire the target character recognition model associated with the business platform, perform noise reduction processing on the target verification code image, and use the noise-reduced target verification code image as the image to be processed.

[0026] The convolutional feature extraction unit is used to input the image to be processed into the convolutional neural network in the target character recognition model, and the convolutional neural network in the target character recognition model extracts the image convolutional features of the image to be processed, and uses the extracted image convolutional features as the target image convolutional features of the target verification code image.

[0027] The sequence feature extraction unit is used to input the convolutional features of the target image into the recurrent neural network in the target character recognition model, and the recurrent neural network in the target character recognition model extracts the sequence features from the convolutional features of the target image, and uses the extracted character sequence features as the target character sequence features corresponding to the target verification code image;

[0028] The sequence feature alignment unit is used to input the target character sequence features into the character classification network in the target character recognition model. The target character sequence features are aligned by the connection time classification network in the character classification network, and the target CAPTCHA characters in the target CAPTCHA image are obtained based on the aligned target character sequence features.

[0029] The target character recognition model includes a recurrent neural network, which is a bidirectional long short-term memory network. The bidirectional long short-term memory network includes a forward long short-term memory network and a backward long short-term memory network. The forward long short-term memory network includes a memory network B. i and memory network B i+1Memory Network B i+1 For memory network B i The next memory network; the inverse long short-term memory network includes memory network C. i+1 and memory network C i ; Memory Network C i+1 For memory network C i The previous memory network; i is a positive integer less than or equal to M; the number of memory networks in both the forward long short-term memory network and the reverse long short-term memory network is M;

[0030] The sequence feature extraction unit includes:

[0031] The forward feature extraction subunit is used to obtain the memory network B in the forward long short-term memory network. i Associated positive history hidden features h i-1 The convolutional features and positive history hidden features h of the target image are combined. i-1 Input memory network B i , by memory network B i The positive target hidden feature h is extracted at time i. i Hide features h of the positive target i Input memory network B with convolutional features of the target image i+1 , by memory network B i+1 The positive target hidden feature h is extracted at time i+1. i+1 ;

[0032] The inverse feature extraction subunit is used to obtain the memory network C in the inverse long short-term memory network. i+1 Associated reverse history hidden features k i+1 The convolutional features of the target image and the inverse history hidden features k i+1 Input memory network C i+1 , by memory network C i+1 The reverse target hidden feature k is extracted at time i+1. i Hide the reverse target feature k i The target image convolutional feature input memory network C i , by memory network C i The reverse target hidden feature k is extracted at time i. i-1 ;

[0033] Feature splicing subunits are used to integrate memory network B i The positive target hidden feature h extracted at time i i With memory network C i The reverse target hidden feature k extracted at time i i-1 Perform feature concatenation to obtain the first concatenated feature, and then use the memory network B.i+1 The positive target hidden feature h extracted at time i+1 i+1 With memory network C i+1 The reverse target hidden feature k extracted at time i+1 i Perform feature concatenation to obtain the second concatenated feature;

[0034] The sequence feature determination subunit is used to determine the target character sequence features corresponding to the target CAPTCHA image extracted from the convolutional features of the target image based on the first concatenation feature and the second concatenation feature.

[0035] The character submission module includes:

[0036] The anti-crawler unit is used to obtain the anti-crawler strategy for the target CAPTCHA characters, and to obtain the sleep duration and character input interval duration indicated by the anti-crawler strategy.

[0037] The character input unit is used to simulate inputting the target verification code characters into the character input area corresponding to the target verification code image according to the character input interval during the sleep period; the character input area includes a character input box and a character submission control.

[0038] The character submission unit is used to respond to a simulated submission operation to the character submission control, perform character verification on the target verification code characters displayed in the character input box, and obtain the character verification result.

[0039] The main information collection module includes:

[0040] The operation voucher display unit is used to output the voucher image display interface associated with the business platform if the character verification result indicates that the verification is successful. The operation voucher image bound to the business link is displayed in the voucher image display interface.

[0041] The voucher interface parsing unit is used to parse the voucher image display interface through a web crawler strategy associated with the business platform to obtain the operational voucher image in the voucher image display interface.

[0042] The optical model calling unit is used to use the operation voucher image as the image to be collected and call the optical character recognition model through the optical character recognition interface.

[0043] The main information collection unit is used to identify the image to be collected using an optical character recognition model, and to use the optical character information identified from the image to be collected as the main information of the operating entity to which the target business object belongs, collected from the operation certificate image.

[0044] The voucher interface parsing unit includes:

[0045] The second query subunit is used to search for the voucher image address that matches the second key element in the second image collection rule in the data structure tree of the voucher image display interface by using the second image collection rule in the web crawler strategy associated with the business platform.

[0046] The Operations Image Extraction Subunit is used to extract the operations voucher image from the voucher display area in the voucher image display interface based on the verification image address.

[0047] The device also includes:

[0048] The association establishment module is used to establish the association between entity information and business links in the entity information database associated with the operating entity, and update the association to the entity information database;

[0049] The business link update module is used to update the business links of the target business object into webpage update links on the business platform by using business links carrying related relationships as webpage update links.

[0050] The device also includes:

[0051] The illegal object detection module is used to identify the operating entity indicated by the main information as an illegal operating entity when the target business object under the webpage update link is found to be an illegal object in the blacklist, based on the association between the main information and the business link.

[0052] The notification information generation module is used to generate notification information associated with illegal operating entities and send the notification messages to the regulatory terminals corresponding to the platform supervisors associated with the business platform.

[0053] One embodiment of this application provides a method for processing verification code characters, including:

[0054] Obtain the original sample images used to train the initial character recognition model; the original sample images are determined based on sample links and web crawling strategies; the sample links are crawled from the business platform using web crawling strategies.

[0055] Obtain a data augmentation strategy for data augmentation of the original sample image, perform data augmentation processing on the original sample image based on the data augmentation strategy, and obtain at least one augmented sample image associated with the original sample image;

[0056] Use the original sample image and at least one enhanced sample image as the target sample image, and use the CAPTCHA characters corresponding to the original sample image as the sample label of the target sample image.

[0057] The target sample image is input into the initial character recognition model, which then recognizes the target sample image and uses the recognized CAPTCHA characters as the predicted label.

[0058] Based on predicted labels and sample labels, the initial character recognition model is iteratively trained to obtain a target character recognition model for recognizing target CAPTCHA characters in the target CAPTCHA image. The target CAPTCHA characters are used to obtain the operation voucher image bound to the business link associated with the target CAPTCHA image after character verification. The operation voucher image is used to collect the main information of the operating entity to which the target business object indicated by the business link belongs.

[0059] One embodiment of this application provides a verification code character processing device, including:

[0060] The original image acquisition module is used to acquire the original sample images for training the initial character recognition model. The original sample images are determined based on sample links and web crawling strategies. The sample links are crawled from the business platform using web crawling strategies.

[0061] An enhanced image generation module is used to obtain a data augmentation strategy for data augmentation of the original sample image, perform data augmentation processing on the original sample image based on the data augmentation strategy, and obtain at least one enhanced sample image associated with the original sample image.

[0062] The sample image determination module is used to use the original sample image and at least one enhanced sample image as the target sample image, and to use the CAPTCHA characters corresponding to the original sample image as the sample label of the target sample image.

[0063] The sample character recognition module is used to input the target sample image into the initial character recognition model, which then recognizes the target sample image and uses the sample verification code characters recognized from the target sample image as the predicted label.

[0064] The model training module is used to iteratively train the initial character recognition model based on the predicted labels and sample labels to obtain a target character recognition model for recognizing target CAPTCHA characters in the target CAPTCHA image. The target CAPTCHA characters are used to obtain the operation voucher image bound to the business link associated with the target CAPTCHA image after character verification. The operation voucher image is used to collect the main information of the operating entity to which the target business object indicated by the business link belongs.

[0065] One embodiment of this application provides a computer device, which includes a processor and a memory;

[0066] The processor is connected to a memory, wherein the memory is used to store computer programs, and the processor is used to invoke the computer programs so that the computer device performs the methods in any aspect of the embodiments of this application.

[0067] One aspect of this application provides a computer-readable storage medium storing a computer program adapted to be loaded and executed by a processor, such that a computer device having a processor performs the method in any aspect of this application.

[0068] One aspect of this application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method in any aspect of this application.

[0069] The computer device involved in this embodiment can intelligently extract target CAPTCHA characters from a target CAPTCHA image by introducing a target character recognition model, thereby improving the real-time performance and accuracy of character extraction. Furthermore, the computer device can simulate inputting the target CAPTCHA characters into the character input area corresponding to the target CAPTCHA image. In response to the simulated submission operation of the target CAPTCHA characters in the character input area, the computer device can perform character verification on the target CAPTCHA characters and obtain a character verification result. Further, if the character verification result indicates successful verification, the computer device can obtain the operating voucher image bound to the business link and collect the entity information of the operating entity to which the target business object belongs from the operating voucher image. Therefore, this embodiment can intelligently perform character verification on the identified target CAPTCHA characters, thereby quickly extracting business entity information (e.g., business license entity information) from the operating voucher image (e.g., business license image) upon successful verification, thus improving the efficiency of entity information acquisition. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a schematic diagram of a network architecture provided in an embodiment of this application;

[0072] Figure 2 This is a schematic diagram illustrating a scenario for intelligently recognizing verification code characters and intelligently collecting subject information, provided in an embodiment of this application.

[0073] Figure 3 This is a flowchart illustrating a verification code character processing method provided in an embodiment of this application;

[0074] Figure 4 This is a schematic diagram of a scenario for obtaining a target verification code image provided in an embodiment of this application;

[0075] Figure 5 This is a schematic diagram of a target character recognition model provided in an embodiment of this application;

[0076] Figure 6 This is a flowchart illustrating a verification code character processing method provided in an embodiment of this application;

[0077] Figure 7 This is a schematic diagram of a data collection scheme based on web crawlers provided in an embodiment of this application;

[0078] Figure 8 This application provides a method for processing verification code characters.

[0079] Figure 9 This is a schematic diagram illustrating a scenario in which an initial character recognition model is trained to obtain a target character recognition model, as provided in an embodiment of this application.

[0080] Figure 10 This is a schematic diagram illustrating a scenario where sample labels are obtained through an attention mechanism network, as provided in an embodiment of this application.

[0081] Figure 11 This is a schematic diagram of the structure of a verification code character processing device provided in an embodiment of this application;

[0082] Figure 12 This is a schematic diagram of the structure of a verification code character processing device provided in an embodiment of this application;

[0083] Figure 13 This is a schematic diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0084] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0085] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0086] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0087] The solutions provided in this application belong to the field of machine learning (ML) within artificial intelligence. Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and many other disciplines. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning typically includes techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0088] Please see Figure 1 , Figure 1 This is a schematic diagram of a network architecture provided in an embodiment of this application. Figure 1As shown, this network architecture is suitable for an e-commerce operation system, which may specifically include computer equipment 2000, a first business server 10a, a first business platform user terminal cluster, a second business server 10b, and a second business platform user terminal cluster. It can be understood that the first platform user terminal cluster can be a user terminal cluster under the first business platform, and the second platform user terminal cluster can be a user terminal cluster under the second business platform. It should be understood that both the first and second platform user terminal clusters may include one or more user terminals; there is no limit to the number of user terminals in the user terminal clusters corresponding to each business platform.

[0089] Among them, such as Figure 1 As shown, the first platform user terminal cluster can include multiple user terminals under the first service platform. Specifically, the multiple user terminals under the first service platform can include... Figure 1 The user terminals shown are 3000a, 3000b, 3000c, ..., 3000n. Figure 1 As shown, user terminals 3000a, 3000b, 3000c, ..., 3000n can connect to the first business server 10a via a network, enabling each user terminal in the first platform user terminal cluster to interact with the first business server 10a through this network connection. For example, each user terminal in the first platform user terminal cluster can interact with the first business server 10a through a first application client (e.g., shopping client A), so that when these user terminals access the first business server 10a through shopping client A, they can return the operational display interface of the operating entity (e.g., a virtual store) registered by these user terminals on the first business platform. It can be understood that each user terminal in the first platform user terminal cluster can include: smartphones, tablets, laptops, desktop computers, smart TVs, and other smart terminals with document loading and display functions.

[0090] Similarly, such as Figure 1 As shown, the second platform user terminal cluster can include multiple user terminals under the second service platform. Specifically, these multiple user terminals under the second service platform can include... Figure 1 The user terminals shown are 4000a, 4000b, 4000c, ..., 4000n. Figure 1As shown, user terminals 4000a, 4000b, 4000c, ..., 4000n can connect to the second business server 10b via a network, allowing each user terminal in the second platform user terminal cluster to interact with the second business server 10b through this network connection. For example, each user terminal in the second platform user terminal cluster can interact with the second business server 10b through a second application client (e.g., shopping client B), so that when these user terminals access the second business server 10b through the shopping client B, they can return the operational display interface of the operating entity (e.g., a virtual store) registered by these user terminals on the second business platform. It can be understood that each user terminal in the second platform user terminal cluster can include: smartphones, tablets, laptops, desktop computers, smart TVs, and other smart terminals with document loading and display functions.

[0091] like Figure 1 As shown, the first business server 10a corresponding to the first business platform and the second business server 10b corresponding to the second business platform can both be connected to the computer device 2000 via the network. In this way, the computer device 2000 can interact with the first business server 10a and the second business server 10b through the product clues (e.g., product types) indicated by the web crawler strategy, so as to crawl the link information of the business objects that match the product clues from the business servers corresponding to these business platforms. For example, the computer device 2000 can crawl the product links of products that match the specified product type from the first business platform and / or the second business platform respectively.

[0092] It is understood that the product links captured by the computer device 2000 may come from the same business platform or from different business platforms. For ease of understanding, this embodiment of the application takes the capture of the link information of a business object (e.g., product S1) from a certain business platform (e.g., the first business platform) as an example to illustrate the specific process of determining the target verification code image to be verified through the captured link information of this business object (e.g., product S1) and the web crawler strategy.

[0093] It should be understood that after acquiring the target verification code image, the computer device 2000 can further acquire a target character recognition model according to an intelligent data acquisition strategy. This model intelligently identifies the target verification code characters in the image, and then simulates inputting the identified characters into the character input area to submit the target verification code. At this point, the computer device 2000 can perform character verification on the submitted verification code. Upon successful verification, it can intelligently acquire an operating certificate image (e.g., a business license image) bound to the link information of this business object (e.g., product S1). From this operating certificate image (e.g., business license image), it can collect the main information of the operating entity (e.g., a virtual store) to which this business object (e.g., product S1) belongs. Since the entire data acquisition process (e.g., character extraction, character input, and character verification) requires no manual intervention, it can fundamentally reduce the time spent on manual interaction, thereby improving the efficiency of acquiring main information.

[0094] Among them, such as Figure 1 The computer device 2000 shown can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. It will not be limited here.

[0095] For further information, please refer to [link / reference]. Figure 2 , Figure 2 This is a schematic diagram illustrating a scenario where intelligent recognition of verification code characters and intelligent collection of subject information are provided in an embodiment of this application. For example... Figure 2 The business server 20b shown can be the first business server 10a corresponding to the first business platform mentioned above. For example... Figure 2 As shown, the first business platform here can be Figure 2 The business platform P is shown. It can be understood that the business server 20b can be used by registered merchant users (e.g., those accessing the platform through the aforementioned shopping client A). Figure 2 The business platform P shown provides online operation services (e.g., store registration service, product listing service) to enterprises, organizations, or individual users.

[0096] For example, when business users such as enterprises or individual users access the first business platform (e.g., through the aforementioned shopping client A) Figure 2When the business platform P is shown, it can provide store registration services for these merchant users. That is, these merchant users can register their stores through the provided store registration service to enter the first business platform (e.g., Figure 2 The business platform P shown above can be used to apply for the creation of a corresponding virtual store. At this time, the business server 20b can be used for the first business platform (e.g., ...). Figure 2 The virtual shops applied for by these merchant users on the business platform P are configured with corresponding shop links, so that one virtual shop corresponds to one shop link. In this way, these merchant users can also list products in the virtual shops corresponding to their respective shop links through the aforementioned product listing service, so that each of the S virtual products (i.e., one virtual object) listed in the corresponding virtual shop can be configured with a product link. Here, S is a positive integer.

[0097] Therefore, it can be seen that the first business platform (for example, Figure 2 The business platform P shown can display virtual products from different virtual stores, and the product links for each virtual product in each virtual store are unique. Thus, the first business platform (e.g., Figure 2 Other users (e.g., purchasing users) on the business platform P shown can accurately access virtual products in a virtual store through a unique product link.

[0098] It is understandable that, in e-commerce operation systems, to facilitate the use of the primary business platform (e.g., Figure 2 Product safety supervision is conducted on the virtual goods presented on the business platform P shown. This application embodiment proposes that product safety can be monitored using product leads (e.g., product type) indicated by a web crawler strategy from the first business platform (e.g., [missing information]). Figure 2 The business platform P shown retrieves product links for N virtual products. These N virtual products can be collectively referred to as N business objects, and their product links can be collectively referred to as the link information of the N business objects, where N is a positive integer.

[0099] For ease of understanding, this application embodiment selects the link information of one business object from the link information of N business objects as the business link of the target business object. The business link of the target business object can be... Figure 2 The business link L1 of the business object N1 shown is as follows. Figure 2As shown, computer device 20a can crawl the business link L1 of business object N1 from the business platform P using a web crawler strategy. Then, by parsing business link L1, it can obtain the store link of the virtual store to which business object N1 (i.e., the target business object) belongs. It is understood that in this embodiment, the store links of the virtual store obtained by parsing business link L1 can be collectively referred to as the operation links of the operating entity to which the target business object belongs. At this time, computer device 20a can obtain the operation display interface of the operating entity (e.g., the homepage of the virtual store) through the operation link. It should be noted that the operation display interface includes a query interface that can be used to query entity information. By simulating the triggering of this query interface, a simulated verification interface can be accessed. The target display area of ​​this simulated verification interface can be used to display information related to... Figure 2 The image shown is an operating certificate image 204a associated with a verification code image.

[0100] For ease of understanding, in this embodiment of the application, the verification code images displayed in the target display area of ​​the simulated verification interface can be collectively referred to as target verification code images. At this time, the computer device 20a can further collect the target verification code images from the simulated verification interface according to the first image collection rules indicated by the web crawler strategy. For example... Figure 2 As shown, the target CAPTCHA image obtained by the collection can be Figure 2 The verification code image 201a is shown. To ensure accurate and real-time recognition of the verification code characters 203a in the verification code image 201a, this application embodiment proposes an innovative method... Figure 2 The character recognition model 202a shown intelligently extracts the verification code characters from the verification code image 201a.

[0101] Among them, such as Figure 2 The character recognition model 202a shown may specifically include a first network and a second network. The first network specifically refers to a convolutional recurrent neural network (CRNN) composed of a convolutional neural network and a recurrent neural network. The convolutional neural network in the CRNN can be a CNN (Convolutional Neural Network), and the recurrent neural network in the CRNN can be a bidirectional long short-term memory network (e.g., Bi-LSTM). The second network specifically refers to a character classification network with character classification capabilities. For example, this character classification network may be a connectionist temporal classification network (e.g., CTC network) or an attention mechanism network.

[0102] For example, by introducing an end-to-end character recognition method based on CRNN and CTC networks, this application embodiment can intelligently identify the target verification code image (i.e., ...) during the entire data acquisition strategy execution process. Figure 2 The target verification code characters (e.g., in the verification code image 201a shown) are shown. Figure 2 The verification code character shown is 203a, which can specifically be the character "3nHf". Additionally, as... Figure 2 As shown, computer device 20a can also perform character verification on the verification code character 203a recognized by intelligent submission, and thus quickly obtain the verification code character 203a when the character verification is successful. Figure 2 The operating voucher image 204a shown is an example. It should be understood that this operating voucher image 204a can specifically be a photo of a business license bound to the aforementioned business link L1. At this time, the computer device can quickly and intelligently extract the entity information of the operating entity to which the aforementioned business object N1 belongs from this business license photo (i.e., operating voucher image 204a). The entity information of the operating entity to which business object N1 belongs can specifically be... Figure 2 The main information shown is 205a.

[0103] It should be understood that computer device 20a obtains the main information of the operating entity (i.e., Figure 2 In the case of the entity information 205a shown, further legality monitoring can be performed on entity information 205a to obtain legality monitoring results. If the legality monitoring result indicates that entity information 205a is not legal, it reflects that the operating entity is an illegal operating entity. This can help platform supervisors accurately identify illegal operating entities on the business platform in a timely manner, so that they can promptly punish the illegal operating entities and ensure the security and reliability of the entire e-commerce operation system. Therefore, the computer device 20a can not only intelligently and in real-time perform character recognition, but also help platform supervisors achieve centralized monitoring and management of operating entities on the business platform P.

[0104] Furthermore, it is understood that the computer device 20a can also add these target CAPTCHA images and the target CAPTCHA characters identified in these images to a priori knowledge base. This allows the added target CAPTCHA images to be retrieved from the priori knowledge base as original sample images, and the target CAPTCHA characters from these images to be retrieved as labeled CAPTCHA characters for annotating the original sample images. The obtained original sample images and labeled CAPTCHA characters can then be used as initial training sample information for further training of the character recognition model 202a (i.e., the target character recognition model). The computer device 20a can then use this initial training sample information and the augmented sample information fitted from the initial training sample information as target training sample information to train a new target character recognition model. The target training sample information here may include a small number of labeled original sample images and augmented sample images fitted during data augmentation. The labeled CAPTCHA characters indicated by these sample labels are the CAPTCHA characters used when data-annotating the aforementioned original sample images, such as the aforementioned target CAPTCHA characters.

[0105] For ease of understanding, in the embodiments of this application, the character recognition model before training can be referred to as the initial character recognition model, and the new character recognition model obtained after training the initial character recognition model can be referred to as the aforementioned target character recognition model.

[0106] Optionally, it should be understood that in a blockchain scenario, the computer device 20a can act as a blockchain node. When the blockchain node trains a new target character recognition model based on the aforementioned target training sample information, it can notify other consensus nodes in the blockchain network to reach a consensus on the model parameters of the newly generated target character recognition model. Upon reaching a consensus, the model parameters of the newly generated target character recognition model can be uploaded to the blockchain. This allows the computer device 20a to obtain the model parameters of the newly generated target character recognition model from the blockchain in real time when it receives a new target verification code image. Based on these updated model parameters, the computer device 20a can update the model parameters of the target character recognition model in the computer device 20a, and then recognize the verification code characters in the new target verification code image based on the updated target character recognition model.

[0107] The specific implementation method of computer device 20a obtaining a target character recognition model by training an initial character recognition model, and recognizing the target verification code characters in the target verification code image and collecting the main information in the operation voucher image through the target character recognition model can be found in the following. Figures 3-10 Description of the corresponding embodiments.

[0108] Further, please see Figure 3 , Figure 3 This is a flowchart illustrating a verification code character processing method provided in an embodiment of this application. It is understood that the method provided in this embodiment can be executed by a computer device, which includes, but is not limited to, a user terminal or business server running an illegal detection module. For ease of understanding, this embodiment uses a user terminal as an example to illustrate the specific process of character extraction using a trained target character recognition model on that user terminal. Figure 3 As shown, the method may include at least the following steps S101-S104:

[0109] Step S101: Use web crawling strategies to capture the business links of the target business objects on the business platform, and determine the target verification code image to be verified based on the business links and web crawling strategies.

[0110] Specifically, the computer device can acquire web crawling strategies associated with the business platform, and use these strategies to crawl link information of N business objects from the platform; N is a positive integer. Further, the computer device can select a target business object from these N objects, using its link information as its business link. Further, the computer device can parse the business link to obtain the operational link of the operating entity to which the target business object belongs, and then output the operating entity's operational display interface based on the operational link. The operational display interface includes a query interface for querying the operating entity's operational voucher image. Further, the computer device can respond to simulated trigger operations on the query interface, outputting a simulated verification interface corresponding to the operational display interface, and then use the web crawling strategy to obtain the target verification code image to be verified from the simulated verification interface.

[0111] The specific process by which the computer device obtains the target verification code image from the simulated verification interface can be described as follows: The computer device can respond to a simulated trigger operation on the query interface and switch the operation display interface to the simulated verification interface. Note that the simulated verification interface here can include a target display area; this target display area can be used to display the target verification code image associated with the operation voucher image; furthermore, the computer device can use the first image collection rule in the web crawler strategy to query the data structure tree corresponding to the simulated verification interface for a verification image address that matches the first key element in the first image collection rule, and then extract the target verification code image from the target display area based on the verification image address.

[0112] For further information, please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic diagram illustrating a scenario for obtaining a target verification code image, as provided in an embodiment of this application. Wherein, as... Figure 4 The business link L2 shown can be from Figure 4 The business platform P shown above (i.e., the aforementioned) Figure 2 Another business object (e.g.,) captured on the business platform P in the corresponding embodiment. Figure 4 The link information for the business object N2 shown. For example, as... Figure 4 As shown, business link L2 can be a product link that can directly point to the product details display interface of business object N2, that is, the product link can be "https: / / detail.abcd.com / item.htm?id=59437980...".

[0113] It should be understood that, since the business object N2 is Figure 4 The products in the virtual store D shown are, in other words, the business object N2 and... Figure 4 The virtual stores D (i.e., the operating entities) shown have a subordinate relationship. Therefore, when the computer device crawls the product link of business object N2 through the web crawler strategy, it can parse the product link (i.e., business link L2) according to the subordinate relationship, so as to obtain the operating link of the virtual store D to which business object N2 belongs from the product link (i.e., business link L2).

[0114] It should be understood that, such as Figure 4 As shown, the operational links here can be... Figure 4 The store link 41a is shown. At this time, the computer device can intelligently access the virtual store D through store link 41a to display the store homepage of virtual store D. For example... Figure 4 As shown, the homepage of the virtual store D can be Figure 4 The operation display interface shown is 400a.

[0115] Among them, such as Figure 4 The operational display interface 400a shown includes a query interface for querying operational certificate images of the operating entity. It can be understood that this query interface can be... Figure 4 The document shown is a credential viewing entry 42a for viewing enterprise qualifications; at this time, the computer device can respond to a simulated trigger operation for this credential viewing entry 42a and output... Figure 4 The simulated verification interface 400b shown can then be used to obtain the verification code image to be verified from the target display area 43a of the simulated verification interface 400b through a web crawler strategy.

[0116] What is understandable is that Figure 4The simulated verification interface 400b shown may include a target display area 43a, a character input area 44a, and a character submission area 45a. Among these, Figure 4 The target display area 43a shown can be used to display a target CAPTCHA image associated with the operating voucher image to be viewed. Figure 4 The character input area 44a shown can be used to simulate entering target CAPTCHA characters extracted from the target CAPTCHA image. Among them, Figure 4 The character submission area 45a shown can be used to simulate submitting characters that have already been entered in the character input area. It should be understood that the character input area 44a and the character submission area 45a can be displayed together in the same area or distributed in different areas; this will not be limited here.

[0117] It should be understood that, regarding the above Figure 4 For the business platform P shown (e.g., QQ Mall shopping platform), the position and code of the target verification code image displayed in the simulated verification interface 400b are relatively fixed. Therefore, as Figure 4 As shown, the computer device can identify the location area of ​​the target CAPTCHA image in the simulated verification interface 400b through the first image acquisition rule in the web crawler strategy. This location area can be... Figure 4 The target display area 43a shown can then be used to collect specific elements in a specific location area from the data structure tree corresponding to the simulated verification interface 400b through the first image collection rule in the web crawler strategy (e.g., a CAPTCHA image extraction rule, where the specific element extraction syntax indicated by the CAPTCHA image extraction rule can be XPath or a CSS selector).

[0118] For example, the computer device can, within the data structure tree (e.g., the XML document corresponding to the XPath path language) corresponding to the simulated verification interface 400b, collectively refer to the node path language corresponding to the location region (i.e., the aforementioned specific location region) as the first key element in the aforementioned first image acquisition rule. Then, it can query the data structure tree corresponding to the simulated verification interface for verification image address characters that match the first key element in the first image acquisition rule. It should be understood that the verification image address character here is used to uniquely point to the specific location region (i.e., the location region) in the simulated verification interface 400b. Figure 4 As shown in the target display area 43a), the computer device can accurately retrieve the information from the verification image address in real time. Figure 4 The target was captured in the display area 43a shown. Figure 4 The target CAPTCHA image shown.

[0119] For example, when the aforementioned business platform P is the Taobao shopping platform, the specific element (i.e., the first key element) collected through XPath path language can be div.imgCaptcha_img. In this case, the verification image address corresponding to this first key element can be: id="nc_1_imgCaptcha_img". Then, based on this verification image address, [the image can be retrieved from...]. Figure 4 Extracted from the target display area 43a shown Figure 4 The target CAPTCHA image shown.

[0120] For example, optionally, when the aforementioned business platform P is the Xiaohongshu shopping platform, the specific element (i.e., the first key element) collected through the XPath path language can be div.center-img. It should be understood that the business platform P here can include, but is not limited to, QQ Mall shopping platform, Taobao shopping platform, Xiaohongshu shopping platform, JD.com shopping platform, Yihaodian shopping platform, Vipshop shopping platform, Pinduoduo shopping platform, Juanpi.com shopping platform, game mall shopping platform, etc. Here, we will not limit the element type of the specific element (i.e., the first key element that each business platform needs to query) specified for each business platform.

[0121] What is understandable is that Figure 4 The online store operator shown in the ONI verification interface 400b can be the merchant user who registered the virtual store D on the business platform P. The merchant user's operating credentials information (e.g., a photo of the operating credentials) can specifically be as described above. Figure 2 The business license image uploaded by the merchant user in the corresponding embodiment to the quality supervision bureau (e.g., the State Administration for Industry and Commerce).

[0122] Step S102: Obtain the target character recognition model associated with the business platform, input the target verification code image into the target character recognition model, and have the target character recognition model recognize the target verification code image to obtain the target verification code characters in the target verification code image;

[0123] Specifically, the computer device can acquire a target character recognition model associated with the business platform, perform denoising processing on the target CAPTCHA image, and use the denoised target CAPTCHA image as the image to be processed. Further, the computer device can input the image to be processed into the convolutional neural network in the target character recognition model, where the convolutional neural network extracts the image convolutional features of the image to be processed, and uses the extracted image convolutional features as the target image convolutional features of the target CAPTCHA image. Further, the computer device can input the target image convolutional features into the recurrent neural network in the target character recognition model, where the recurrent neural network extracts the sequence features from the target image convolutional features, and uses the extracted character sequence features as the target character sequence features corresponding to the target CAPTCHA image. Further, the computer device can input the target character sequence features into the character classification network in the target character recognition model, where the connection time classification network in the character classification network aligns the target character sequence features, and the target CAPTCHA characters in the target CAPTCHA image are obtained based on the aligned target character sequence features.

[0124] Optionally, and understandably, the specific process by which a computer device identifies target CAPTCHA characters through a target character recognition model can also be described as follows: When the computer device acquires a target character recognition model associated with the business platform, it can comprehensively identify the target CAPTCHA characters in the target CAPTCHA image by combining the first network and the second network in the target character recognition model. It should be understood that the first network here can be the aforementioned convolutional recurrent neural network (i.e., the aforementioned CRNN network), which specifically includes the aforementioned convolutional neural network (e.g., the aforementioned CNN network) and recurrent neural network (e.g., the aforementioned Bi-LSTM). The second network here can specifically be the aforementioned character classification network; for example, the character classification network can include any one of the aforementioned connection-time classification network (e.g., the aforementioned CTC network) and the aforementioned attention mechanism network that has character classification functionality.

[0125] For ease of understanding, this example uses the connection time classification network (e.g., the CTC network) as the second network to illustrate the process of recognizing variable-length CAPTCHAs in a target CAPTCHA image using the CRNN network and the CTC network. Further details can be found in [link to documentation]. Figure 5 , Figure 5 This is a schematic diagram of a target character recognition model provided in an embodiment of this application. For example... Figure 5 As shown, the target verification code image extracted by the computer device through the above step S101 can be... Figure 5 The verification code image shown is 51a. Among them, Figure 5The network 52a shown can be the aforementioned CNN network, which can be used to extract image convolutional features from the CAPTCHA image 51a. For example, the image convolutional features extracted from the CAPTCHA image 51a by this network 52a can be... Figure 5 The image convolution features shown.

[0126] Optionally, to improve the reliability of feature extraction, the computer device can also pre-define the image based on the image denoising strategy. Figure 5 The CAPTCHA image 51a (target CAPTCHA image) shown is subjected to denoising processing, and the denoised CAPTCHA image 51a (i.e., the denoised target CAPTCHA image) can then be used as the image to be processed; furthermore, the computer device can input the image to be processed into... Figure 5 The network 52a shown (i.e., the CNN network mentioned above) extracts the image convolutional features of the image to be processed, and then the extracted image convolutional features (e.g., Figure 5 The image convolution feature shown in 53a) is used as the target image convolution feature of the target CAPTCHA image.

[0127] It should be understood that the image denoising strategy involved in the embodiments of this application may specifically include the binarization algorithm corresponding to the binarization operation and the grayscale algorithm corresponding to the grayscale operation. For example, the embodiments of this application, through... Figure 5 Binarization or grayscale conversion of the verification code image 51a can filter out noise interference such as background texture in the verification code image 51a.

[0128] It should be understood that, when the target character recognition model is a character recognition model composed of the aforementioned CRNN network and CTC network, in order to address the inherent limitations of the CTC network (e.g., the dimension of the output features of network 52c should be consistent with the dimension of the input features of network 52c), this application embodiment proposes to reduce the pooling layers of the CNN network in the CRNN network. For example, if the original CNN network had 3 pooling layers, to prevent the phenomenon of excessively rapid feature dimensionality reduction caused by too many pooling layers, this application embodiment proposes to retain a CNN network with fewer pooling layers, for example, a CNN network with a single pooling layer.

[0129] Optionally, it should be understood that if the target character recognition model is a character recognition model composed of the CRNN network and the attention mechanism network, then there is no need to reduce the pooling layers of the CNN network in the CRNN network. That is, in this case, the pooling layers of the CNN network in the CRNN network can be the three common pooling layers.

[0130] in, Figure 5The network 52b shown can be the recurrent neural network described above. For example, the recurrent neural network here mainly refers to a bidirectional long short-term memory network. Specifically, this bidirectional long short-term memory network can include a forward long short-term memory network and a backward long short-term memory network. For example... Figure 5 As shown, the forward long short-term memory network can be used to calculate the forward target hidden features of the convolutional image feature 53a at each time step along the first feature calculation direction from left to right; for example, the forward hidden features extracted by the computer device at any two adjacent time steps (e.g., time step i and time step i+1) may include: the forward target hidden feature h at time step i. i and the positive target hidden feature h at time i+1 i+1 It should be understood that, since this forward long short-term memory network is essentially a recurrent neural network, the forward hidden features (e.g., forward target hidden features h) extracted by the computer device at time i are... i Essentially, this can serve as the input feature for the next time step (i.e., time step i+1) after time step i. Therefore, the positive hidden features extracted by the computer device at time step i (e.g., the positive target hidden feature h) i Essentially, the extracted positive historical hidden features h i-1 This is jointly determined by the image convolutional features 53a. Similarly, the positive hidden features extracted by the computer device at time i+1 (e.g., the positive target hidden feature h) i+1 Essentially, it is the positive target hidden feature h extracted from the time step preceding time step i (i.e., time step i) at time step i+1. i This is jointly determined by the image convolution feature 53a. It should be understood that this forward long short-term memory network can contain M memory networks, and any two adjacent memory networks among these M memory networks can contain memory network B. i and memory network B i+1 Furthermore, in the aforementioned first feature calculation direction, the memory network B i+1 For memory network B i The next memory network. The memory network used by the computer device at time i can be memory network B. i Similarly, the memory network used by this computer device at time i can be memory network B. i+1 .

[0131] Similarly, such as Figure 5As shown, the inverse long short-term memory network can be used to calculate the inverse target hidden features of the convolutional image feature 53a at each time step along the second feature calculation direction from right to left. For example, the inverse hidden features extracted by the computer device at any two adjacent time steps (e.g., time step i and time step i+1) may include: the inverse hidden feature k at time step i+1. i And the reverse hidden feature k at time i i-1 It should be understood that, since this inverse long short-term memory network is essentially a recurrent neural network, the inverse hidden features (e.g., inverse target hidden features k) extracted by the computer device at time i are... i-1 Essentially, it is the inverse target hidden feature k extracted from the next time step after time i (i.e., time i+1). i This is jointly determined by the image convolutional features 53a. Similarly, the inverse hidden features (e.g., inverse target hidden features k) extracted by the computer device at time i+1 are also considered. i Essentially, it is the reverse history hidden feature k extracted from the next time step after time i+1 (i.e., time i+2). i+1 This is jointly determined by the image convolution feature 53a. It should be understood that this inverse long short-term memory network can also contain M memory networks, and in this inverse long short-term memory network, any two adjacent memory networks among the M memory networks can contain memory network C. i and memory network C i+1 Furthermore, in the aforementioned second feature calculation direction, the memory network C i+1 For memory network C i The previous memory network. The memory network used by this computer device at time i can be memory network C. i Similarly, the memory network used by this computer device at time i can be a memory network C. i+1 .

[0132] Therefore, it can be seen that computer devices can acquire memory network B in the aforementioned positive long short-term memory network. i Associated positive history hidden features h i-1 At that time, the target image can be convolved with features (e.g., Figure 5 The image convolutional features shown) and the positive history hiding features h i-1 Input memory network B i , by memory network B i At time i, the aforementioned positive target hidden feature h is extracted. i Furthermore, computer devices can hide features h of forward-facing targets. i Input memory network B with convolutional features of the target image i+1 (i.e., memory network B)i The next memory network), and by that memory network B i+1 The above positive target hidden feature h is extracted at time i+1. i+1 .

[0133] Similarly, computer devices can also acquire memory network C in the aforementioned reverse long short-term memory network. i+1 Associated reverse history hidden features k i+1 At that time, the target image features are convolved (e.g., Figure 5 The image convolutional features shown) and the inverse history hidden features k i+1 Input memory network C i+1 , by memory network C i+1 The reverse target hidden feature k is extracted at time i+1. i Furthermore, computer equipment can hide the reverse target features k i The target image convolutional feature input memory network C i , by memory network C i The reverse target hidden feature k is extracted at time i. i-1 .

[0134] Furthermore, such as Figure 5 As shown, computer devices can use memory network B i The positive hidden features at time i (e.g., the positive target hidden features h mentioned above) i ) and memory network C i The inverse hidden features at time i (e.g., the inverse target hidden features k) i-1 The features are spliced ​​together to obtain the spliced ​​features at time i. It should be understood that, in the embodiments of this application, the spliced ​​features obtained at time i can be collectively referred to as the first spliced ​​features. Similarly, the computer device can use the memory network B i+1 The positive hidden features at time i+1 (e.g., the positive target hidden features h mentioned above) i+1 ) and memory network C i+1 The inverse hidden features at time i+1 (e.g., the inverse target hidden features k) i The features are spliced ​​together to obtain the spliced ​​features at time i+1. It should be understood that the spliced ​​features obtained at time i+1 can be collectively referred to as the second spliced ​​features in the embodiments of this application.

[0135] like Figure 5 As shown, a computer device can determine the target character sequence features corresponding to the target CAPTCHA image extracted from the convolutional features of the target image based on the first and second concatenation features. For example, as... Figure 5As shown, the computer device can use the first stitched features as new convolutional image features, and input these new convolutional image features into the aforementioned memory network B. i and memory network C i Similarly, the computer device can use the second stitched feature as another new convolutional image feature, and input this new convolutional image feature into the aforementioned memory network B. i+1 and memory network C i+1 Among them, memory network B i and memory network C i For a detailed explanation of how to extract features from new convolutional image features, please refer to the above explanation of memory network B. i and memory network C i The description of that will not be repeated here. Furthermore, the memory network B... i+1 and memory network C i+1 For a detailed implementation of feature extraction for another new convolutional image feature, please refer to the above description of the memory network B. i+1 and memory network C i+1 The description will not be repeated here.

[0136] like Figure 5 As shown, computer equipment can be accessed via... Figure 5 The network 52b shown extracts the convolutional features of the target image (i.e., Figure 5 The sequence features in the convolutional image features 53b) shown can be used as the target character sequence features corresponding to the target CAPTCHA image. These target character sequence features can be... Figure 5 The sequence feature 53b shown has a character feature dimension (i.e., the length of characters extracted from the target CAPTCHA characters) of 10. These 10 characters can specifically be... Figure 10 The string shown is " <space> 3 <space>nn <space>HH <space>f". Further, the computer device inputs this sequence feature 53b to... Figure 5 The network 52c shown can be the connection time classification network in the character classification network described above. In this case, the computer device can use the connection time classification network in the character classification network to align the sequence feature 53b, and then obtain the desired result based on the aligned sequence feature 53b. Figure 5 The verification code character 51b in the verification code image 51a shown is used to facilitate the subsequent execution of step S103. The character " <space>"Can be used to represent the separator "-".

[0137] Step S103: Simulate input of the target verification code character into the character input area corresponding to the target verification code image. In response to the simulated submission operation of the target verification code character in the character input area, perform character verification on the target verification code character and obtain the character verification result.

[0138] Specifically, to improve the success rate of data collection, the computer device can acquire anti-crawler strategies targeting the target CAPTCHA character, and then acquire the sleep duration (e.g., 1-2 seconds) and character input interval duration (e.g., random time intervals when simulating user character input) indicated by the anti-crawler strategy; further, the computer device can simulate inputting the target CAPTCHA character into the character input area corresponding to the target CAPTCHA image within the sleep duration and according to the character input interval duration; further, the computer device can respond to the simulated submission operation of the character submission control, perform character verification on the target CAPTCHA character displayed in the character input box, and obtain the character verification result.

[0139] It can be understood that the character input area here can specifically include a character input box and a character submit control; the character input box can be distributed in the above-mentioned... Figure 4 The character input area 44a in the corresponding embodiment, where the character submission controls can be distributed in the above-mentioned... Figure 4 Within the character input area 44a in the corresponding embodiment, the elements can also be independently distributed as described above. Figure 4 Within the character submission area 45a in the corresponding embodiment. The specific location of the character submission control will not be limited here.

[0140] It should be understood that after the computer device simulates a user submitting the target CAPTCHA characters, it can perform character verification on the target CAPTCHA characters displayed in the character input box. For example, the computer device can compare the character features of the real CAPTCHA characters in the target verification image with the currently recognized target CAPTCHA characters. If the similarity between the two meets the character verification conditions, a first character feature comparison result can be obtained. This first character feature comparison result can be used to indirectly reflect that the currently recognized target CAPTCHA characters have high accuracy. Conversely, if the similarity between the two does not meet the character verification conditions, the computer device can obtain a second character feature comparison result. This second character feature comparison result can be used to intelligently correct the currently recognized target CAPTCHA characters.

[0141] It should be understood that, in the embodiments of this application, the above-mentioned first character feature comparison result or second character feature comparison result can be collectively referred to as the character verification result obtained after character verification of the target verification code character. Therefore, if the character verification result is the first character feature comparison result, the following step S104 can be further executed.

[0142] Step S104: If the character verification result indicates successful verification, obtain the operation voucher image bound to the business link, and collect the main information of the operation entity to which the target business object belongs from the operation voucher image.

[0143] Specifically, if the character verification result indicates successful verification, the computer device can output a voucher image display interface associated with the business platform, displaying the operational voucher image bound to the business link. Furthermore, the computer device can use a web crawler strategy associated with the business platform to parse and process the voucher image display interface, obtaining the operational voucher image from it. This operational voucher image is then used as the image to be collected, and an optical character recognition (OCR) model is called through the OCR interface. Further, the computer device can use the OCR model to recognize the image to be collected, and the OCR information recognized from the image to be collected is used as the subject information of the operating entity to which the target business object belongs, collected from the operational voucher image.

[0144] In this context, it can be understood that the specific process by which the computer device extracts the operational voucher image from the voucher image display interface can be described as follows: The computer device can use the second image collection rule (e.g., the business license image extraction rule) in the web crawler strategy associated with the business platform to search for the voucher image address that matches the second key element in the second image collection rule in the data structure tree of the voucher image display interface. Then, based on the verified image address, the operational voucher image can be extracted from the voucher display area in the voucher image display interface. For example, the operational voucher image here can be the aforementioned... Figure 2 The corresponding embodiment shows the operation certificate image 204a.

[0145] It is understandable that the specific implementation method of the computer device extracting the operation certificate image through the second image acquisition rule can be found in the above. Figure 4 The specific process of extracting the target verification image in the corresponding embodiments will not be repeated here.

[0146] Therefore, the computer device in this embodiment can crawl business links of target business objects on a business platform using web crawling strategies, and determine the target CAPTCHA image to be verified based on the business links and the web crawling strategy. Furthermore, the computer device can acquire a target character recognition model associated with the business platform, and input the target CAPTCHA image into the target character recognition model, which then recognizes the target CAPTCHA image to obtain the target CAPTCHA characters. It should be understood that by introducing a target character recognition model, this embodiment can intelligently extract the target CAPTCHA characters from the target CAPTCHA image, thereby improving the real-time performance and accuracy of character extraction. Furthermore, the computer device can simulate inputting the target CAPTCHA characters into the character input area corresponding to the target CAPTCHA image, and then, in response to the simulated submission operation of the target CAPTCHA characters in the character input area, perform character verification on the target CAPTCHA characters to obtain the character verification result. Further, if the character verification result indicates successful verification, the computer device can acquire an operating voucher image bound to the business link, and collect the entity information of the operating entity to which the target business object belongs from the operating voucher image. Therefore, the embodiments of this application can intelligently verify the identified target verification code characters when they are recognized. This allows for the rapid extraction of business entity information (e.g., business license entity information) from the operating voucher image (e.g., business license image) upon successful verification, thereby improving the efficiency of obtaining entity information.

[0147] For further details, please see Figure 6 , Figure 6 This is a flowchart illustrating a verification code character processing method provided in an embodiment of this application. It should be understood that the method provided in this embodiment can be executed by a computer device, which includes, but is not limited to, a user terminal or a server. Figure 6 As shown, the method may include at least the following steps S201-S211;

[0148] Step S201: Obtain the web crawler strategy associated with the business platform, and use the web crawler strategy to crawl the link information of N business objects from the business platform.

[0149] Where N is a positive integer;

[0150] Step S202: Obtain the target business object from N business objects, and use the link information of the target business object as the business link of the target business object;

[0151] Step S203: Parse the business link to obtain the operation link of the operation entity to which the target business object belongs, and output the operation display interface of the operation entity based on the operation link;

[0152] The operation display interface includes a query interface for querying operation certificate images of the operating entity;

[0153] Step S204: Respond to the simulated trigger operation for the query interface, output the simulated verification interface corresponding to the operation display interface, and obtain the target verification code image to be verified from the simulated verification interface through a web crawler strategy.

[0154] The specific implementation methods of steps S201-S204 can be found in the above. Figure 3 The description of step S101 in the corresponding embodiments will not be repeated here.

[0155] Step S205: Obtain the target character recognition model associated with the business platform, input the target verification code image into the target character recognition model, and have the target character recognition model recognize the target verification code image to obtain the target verification code characters in the target verification code image;

[0156] Step S206: Simulate input of the target verification code characters into the character input area corresponding to the target verification code image; in response to the simulated submission operation of the target verification code characters in the character input area, perform character verification on the target verification code characters and obtain the character verification result.

[0157] Step S207: If the character verification result indicates successful verification, obtain the operation voucher image bound to the business link, and collect the main information of the operation entity to which the target business object belongs from the operation voucher image.

[0158] The specific implementation methods of steps S205-S207 can be found in the above. Figure 3 The descriptions of steps S102-S104 in the corresponding embodiments will not be repeated here.

[0159] Step S208: In the entity information database associated with the operating entity, establish the association between entity information and business links, and update the association to the entity information database.

[0160] Step S209: Use the business link carrying the association relationship as the webpage update link, and update the business link of the target business object as the webpage update link on the business platform.

[0161] Step S210: When it is found that the target business object under the webpage update link is an illegal object in the blacklist, the operating entity indicated by the main information is regarded as an illegal operating entity based on the association between the main information and the business link.

[0162] Step S211: Generate notification information associated with the illegal operating entity and send the notification message to the monitoring terminal corresponding to the platform supervisor associated with the business platform.

[0163] For further information, please refer to [link / reference]. Figure 7 , Figure 7 This is a schematic diagram illustrating a scenario of an overall data collection scheme based on web crawlers provided in an embodiment of this application. For example... Figure 7 All steps of the data acquisition scheme shown rely on a web crawler strategy. It should be understood that the web crawler strategy here specifically refers to a program or script that automatically retrieves information from the World Wide Web according to certain rules. Therefore, in the business scenario of intelligent supervision of e-commerce operation systems involved in the embodiments of this application, the following can be implemented... Figure 7 In step S11, a web crawler strategy is used to collect link information of one or more business objects from the platform 71a. This collected link information can then be collectively referred to as the link information of N business objects crawled from the business platform. For example, in an e-commerce operation system, the link information of one business object can specifically be the product link of a single product. This means that the computer device can use the link acquisition rules involved in the web crawler strategy (e.g., the aforementioned product leads) to crawl the product links of N listed products from the platform 71a (e.g., the aforementioned QQ Mall shopping platform). Each crawled product can be treated as a target business object, and these N product links can be collectively referred to as the business links of the target business object. This allows for the legitimate monitoring of the operating entities of a large number of products on the platform 71a, preventing the illegal operation of unauthorized entities.

[0164] like Figure 7 For details on how the computer device obtains the verification code image, please refer to the above. Figure 3 or Figure 6 The corresponding embodiment describes the specific process of obtaining the verification code image. Further, the computer device can input the currently obtained verification code image (i.e., the aforementioned target verification code image). Figure 7 The target character recognition model 72a shown is used to perform... Figure 7 As shown in step S12, the target character recognition model can be used to identify the verification code characters in the currently acquired verification code image (i.e., the target verification code image mentioned above).

[0165] It should be understood that the computer device inputs the currently obtained verification code image (i.e., the aforementioned target verification code image) into... Figure 7 In the process of recognizing the CAPTCHA using the target character recognition model 72a shown, the target CAPTCHA image needs to be located in a specific area from the simulated verification interface beforehand using the first image acquisition rules involved in the web crawler strategy. Then, the target CAPTCHA image can be extracted from the simulated verification interface. The specific implementation method for the computer device to extract the target CAPTCHA image from the simulated verification interface can be found in the above... Figure 3 The description of the first image acquisition rules in the corresponding embodiments will not be repeated here.

[0166] Furthermore, when the computer device recognizes the target verification code characters in the target verification code image using the target character recognition model, it can perform character verification on the target verification code characters. If the verification is successful, it can then obtain the operational voucher images bound to each product link (i.e., the aforementioned business links). At this point, the computer device can extract the operational voucher images bound to each business link from the aforementioned voucher image display interface using the second image collection rules involved in the web crawling strategy. The extracted operational voucher images can then be input into... Figure 7 The OCR interface shown is used to further execute step S13, thereby obtaining corresponding optical character information from each operation voucher image through the OCR interface, and using the corresponding optical character information obtained from each operation voucher image as the subject information of the operation entity to which each target business object belongs.

[0167] Furthermore, such as Figure 7 As shown, when computer equipment obtains the main information of each operating entity, it can... Figure 7 In the entity information database 74a shown, entity information of the operating entity associated with the corresponding business link is added or updated in the list column where each business link is located, thereby establishing the association between the entity information of each operating entity and each business link.

[0168] It should be understood that if a product link (e.g., business link L3) retrieved by the computer device from the platform to be collected is already associated with the operating entity to which the product belongs (e.g., the aforementioned virtual store D), then the computer device will not need to repeatedly establish the association between business link L3 and the entity information of virtual store D in the entity information database 74a. Conversely, if a product link (e.g., business link L4) retrieved by the computer device from the platform to be collected is not associated with the operating entity to which the product belongs (e.g., the aforementioned virtual store D), then when the computer device obtains the entity information of virtual store D, it will... Figure 7 The database 74a shows the establishment of a relationship between the business link L4 and the main information of the virtual store D. This allows for direct association between the product link L3 and the main information of the virtual store. Consequently, regulatory authorities associated with the data collection platform 71a can use the relationships between business links and main information to legally monitor various operating entities. This enables them to quickly and effectively identify illegal operating entities on the data collection platform 71a, thus addressing the root cause of the problem that the main information of operating entities cannot be obtained directly through the aforementioned product clues. In other words, computer equipment can use the established relationships to help regulatory authorities efficiently supervise the platform operations of illegal operating entities.

[0169] Therefore, this application embodiment, by introducing a target character recognition model, can intelligently extract target CAPTCHA characters from target CAPTCHA images, thereby improving the real-time performance and accuracy of character extraction. Furthermore, this application embodiment can intelligently perform character verification on the identified target CAPTCHA characters. Upon successful verification, it can quickly extract business entity information (e.g., business license entity information) from the operating voucher image (e.g., a watermarked business license image), thereby improving the efficiency of obtaining entity information and fundamentally helping to solve the problem that the entity information of the operating entity cannot be directly associated with the aforementioned product clues.

[0170] For further details, please see Figure 8 , Figure 8 This application provides a verification code character processing method, which can be executed by the aforementioned computer device. The method may include the following steps S301-S305.

[0171] Step S301: Obtain the original sample images used to train the initial character recognition model;

[0172] The original sample images were determined based on the sample links and web crawling strategies; the sample links were obtained from the business platform through web crawling strategies.

[0173] Specifically, when a computer device retrieves product links from a business platform using a web crawler strategy, it can use these retrieved product links as sample links. It can further search for historical CAPTCHA images determined through the web crawler strategy and sample links, and then use these historical CAPTCHA images as raw sample images for training the initial character recognition model. It should be understood that when the computer device obtains the raw sample image, it can annotate the CAPTCHA characters in the raw sample image (for example, it can use CAPTCHA characters identified from the raw sample image as annotated CAPTCHA characters to annotate the raw sample image). Based on the annotated raw sample image, it can further execute step S302, thereby quickly generating a large number of enhanced sample images with different CAPTCHA styles from the currently annotated small number of CAPTCHA images (i.e., the aforementioned annotated raw sample images or the aforementioned raw sample images).

[0174] Step S302: Obtain a data augmentation strategy for data augmentation of the original sample image, perform data augmentation processing on the original sample image based on the data augmentation strategy, and obtain at least one augmented sample image associated with the original sample image;

[0175] Note that the data augmentation strategies here can specifically include character distortion, character enlargement, character rotation, and character offset strategies. Specifically, when a computer device acquires these data augmentation strategies, it can apply them to the currently labeled original sample image to quickly obtain augmented sample images with different CAPTCHA styles, thereby generating at least one augmented sample image associated with the original sample image.

[0176] Specifically, the CAPTCHA styles mentioned here can include: character distortion styles corresponding to character distortion strategies, character method styles corresponding to character enlargement strategies, character rotation styles corresponding to character rotation strategies, and horizontal or vertical character position offset styles corresponding to character offset strategies. It should be understood that, after performing data augmentation processing on the obtained original sample images based on the above data augmentation strategies, this application embodiment can obtain augmented sample images with different CAPTCHA styles. This allows for the generation of a large amount of augmented data (i.e., the aforementioned augmented sample images with different CAPTCHA styles) that can be used for training from a small amount of labeled data (i.e., labeled original sample images), so that a target character recognition model capable of recognizing different CAPTCHA styles can be trained.

[0177] Step S303: Use the original sample image and at least one enhanced sample image as the target sample image, and use the CAPTCHA characters corresponding to the original sample image as the sample label of the target sample image.

[0178] It is understood that, in this embodiment of the application, when obtaining target sample images for training the initial character recognition model, it is necessary to assign corresponding sample labels to these target sample images. Since the enhanced sample images are obtained by distorting, enlarging, rotating, or shifting characters in the original sample images, this embodiment of the application can use the labeled verification code characters of the labeled original sample images as the sample labels of these enhanced sample images, thereby fundamentally reducing the cost of simulation learning.

[0179] Step S304: Input the target sample image into the initial character recognition model, and the initial character recognition model will recognize the target sample image and use the sample verification code characters recognized from the target sample image as the predicted label.

[0180] Step S305: Based on the predicted labels and sample labels, the initial character recognition model is iteratively trained to obtain a target character recognition model for recognizing target CAPTCHA characters in the target CAPTCHA image;

[0181] The target verification code character is used to obtain the operation certificate image bound to the business link associated with the target verification code image after character verification; the operation certificate image is used to collect the main information of the operating entity to which the target business object indicated by the business link belongs.

[0182] Specifically, the computer device can determine an initial loss function for adjusting the model parameters in the initial character recognition model based on the predicted probability value corresponding to the predicted label and the true probability value corresponding to the sample label. This initial loss function is jointly determined by the loss functions of the first and second networks. Further, if the computer device determines that the value of the initial loss function does not meet the model convergence condition, it adjusts the model parameters of each network in the initial character recognition model based on the value of the initial loss function. Then, iteratively trains the adjusted initial character recognition model using the target sample image to obtain the target loss function of the iteratively trained initial character recognition model. Further, if the value of the target loss function meets the model convergence condition, the computer device can determine the iteratively trained initial character recognition model that meets the model convergence condition as the target character recognition model. The target character recognition model can be used to recognize target CAPTCHA characters in target CAPTCHA images in real time.

[0183] In this context, it's understandable that after predicting the label, the computer device can compare it with the true label of the target sample image to obtain the initial loss function value of the model. It's also understandable that if this initial loss function value does not meet the aforementioned model convergence condition (for example, if the initial loss function value is not the minimum loss function value during model training), the computer device can adjust the model parameters of each network in the initial character recognition model based on this initial loss function value. This allows for iterative training of the adjusted initial character recognition model using new target sample images until the target loss function value of the iteratively trained initial character recognition model meets the model convergence condition. The iteratively trained initial character recognition model that meets the model convergence condition is then determined as the target character recognition model.

[0184] For further information, please refer to [link / reference]. Figure 9 , Figure 9 This is a schematic diagram illustrating a scenario where an initial character recognition model is trained to obtain a target character recognition model, as provided in an embodiment of this application. Figure 9 Steps S921-S927, as shown, represent the model training phase. Figure 9 As shown, during the model training phase, the few CAPTCHA images annotated in step S921 can be the original sample images annotated above. Figure 9 As shown, the computer device can further execute step S922 to perform data augmentation processing on the original sample image by means of distortion, rotation, etc., so as to obtain an augmented sample image that is associated with a small number of labeled CAPTCHA images and has a different CAPTCHA style.

[0185] Furthermore, such as Figure 9 As shown, to reduce the cost of model learning, computer devices can perform... Figure 9 Step S923, as shown, involves denoising the target sample images used in training (e.g., the aforementioned few labeled CAPTCHA images and enhanced sample images with different CAPTCHA styles), thereby using the denoised target sample images as training sample images. At this point, the computer device can further execute step S924 to input the training sample image into the convolutional neural network (e.g., a CNN network) in the initial character recognition model, so as to extract the image convolutional features of the training sample image (i.e., the sample image convolutional features of the target sample image) through the convolutional neural network in the initial character recognition model. Further, the computer device can execute step S925 to extract character sequence features from the sample image convolutional features through the recurrent neural network (e.g., a Bi-LSTM network) in the target character recognition model, and the extracted character sequence features from the sample image convolutional features can be collectively referred to as the sample character sequence features corresponding to the target sample image. Furthermore, the computer device can execute step S926 to align the sample character sequence features obtained in step S925 through a connected temporal classification network, and then output the sample CAPTCHA characters in the target sample image based on the aligned sample character sequence features. It should be understood that during the model training phase, the sample CAPTCHA characters output by the initial character recognition model are the predicted labels mentioned above. At this time, the computer device can execute step S927 to train the initial character recognition model using the predicted labels and sample labels, and then use the trained initial character recognition model as the target character recognition model in the model application phase.

[0186] Among them, such as Figure 9 As shown, in the model application stage, the computer device can execute step S911 to parse the above-mentioned store link and output the simulated verification interface corresponding to the above-mentioned operation display interface, and then obtain the verification code image to be verified (i.e., the above-mentioned target verification code image to be verified) from the simulated verification interface. It should be understood that the target character recognition model here is a character recognition model based on an end-to-end algorithm. Therefore, after the computer device executes step S912, the target character recognition model can identify the target verification code characters in the target verification code image in real time, and then execute the anti-crawler strategy. Figure 9 Step S913, as shown, requires simulating user input of characters one by one within a set delay of 1-2 seconds (i.e., the aforementioned sleep duration), according to a randomly set character input interval. This allows for subsequent character verification of the input characters, and if the character verification is successful, execution can proceed. Figure 9 Step S914, as shown, involves obtaining the main information from the operating voucher image. The method for obtaining the main information from the operating voucher image can be found above. Figure 3 The specific process of obtaining subject information described in the corresponding embodiments will not be repeated here.

[0187] Optionally, when executing step S926, the computer device may also serially output the sample verification code characters through the aforementioned attention mechanism network. For further understanding, please refer to [link to relevant documentation]. Figure 10 , Figure 10 This is a schematic diagram illustrating a scenario where sample labels are obtained through an attention mechanism network, as provided in an embodiment of this application. It should be understood that during the model training phase, such as... Figure 10 The verification code image 101a shown can be the target sample image mentioned above. The convolutional recurrent neural network shown in the figure can specifically include... Figure 9 The diagram shows a convolutional neural network and a recurrent neural network. When a computer device outputs data through this convolutional and recurrent neural network... Figure 10 When the character sequence feature 103 is shown, this character sequence feature 103a can be used as the aforementioned sample character sequence feature. Thus, when the sample character sequence feature is input to... Figure 10 When using the attention mechanism network 104a, the attention map features of the sample character sequence features at each time step can be encoded by the attention mechanism network. At this time, the computer device can use the speech extraction network in the attention mechanism network to extract character features from the labeled CAPTCHA characters corresponding to the sample labels, thereby obtaining the CAPTCHA character features corresponding to each character element in the labeled CAPTCHA characters. Then, based on the extracted CAPTCHA character features of each character element and the aforementioned attention map features, the computer device can decode and output the prediction elements corresponding to each character element through the decoding functions (e.g., decoding functions C1, C2, and C3) in the attention mechanism network. For example, the prediction elements may specifically include prediction elements "3", "n", "H", and "f". Furthermore, based on the string formed by these predicted elements, the computer device can determine... Figure 10 The predicted label for the CAPTCHA image 101a shown, for example, the sample CAPTCHA characters corresponding to the predicted label can be... Figure 10 The verification code shown is 105a.

[0188] Therefore, it can be seen that the computer device involved in the embodiments of this application can, before obtaining the target character recognition model, also use the above-mentioned... Figure 9 or Figure 10 The character classification network in the corresponding embodiment trains the initial character recognition model. It should be understood that in this embodiment, the computer device can annotate the characters in the acquired original sample images when it obtains them for training the initial character recognition model. Then, it can perform data augmentation on the annotated original sample images using data augmentation strategies to quickly generate at least one augmented sample image associated with the original sample image. This way, only a small number of original sample images need to be annotated during the model training phase, thereby reducing the annotation work on the large number of target sample images involved in training and lowering the cost of model training. In other words, this embodiment can quickly generate augmented sample images of different CAPTCHA styles using a small number of annotated original sample images. Thus, during the model training of the initial character recognition model, target sample images of different CAPTCHA styles can be quickly obtained based on the original sample images and at least one augmented sample image. At this time, the computer device can directly use the CAPTCHA characters of the annotated original sample images as sample labels for the target sample images, thereby quickly and accurately configuring corresponding sample labels for these target sample images involved in training. Furthermore, after inputting the target sample image into the initial character recognition model, the computer device can intelligently recognize the target sample image and use the identified CAPTCHA characters as predicted labels. At this point, the computer device can iteratively train the initial character recognition model based on the true label (i.e., the aforementioned sample label) and the predicted label of the target sample image to obtain a target character recognition model for recognizing the target CAPTCHA characters in the target CAPTCHA image. This target character recognition model can be applied to various CAPTCHA scenarios and can also solve the problem of recognizing variable-length CAPTCHAs through convolutional recurrent neural networks and character classification networks within the target character recognition model, thereby improving the accuracy and reliability of character recognition at low cost and high efficiency. In addition, it should be understood that the specific implementation method of the computer device training the target character recognition model to recognize the target CAPTCHA characters in the target CAPTCHA image can be found in the above description. Figure 3 or Figure 6 The description of the target character recognition model in the corresponding embodiments will not be repeated here.

[0189] Further, please see Figure 11 , Figure 11 This is a schematic diagram of the structure of a verification code character processing device provided in an embodiment of this application. The aforementioned verification code character processing device 1 can be a computer program (including program code) running on a computer device, for example, the verification code character processing device 1 can be an application software; the device can be used to execute the corresponding steps in the method provided in the embodiment of this application. The verification code character processing device 1 may include: a target image determination module 11, a target character recognition module 12, a character submission module 13, and a main information collection module 14; optionally, the verification code character processing device 1 may further include: an association relationship establishment module 15, a business link update module 16, an illegal object detection module 17, and a notification information generation module 18.

[0190] The target image determination module 11 uses a web crawler strategy to crawl the business links of the target business objects on the business platform, and determines the target verification code image to be verified based on the business links and the web crawler strategy.

[0191] The target image determination module 11 includes: a link capture unit 111, a link determination unit 112, a link parsing unit 113, and a verification interface output unit 114.

[0192] Link crawling unit 111 is used to obtain the web crawling strategy associated with the business platform, and crawl the link information of N business objects from the business platform through the web crawling strategy; N is a positive integer;

[0193] Link determination unit 112 is used to obtain the target business object from N business objects and use the link information of the target business object as the business link of the target business object;

[0194] The link parsing unit 113 is used to parse the business link, obtain the operation link of the operation entity to which the target business object belongs, and output the operation display interface of the operation entity based on the operation link; the operation display interface includes a query interface for querying the operation voucher image of the operation entity.

[0195] The verification interface output unit 114 is used to respond to the simulated trigger operation for the query interface, output the simulated verification interface corresponding to the operation display interface, and obtain the target verification code image to be verified from the simulated verification interface through a web crawler strategy.

[0196] The verification interface output unit 114 includes: an interface triggering subunit 1141, a first query subunit 1142, and a verification image extraction subunit 1143.

[0197] The interface triggering subunit 1141 is used to respond to the simulated triggering operation for the query interface and switch the operation display interface to the simulated verification interface; the simulated verification interface includes a target display area; the target display area is used to display the target verification code image associated with the operation voucher image;

[0198] The first query subunit 1142 is used to query the verification image address that matches the first key element in the first image collection rule in the data structure tree corresponding to the simulated verification interface, based on the first image collection rule in the web crawler strategy.

[0199] The verification image extraction subunit 1143 is used to extract the target verification code image from the target display area based on the verification image address.

[0200] The specific implementation methods of the interface triggering subunit 1141, the first query subunit 1142, and the verification image extraction subunit 1143 can be found in the above description. Figure 3 The specific process of extracting the target verification code image described in the corresponding embodiments will not be repeated here.

[0201] The specific implementation methods of the link capture unit 111, link determination unit 112, link parsing unit 113, and verification interface output unit 114 can be found in the above description. Figure 3 The description of step S101 in the corresponding embodiment will not be repeated here.

[0202] The target character recognition module 12 obtains the target character recognition model associated with the business platform, inputs the target verification code image into the target character recognition model, and the target character recognition model recognizes the target verification code image to obtain the target verification code character corresponding to the target verification code image;

[0203] The target character recognition module 12 includes: a target image acquisition unit 121, a convolutional feature extraction unit 122, a sequence feature extraction unit 123, and a sequence feature alignment unit 124.

[0204] The target image acquisition unit 121 is used to acquire the target character recognition model associated with the business platform, perform noise reduction processing on the target verification code image, and use the noise-reduced target verification code image as the image to be processed.

[0205] The convolutional feature extraction unit 122 is used to input the image to be processed into the convolutional neural network in the target character recognition model, and extract the image convolutional features of the image to be processed by the convolutional neural network in the target character recognition model, and use the extracted image convolutional features as the target image convolutional features of the target verification code image.

[0206] The sequence feature extraction unit 123 is used to input the convolutional features of the target image into the recurrent neural network in the target character recognition model, and extract the sequence features from the convolutional features of the target image by the recurrent neural network in the target character recognition model, and use the extracted character sequence features as the target character sequence features corresponding to the target verification code image;

[0207] The target character recognition model includes a recurrent neural network, which is a bidirectional long short-term memory network. The bidirectional long short-term memory network includes a forward long short-term memory network and a backward long short-term memory network. The forward long short-term memory network includes a memory network B. i and memory network B i+1 Memory Network B i+1 For memory network B i The next memory network; the inverse long short-term memory network includes memory network C. i+1 and memory network C i ; Memory Network C i+1 For memory network C i The previous memory network; i is a positive integer less than or equal to M; the number of memory networks in both the forward long short-term memory network and the reverse long short-term memory network is M;

[0208] The sequence feature extraction unit 123 includes: a forward feature extraction subunit 1231, a reverse feature extraction subunit 1232, a feature splicing subunit 1233, and a sequence feature determination subunit 1234;

[0209] The forward feature extraction subunit 1231 is used to obtain the memory network B in the forward long short-term memory network. i Associated positive history hidden features h i-1 The convolutional features and positive history hidden features h of the target image are combined. i-1 Input memory network B i , by memory network B i The positive target hidden feature h is extracted at time i. i Hide features h of the positive target i Input memory network B with convolutional features of the target image i+1 , by memory network B i+1 The positive target hidden feature h is extracted at time i+1. i+1 ;

[0210] The reverse feature extraction subunit 1232 is used to obtain the memory network C in the reverse long short-term memory network. i+1 Associated reverse history hidden features k i+1 The convolutional features of the target image and the inverse history hidden features k i+1 Input memory network C i+1 , by memory network C i+1 The reverse target hidden feature k is extracted at time i+1. i Hide the reverse target feature k i The target image convolutional feature input memory network C i , by memory network C i The reverse target hidden feature k is extracted at time i. i-1 ;

[0211] Feature splicing subunit 1233 is used to integrate memory network B i The positive target hidden feature h extracted at time i i With memory network C i The reverse target hidden feature k extracted at time i i-1 Perform feature concatenation to obtain the first concatenated feature, and then use the memory network B. i+1 The positive target hidden feature h extracted at time i+1 i+1 With memory network C i+1 The reverse target hidden feature k extracted at time i+1 i Perform feature concatenation to obtain the second concatenated feature;

[0212] The sequence feature determination subunit 1234 is used to determine the target character sequence features corresponding to the target verification code image extracted from the convolutional features of the target image based on the first concatenation feature and the second concatenation feature.

[0213] The specific implementation methods of the forward feature extraction subunit 1231, the reverse feature extraction subunit 1232, the feature concatenation subunit 1233, and the sequence feature determination subunit 1234 can be found in the above description. Figure 3 The description of the bidirectional long short-term memory network in the corresponding embodiments will not be repeated here.

[0214] The sequence feature alignment unit 124 is used to input the target character sequence features into the character classification network in the target character recognition model, and perform alignment processing on the target character sequence features through the connection time classification network in the character classification network, and obtain the target verification code characters in the target verification code image based on the aligned target character sequence features.

[0215] The specific implementation methods of the target image acquisition unit 121, the convolutional feature extraction unit 122, the sequence feature extraction unit 123, and the sequence feature alignment unit 124 can be found above. Figure 3 The description of step S102 in the corresponding embodiment will not be repeated here.

[0216] The character submission module 13 is used to simulate inputting the target verification code characters into the character input area corresponding to the target verification code image. In response to the simulated submission operation of the target verification code characters in the character input area, the module performs character verification on the target verification code characters and obtains the character verification result.

[0217] The character submission module 13 includes: an anti-crawler unit 131, a character input unit 132, and a character submission unit 133;

[0218] Anti-crawler unit 131 is used to obtain anti-crawler strategies for the target verification code characters, and to obtain the sleep duration and character input interval duration indicated by the anti-crawler strategy.

[0219] The character input unit 132 is used to simulate inputting the target verification code characters into the character input area corresponding to the target verification code image according to the character input interval during the sleep period; the character input area includes a character input box and a character submission control.

[0220] The character submission unit 133 is used to respond to the simulated submission operation of the character submission control, perform character verification on the target verification code characters displayed in the character input box, and obtain the character verification result.

[0221] The specific implementation methods of the anti-crawler unit 131, the character input unit 132, and the character submission unit 133 can be found in the above description. Figure 3 The description of step S103 in the corresponding embodiment will not be repeated here.

[0222] The main information collection module 14 is used to obtain the operation voucher image bound to the business link if the character verification result indicates successful verification, and to collect the main information of the operation entity to which the target business object belongs from the operation voucher image.

[0223] The main information collection module 14 includes: an operation voucher display unit 141, a voucher interface parsing unit 142, an optical model calling unit 143, and a main information collection unit 144.

[0224] The operation voucher display unit 141 is used to output the voucher image display interface associated with the business platform if the character verification result indicates that the verification is successful. The operation voucher image bound to the business link is displayed in the voucher image display interface.

[0225] The voucher interface parsing unit 142 is used to parse the voucher image display interface through a web crawler strategy associated with the business platform to obtain the operational voucher image in the voucher image display interface.

[0226] The voucher interface parsing unit 142 includes: a second query subunit 1421 and an operation image extraction subunit 1422;

[0227] The second query subunit 1421 is used to search for a voucher image address that matches the second key element in the second image collection rule in the data structure tree of the voucher image display interface by using the second image collection rule in the web crawler strategy associated with the business platform.

[0228] The operation image extraction subunit 1422 is used to extract the operation voucher image from the voucher display area in the voucher image display interface based on the verification image address.

[0229] The specific implementation methods of the second query subunit 1421 and the operation image extraction subunit 1422 can be found in the above. Figure 3 The specific process of obtaining the operation certificate image in the corresponding embodiment will not be repeated here.

[0230] The optical model calling unit 143 is used to use the operation certificate image as the image to be collected and call the optical character recognition model through the optical character recognition interface.

[0231] The main information collection unit 144 is used to recognize the image to be collected through an optical character recognition model, and to use the optical character information recognized from the image to be collected as the main information of the operating entity to which the target business object belongs, collected from the operation certificate image.

[0232] The specific implementation methods of the operation voucher display unit 141, the voucher interface parsing unit 142, the optical model calling unit 143, and the main information collection unit 144 can be found in the above description. Figure 3 The description of step S104 in the corresponding embodiment will not be repeated here.

[0233] Optionally, the association establishment module 15 is used to establish the association between entity information and business links in the entity information database associated with the operating entity, and update the association to the entity information database.

[0234] Optionally, the business link update module 16 is used to update the business link of the target business object into a webpage update link on the business platform by using the business link carrying the association relationship as the webpage update link.

[0235] Optionally, the illegal object detection module 17 is used to identify the operating entity indicated by the main information as an illegal operating entity when the target business object under the webpage update link is found to be an illegal object in the blacklist, based on the association between the main information and the business link.

[0236] The notification information generation module 18 is used to generate notification information associated with illegal operating entities and send the notification message to the regulatory terminal corresponding to the platform regulator associated with the business platform.

[0237] The specific implementation methods of the target image determination module 11, target character recognition module 12, character submission module 13, and main body information collection module 14 can be found above. Figure 3 The descriptions of steps S101-S104 in the corresponding embodiments will not be repeated here. Furthermore, the specific implementations of the association establishment module 15, the business link update module 16, the illegal object detection module 17, and the notification information generation module 18 can be found above. Figure 6 The descriptions of steps S201-S211 in the corresponding embodiments will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated.

[0238] For further details, please see Figure 11 , Figure 12 This is a schematic diagram of the structure of a verification code character processing device provided in an embodiment of this application. The verification code character processing device 2 can be a computer program (including program code) running on a computer device; for example, the verification code character processing device 2 can be an application software. The device can be used to execute the corresponding steps in the method provided in the embodiment of this application. The verification code character processing device 2 can include: an original image acquisition module 21, an enhanced image generation module 22, a sample image determination module 23, a sample character recognition module 24, and a model training module 25.

[0239] The original image acquisition module 21 is used to acquire the original sample images for training the initial character recognition model; the original sample images are determined based on the sample links and web crawling strategies; the sample links are crawled from the business platform through web crawling strategies.

[0240] The enhanced image generation module 22 is used to obtain a data augmentation strategy for data augmentation of the original sample image, perform data augmentation processing on the original sample image based on the data augmentation strategy, and obtain at least one enhanced sample image associated with the original sample image.

[0241] The sample image determination module 23 is used to use the original sample image and at least one enhanced sample image as the target sample image, and to use the annotation verification code characters corresponding to the original sample image as the sample label of the target sample image.

[0242] The sample character recognition module 24 is used to input the target sample image into the initial character recognition model, and the initial character recognition model recognizes the target sample image and uses the sample verification code characters recognized from the target sample image as the predicted label.

[0243] The model training module 25 is used to iteratively train the initial character recognition model based on the predicted labels and sample labels to obtain a target character recognition model for recognizing target CAPTCHA characters in the target CAPTCHA image; the target CAPTCHA characters are used to obtain the operation voucher image bound to the business link associated with the target CAPTCHA image after character verification; the operation voucher image is used to collect the main information of the operating entity to which the target business object indicated by the business link belongs.

[0244] The specific implementation methods of the original image acquisition module 21, the enhanced image generation module 22, the sample image determination module 23, the sample character recognition module 24, and the model training module 25 can be found above. Figure 8 The descriptions of steps S301-S305 in the corresponding embodiments will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated.

[0245] Further, please see Figure 13 , Figure 13 This is a schematic diagram of a computer device provided in an embodiment of this application. Figure 13 The computer device 1000 shown may include: at least one processor 1001, such as a CPU; at least one network interface 1004; a user interface 1003; a memory 1005; and at least one communication bus 1002. The communication bus 1002 is used to enable communication between these components. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as at least one disk drive. Optionally, the memory 1005 may also be at least one storage device located remotely from the aforementioned processor 1001. Figure 13 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a device control application program.

[0246] exist Figure 13 In the computer device 1000 shown, the network interface 1004 is mainly used to provide network communication functions; while the user interface 1003 is mainly used to provide an input interface for the user; and the processor 1001 can be used to call the device control application stored in the memory 1005 to execute the aforementioned... Figure 3 or Figure 6 or Figure 8 The description of the verification code character processing method in the corresponding embodiment can also be executed as described above. Figure 11 The description of the verification code character processing device 1 in the corresponding embodiment can also be executed as described above. Figure 12 The description of the verification code character processing device 2 in the corresponding embodiments will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated.

[0247] Furthermore, it should be noted that this application embodiment also provides a computer-readable storage medium, which stores a computer program executed by the computer device 1000 mentioned above. The computer program includes program instructions, and when the processor executes the program instructions, it can execute the aforementioned... Figure 3 or Figure 6 or Figure 8 The description of the verification code character processing method in the corresponding embodiments is already provided and will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer-readable storage medium embodiments related to this application, please refer to the description of the method embodiments of this application.

[0248] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0249] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.< / space> < / space> < / space> < / space> < / space>

Claims

1. A method for processing verification code characters, characterized in that, include: The web crawler strategy retrieves business links of target business objects on the business platform based on product leads. Then, based on these business links and the web crawler strategy, it determines target verification code images associated with the operational voucher images to be viewed. The target business objects refer to N virtual products in virtual stores on the business platform that are subject to security monitoring; N is a positive integer. According to the intelligent data acquisition strategy, a target character recognition model for recognizing variable-length CAPTCHAs based on an end-to-end algorithm is obtained; the target character recognition model includes a convolutional neural network, a recurrent neural network, and a character classification network with character classification function; the character classification network includes any one of a connection time classification network and an attention mechanism network; the attention mechanism network is used to indicate how to extract the image convolutional features of the target CAPTCHA image through the convolutional neural network without reducing the pooling layers of the convolutional neural network; When the character classification network is the connection time classification network, the pooling layers in the convolutional neural network are reduced to obtain a convolutional neural network with reduced pooling layers. The target CAPTCHA image is input into the convolutional neural network with reduced pooling layers in the target character recognition model. The target image convolutional features of the target CAPTCHA image are extracted through the convolutional neural network with reduced pooling layers. The convolutional features of the target image are input into the recurrent neural network. The recurrent neural network calculates the forward target hidden features of the target image convolutional features at each time step along the first feature calculation direction, and calculates the reverse target hidden features of the target image convolutional features at each time step along the second feature calculation direction. The forward target hidden features and the corresponding reverse target hidden features at each time step are concatenated to obtain the concatenated features at each time step. The target character sequence features corresponding to the target CAPTCHA image are extracted from the target image convolutional features using the concatenated features at each time step. The target character sequence features are input into the connection time classification network, which aligns the target character sequence features and obtains the target CAPTCHA characters in the target CAPTCHA image based on the aligned target character sequence features; the feature dimension of the aligned target character sequence features is consistent with the feature dimension of the target character sequence features. The target verification code character is simulated and input into the character input area corresponding to the target verification code image. In response to the simulated submission operation of the target verification code character in the character input area, the target verification code character is verified to obtain the character verification result. If the character verification result indicates successful verification, then obtain the operation certificate image bound to the business link, and collect the main information of the operation entity to which the target business object belongs from the operation certificate image; The entity information is used to conduct security supervision of the virtual store when the operating entity is the virtual store.

2. The method according to claim 1, characterized in that, The process of retrieving the business link of the target business object on the product lead crawling business platform indicated by the web crawling strategy, and determining the target verification code image associated with the operational voucher image to be viewed based on the business link and the web crawling strategy, includes: Obtain a web crawling strategy associated with the business platform, and use the web crawling strategy to crawl link information of N business objects from the business platform; the N business objects are the N virtual products; Obtain the target business object from the N business objects, and use the link information of the target business object as the business link of the target business object; The business link is parsed to obtain the operation link of the operating entity to which the target business object belongs, and the operation display interface of the operating entity is output based on the operation link; the operation display interface includes a query interface for querying the operation certificate image of the operating entity; In response to the simulated trigger operation of the query interface, the simulated verification interface corresponding to the operation display interface is output, and the target verification code image to be verified is obtained from the simulated verification interface through the web crawler strategy.

3. The method according to claim 2, characterized in that, The response simulates a trigger operation for the query interface, outputs a simulated verification interface corresponding to the operation display interface, and obtains the target verification code image to be verified from the simulated verification interface using the web crawler strategy, including: In response to a simulated trigger operation on the query interface, the operation display interface is switched to a simulated verification interface; the simulated verification interface includes a target display area; the target display area is used to display a target verification code image associated with the operation voucher image; Using the first image collection rule in the web crawler strategy, query the data structure tree corresponding to the simulated verification interface to find the verification image address that matches the first key element in the first image collection rule. The target verification code image is extracted from the target display area based on the verification image address.

4. The method according to claim 1, characterized in that, The step of inputting the target verification code image into the convolutional neural network with reduced pooling layers in the target character recognition model includes: The target verification code image is denoised, and the denoised target verification code image is used as the image to be processed. The image to be processed is input into the convolutional neural network with reduced pooling layers in the target character recognition model.

5. The method according to claim 1, characterized in that, The recurrent neural network in the target character recognition model is a bidirectional long short-term memory (LSTM) network; the bidirectional LSTM network includes a forward LSTM network and a backward LSTM network; the forward LSTM network includes a memory network B. i and memory network B i+1 The memory network B i+1 For the memory network B i The next memory network; the reverse long short-term memory network includes memory network C. i+1 and memory network C i The memory network C i+1 For the memory network C i The previous memory network; i is a positive integer less than or equal to M; the number of memory networks in the forward long short-term memory network and the reverse long short-term memory network is both M; The process involves inputting the convolutional features of the target image into the recurrent neural network. The recurrent neural network calculates the forward target hidden features of the target image convolutional features at each time step along a first feature calculation direction, and calculates the reverse target hidden features of the target image convolutional features at each time step along a second feature calculation direction. The forward target hidden features and the corresponding reverse target hidden features at each time step are then concatenated to obtain the concatenated features at each time step. The target character sequence features corresponding to the target CAPTCHA image are extracted from the target image convolutional features using these concatenated features. This process includes: The convolutional features of the target image are input into the recurrent neural network, which then obtains the memory network B in the forward long short-term memory network along the first feature calculation direction. i Associated positive history hidden features h i-1 The convolutional features of the target image and the positive history hiding features h are combined. i-1 Input the memory network B i The memory network B i The positive target hidden feature h is extracted at time i. i The positive target hiding feature h i The convolutional features of the target image are input into the memory network B. i+1 The memory network B i+1 The positive target hidden feature h is extracted at time i+1. i+1 ; The memory network C in the reverse long short-term memory network is obtained by the recurrent neural network along the second feature calculation direction. i+1 Associated reverse history hidden features k i+1 The convolutional features of the target image and the reverse history hiding features k are combined. i+1 Input the memory network C i+1 The memory network C i+1 The reverse target hidden feature k is extracted at time i+1. i The reverse target hiding feature k i The convolutional features of the target image are input into the memory network C. i The memory network C i The reverse target hidden feature k is extracted at time i. i-1 ; The memory network B i The positive target hidden feature h extracted at time i i With the memory network C i The reverse target hiding feature k extracted at time i i-1 Perform feature concatenation to obtain the first concatenated feature, and then combine the memory network B. i+1 The positive target hidden feature h extracted at time i+1 i+1 With the memory network C i+1 The reverse target hidden feature k extracted at time i+1 i Perform feature concatenation to obtain the second concatenated feature; Based on the first concatenation feature and the second concatenation feature, the target character sequence feature corresponding to the target verification code image is determined from the convolutional features of the target image.

6. The method according to claim 1, characterized in that, If the character verification result indicates successful verification, then an operation certificate image bound to the business link is obtained, and the entity information of the operating entity to which the target business object belongs is collected from the operation certificate image, including: If the character verification result indicates successful verification, the voucher image associated with the business platform will be output to the voucher image display interface, where the operation voucher image bound to the business link will be displayed. The voucher image display interface is parsed and processed using a web crawler strategy associated with the business platform to obtain the operational voucher image in the voucher image display interface. The operational voucher image is used as the image to be collected, and the optical character recognition model is called through the optical character recognition interface. The optical character recognition model is used to identify the image to be collected, and the optical character information identified from the image to be collected is used as the subject information of the operating entity to which the target business object belongs, which is collected from the operation certificate image.

7. The method according to claim 6, characterized in that, The step of parsing the voucher image display interface using a web crawler strategy associated with the business platform to obtain the operational voucher image in the voucher image display interface includes: Using the second image collection rule in the web crawler strategy associated with the business platform, the voucher image address character that matches the second key element in the second image collection rule is searched in the data structure tree of the voucher image display interface. The operation voucher image is extracted from the voucher display area in the voucher image display interface based on the verification image address.

8. The method according to claim 1, characterized in that, The method further includes: In the entity information database associated with the operating entity, establish the association between the entity information and the business link, and update the association to the entity information database; The business link carrying the aforementioned relationship is used as the webpage update link, and the business link of the target business object is updated to the webpage update link on the business platform.

9. The method according to claim 8, characterized in that, The method further includes: When it is found that the target business object under the webpage update link is an illegal object in the blacklist, the operating entity indicated by the main body information is regarded as an illegal operating entity based on the association between the main body information and the business link. Generate a notification message associated with the illegal operating entity and send the notification message to the monitoring terminal corresponding to the platform supervisor associated with the business platform.

10. A method for processing verification code characters, characterized in that, include: Obtain raw sample images for training the initial character recognition model; the raw sample images are determined based on sample links and web crawling strategies; The sample links were obtained from the business platform using the web crawler strategy. A data augmentation strategy for augmenting the original sample image is obtained, and the original sample image is augmented based on the data augmentation strategy to obtain at least one augmented sample image associated with the original sample image. The original sample image and the at least one enhanced sample image are used as target sample images, and the CAPTCHA characters corresponding to the original sample image are used as sample labels for the target sample images. The target sample image is input into the initial character recognition model, which then recognizes the target sample image and uses the sample verification code characters recognized from the target sample image as the predicted label. Based on the predicted label and the sample label, the initial character recognition model is iteratively trained to obtain the target character recognition model as claimed in any one of claims 1-9.

11. A verification code character processing device, characterized in that, include: The target image determination module retrieves business links of target business objects on the business platform based on product leads indicated by the web crawling strategy, and determines the target verification code image associated with the operational voucher image to be viewed based on the business link and the web crawling strategy; the target business object refers to N virtual products in the virtual store to be subject to security supervision presented on the business platform; N is a positive integer; The target character recognition module acquires a target character recognition model for recognizing variable-length CAPTCHAs based on an end-to-end algorithm according to an intelligent data acquisition strategy. The target character recognition model includes a convolutional neural network, a recurrent neural network, and a character classification network with character classification function. The character classification network includes any one of a connection time classification network and an attention mechanism network. The attention mechanism network is used to indicate how to extract the image convolutional features of the target CAPTCHA image through the convolutional neural network without reducing the pooling layers of the convolutional neural network. The target character recognition module is further configured to reduce the pooling layers in the convolutional neural network when the character classification network is the connection time classification network, to obtain a convolutional neural network with reduced pooling layers, input the target CAPTCHA image into the convolutional neural network with reduced pooling layers in the target character recognition model, and extract the target image convolutional features of the target CAPTCHA image through the convolutional neural network with reduced pooling layers. The target character recognition module is further configured to input the convolutional features of the target image into the recurrent neural network, and the recurrent neural network calculates the forward target hidden features of the convolutional features of the target image at each time step along the first feature calculation direction, and calculates the reverse target hidden features of the convolutional features of the target image at each time step along the second feature calculation direction, and concatenates the forward target hidden features and the corresponding reverse target hidden features at each time step to obtain the concatenated features at each time step, and determines the target character sequence features corresponding to the target verification code image extracted from the convolutional features of the target image through the concatenated features at each time step; The target character recognition module is further configured to input the target character sequence features into the connection time classification network, whereby the connection time classification network aligns the target character sequence features and, based on the aligned target character sequence features, obtains the target verification code character corresponding to the target verification code image; the feature dimension of the aligned target character sequence features is consistent with the feature dimension of the target character sequence features. The character submission module is used to simulate inputting the target verification code character into the character input area corresponding to the target verification code image, and in response to the simulated submission operation of the target verification code character in the character input area, to perform character verification on the target verification code character and obtain the character verification result; The main information collection module is used to obtain the operation certificate image bound to the business link if the character verification result indicates successful verification, and to collect the main information of the operation entity to which the target business object belongs from the operation certificate image; The entity information is used to conduct security supervision of the virtual store when the operating entity is the virtual store.

12. A verification code character processing device, characterized in that, include: The original image acquisition module is used to acquire original sample images for training the initial character recognition model; the original sample images are determined based on sample links and web crawling strategies; the sample links are crawled from the business platform through the web crawling strategy. An enhanced image generation module is used to obtain a data augmentation strategy for data augmentation of the original sample image, and to perform data augmentation processing on the original sample image based on the data augmentation strategy to obtain at least one enhanced sample image associated with the original sample image. The sample image determination module is used to use the original sample image and the at least one enhanced sample image as target sample images, and to use the CAPTCHA characters corresponding to the original sample image as sample labels for the target sample images. The sample character recognition module is used to input the target sample image into the initial character recognition model, and the initial character recognition model recognizes the target sample image, and uses the sample verification code characters recognized from the target sample image as the prediction label; The model training module is used to iteratively train the initial character recognition model based on the predicted label and the sample label to obtain the target character recognition model as claimed in any one of claims 1-9.

13. A computer device, characterized in that, include: Processor and memory; The processor is connected to a memory, wherein the memory is used to store a computer program, and the processor is used to invoke the computer program to cause the computer device to perform the method according to any one of claims 1-10.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded and executed by a processor to cause a computer device having the processor to perform the method according to any one of claims 1-10.

Images