Collection account extraction method and device, equipment, medium and program product

Abstract

The invention provides a collection account extraction method and device, equipment, a medium and a program product, and the method comprises the steps: carrying out the navigation to a registration page of a target application through recognizing a specific element of the registration page; registering an account on a registration page by submitting registration information and verifying the verification code; and when it is monitored that the account number is registered successfully, a recharge page of the target application is navigated to by submitting a recharge order, and collection account number information is extracted from the recharge page. The collection account extraction method provided by the invention comprises the following steps: firstly, automatically and accurately navigating to a registration page of a target application; automatic registration operation is realized through registration information submission and verification code verification; and finally, automatically triggering a recharging order in a payment link, analyzing a recharging page structure, and extracting collection account information. The series of automatic operation breaks through the efficiency bottleneck of traditional manual operation, end-to-end automation from registration to account information acquisition is achieved, and meanwhile a large amount of manpower and time cost is saved.

Description

Technical Field

[0001] The present invention relates to the technical field of account extraction, and in particular to a method, device, equipment, medium and program product for extracting a payment account. Background Art

[0002] With the rapid development of internet platforms and digital payments, the number of risky applications leveraging these platforms has increased, and these applications are used to conduct unusual online activities. Due to their convenience and concealment, such activities often have far-reaching consequences without users being fully aware of them.

[0003] To ensure successful top-up for as many users as possible, even those unfamiliar with mobile payments, risky apps typically simplify the payment process: select a top-up method; choose the payment amount; and then be redirected to the appropriate payment channel to complete the payment. Platforms that provide fake payment accounts and redirect to real payment channels are known as fourth-party payment platforms. Compared to traditional third-party payment platforms, these platforms incorporate an additional party: the payment agent. The ability of risky apps to facilitate quick and convenient payment collection while concealing their true payment account information is entirely due to the effectiveness of fourth-party payment platforms.

[0004] A core strategy for addressing unusual online activity in risky applications is to block the inflow of funds. Given the diversity and complexity of risky applications, tracking and verifying deposit information is a complex process, often requiring multiple steps, including user registration and subsequent guidance. Consequently, manual review is often required to gather relevant information. However, relying on manual review lacks real-time response capabilities, is time-consuming, and is costly, making it difficult to rapidly respond to the rapidly growing number of risky applications. Summary of the Invention

[0005] The present invention provides a method, device, equipment, medium and program product for extracting a collection account, which is used to solve the problem that the existing technology manually reviews the recharge information of risk applications, which is time-consuming and costly, and is difficult to quickly respond to the rapidly growing risk applications.

[0006] In a first aspect, the present invention provides a method for extracting a payment account, comprising: Navigate to the registration page of the target application by identifying the unique elements of the registration page; Register an account on the registration page by submitting registration information and verifying the verification code; When the account registration is detected to be successful, a recharge order is submitted to navigate to the recharge page of the target application, and the payment account information is extracted from the recharge page.

[0007] In one embodiment, navigating to the registration page of the target application by identifying a registration page-specific element includes: Navigate to the target application's registration page to be identified; If the registration page to be identified includes a registration page-specific element, determining that the registration page to be identified is the registration page of the target application; If the registration page to be identified does not contain a registration page-specific element, a registration button is searched for in the registration page to be identified, and interference elements are eliminated during the search process. Based on the registration button, the registration page of the target application is navigated to.

[0008] In one embodiment, searching for a registration button in the registration page to be identified, eliminating interference elements during the search process, and navigating to the registration page of the target application based on the registration button includes: Traversing and clicking valid buttons in the registration page to be identified; the valid buttons include at least a registration button; After each click is detected, the page after the click is determined to be the registration page of the target application by identifying the unique elements of the registration page; If the page after the click is not the registration page of the target application, a close button is searched for on the page after the click, and the interference element in the page after the click is eliminated by clicking the close button.

[0009] In one embodiment, registering an account on the registration page by submitting registration information and verifying the verification code includes: Submit registration information on the registration page; If registration prompt information is displayed after submitting the registration information, identifying the prompt category of the registration prompt information; If the prompt type is a verification code prompt, the verification code is successfully verified based on the registration prompt information; If the prompt type is a prompt indicating that the registration information is incorrect, the registration information is modified, and the step of submitting the registration information on the registration page is iteratively executed until the verification code is successfully verified.

[0010] In one embodiment, the successfully verifying the verification code based on the registration prompt information includes: If the verification code image in the registration prompt information is a point-to-point verification code, segmenting the verification code image into a plurality of character image regions; Perform multi-angle rotation sampling on each character image area to obtain a rotation image set corresponding to each character image area; Performing character recognition based on each rotation angle image in each rotation image set to obtain a character recognition result for each character image area; Based on the requirement of clicking the verification code in the registration prompt information, the verification code is successfully verified by clicking on multiple target character recognition results.

[0011] In one embodiment, when character recognition is performed based on each rotation angle image in each rotation image set to obtain a character recognition result for each character image area, the following steps are performed for each character image area: Performing character recognition on each rotation angle image in the rotation image set to obtain a plurality of candidate characters for each rotation angle image and a confidence level of each candidate character; The confidence scores of the same character in each candidate character of each rotation angle image are weighted and summed to obtain target confidence scores of multiple different candidate characters; The candidate character corresponding to the highest target confidence is determined as the character recognition result of the character image area.

[0012] In one embodiment, the successfully verifying the verification code based on the registration prompt information further includes: If the verification code image in the registration prompt information is a slider verification code, inputting the mouse state space into a pre-trained reinforcement learning model to obtain the acceleration variation range of the mouse during sliding output by the pre-trained reinforcement learning model; the mouse state space is determined based on the mouse position and the mouse sliding trajectory; The verification code is successfully verified by controlling the mouse to slide according to the mouse sliding trajectory and the acceleration variation range.

[0013] In one embodiment, the target reward function used by the reinforcement learning model during pre-training is composed of a first reward function, a second reward function, and a third reward function; The first reward function is a function that gives a positive reward when the agent successfully passes the slider verification code and the sliding completion time is within a preset optimal time interval; The second reward function is a function that gives a negative reward when the agent fails to pass the slider verification code; The third reward function is a function that penalizes sudden acceleration changes when the agent slides.

[0014] In a second aspect, the present invention further provides a payment account extraction device, comprising: Pathfinding module, used to navigate to the registration page of the target application by identifying the unique elements of the registration page; A detection module is used to register an account on the registration page by submitting registration information and verifying the verification code; The extraction module is used to, when monitoring the successful registration of an account, navigate to the recharge page of the target application by submitting a recharge order, and extract the payment account information from the recharge page.

[0015] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described methods for extracting a payment account are implemented.

[0016] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the payment account extraction method as described in any one of the above are implemented.

[0017] In a fifth aspect, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by the processor, implements the steps of any of the above-mentioned payment account extraction methods.

[0018] The collection account extraction method, device, equipment, medium, and program product provided by the present invention first automatically navigate to the registration page of the target application based on page element feature recognition, which can effectively avoid the problem of different registration page routing methods for different applications and achieve accurate positioning of the registration page; then, by simulating real user behavior, the full registration process of registration information submission and verification code verification is completed to ensure the validity of the registered account; finally, in the payment link, the recharge order is automatically triggered and the recharge page structure is parsed to extract the collection account information embedded therein. This series of automated operations breaks through the efficiency bottleneck of traditional manual operations and realizes end-to-end automation from registration to account information acquisition. Especially in the case of rapidly growing risk applications, it can efficiently and accurately obtain the collection account of the target application while saving a lot of manpower and time costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 This is one of the flow charts of the payment account extraction method provided by the present invention.

[0021] Figure 2This is a payment flow diagram provided by the present invention.

[0022] Figure 3 It is a schematic diagram of the risk platform payment replacement method provided by the present invention.

[0023] Figure 4 This is the second flow chart of the payment account extraction method provided by the present invention.

[0024] Figure 5 It is a schematic diagram of traversing and searching for an effective button provided by the present invention.

[0025] Figure 6 Schematic diagram of interference element detection provided by the present invention.

[0026] Figure 7 It is a schematic diagram of the account combination provided by the present invention.

[0027] Figure 8 It is a schematic diagram of the multi-angle rotation of the character image area provided by the present invention.

[0028] Figure 9 It is a schematic diagram of the slider verification code provided by the present invention.

[0029] Figure 10 It is a structural diagram of the payment account extraction device provided by the present invention.

[0030] Figure 11 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0031] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0032] The terms "first," "second," and the like in the present invention are used to distinguish similar objects and are not used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present invention can be implemented in orders other than those illustrated or described herein.

[0033] The following combination Figures 1-11 The present invention describes the method, device, equipment, medium and program product for extracting the payment account provided by the present invention.

[0034] It's important to note that existing technologies rely on manual review, lack real-time response capabilities, and are time-consuming and costly to detect, making it difficult to quickly address rapidly growing risks. To address the inefficiencies and high costs associated with manual review, automated processes are being adopted. However, these processes also present challenges. First, the user interfaces and payment processes for different risk applications vary. Furthermore, registration, verification, and payment processes, as well as interface formats, vary across different fourth-party payment platforms. These technologies are poorly adaptable to website structures, and subtle changes can lead to ineffective recognition and data collection. Automated tools with fixed processes struggle to fully address these challenges. Second, current automated technologies focus on specific tasks, such as verification code recognition, resulting in limitations in the type of operations they perform and a lack of comprehensive solutions covering the entire process, from registration to top-up. Furthermore, current automated technologies are limited in their ability to process non-textual information, which remains a technical challenge. This is particularly true for non-textual information, such as image verification codes, as the diverse modalities make unified and efficient solutions lacking.

[0035] To overcome the above challenges, an embodiment of the present invention proposes a method for extracting payment accounts based on an intelligent automation framework. It is mainly aimed at extracting payment accounts for risky applications, and aims to efficiently parse and process-basedly crack down on fourth-party payment platforms related to various risky applications. By deeply analyzing the operating procedures of each risky application, general rules are extracted, and modular components are developed. These components can be flexibly combined to adapt to different automation needs. Each module focuses on a specific operation and adopts a variety of intelligent technologies such as image recognition, text matching, and reinforcement learning to ensure that the system can accurately process various complex page information. In addition, the method accurately tracks the payment account by simulating the actual user operation process, thereby effectively interrupting the flow of funds in the payment account. This method not only improves data processing efficiency, but also reduces maintenance costs, providing strong support for the risk governance platform.

[0036] It should be noted that all actions of acquiring information and data in the embodiments of the present invention are performed in compliance with the corresponding data protection laws and policies of the country where they are located and with authorization from the corresponding device owner.

[0037] The embodiment of the present invention takes the payment account extraction device as the execution body to describe the payment account extraction method.

[0038] Combine Figure 1 , Figure 1 This is one of the flow charts of the payment account extraction method provided by the present invention.

[0039] like Figure 1 As shown, the method for extracting the payment account includes the following steps: Step 101: Navigate to the registration page of the target application by identifying unique elements of the registration page; Step 102: Register an account on the registration page by submitting registration information and verifying the verification code; Step 103: When the account registration is detected to be successful, the recharge order is submitted to navigate to the recharge page of the target application, and the payment account information is extracted from the recharge page.

[0040] Specifically, with the continuous development of the mobile internet, mobile application functionality has become increasingly diverse. Numerous mobile application scenarios now involve services such as payment, transfers, and top-ups. Consequently, mobile payment has become an essential, fundamental function for mobile applications. In addition to traditional online banking and credit card payments, quick payment through third-party platforms has become a mainstream mobile payment method. Providing users with convenient and easy payment options is essential for risky applications. However, developers and operators of risky applications often need to conceal the flow of collected funds to evade tracking. Consequently, risky applications often rely on fourth-party payment platforms for fund collection. This example demonstrates how to automatically retrieve payment information from fourth-party payment platforms and locate it to specific accounts for effective control.

[0041] In the third-party payment process, the main participants include users, merchants and payment platforms. When a user places an order with a merchant and initiates the payment process, the user will jump to the payment platform under the guidance of the merchant to make the payment. After the user completes the payment operation, the merchant will ship the goods to the user; depending on the purchase method, the payment platform will complete the settlement of funds before or after the merchant completes the delivery. The biggest difference between fourth-party payment and third-party payment is that in addition to users, merchants (risk applications here) and payment platforms, the fourth-party payment platform is also added. At the same time, in order to hide the real payee, the fourth-party payment platform often uses a large number of fake merchants and collects payments in the form of account pools. The whole process is as follows Figure 2 As shown, Figure 2 This is a payment flow diagram provided by the present invention. In the fourth-party payment process, these fake merchant accounts are controlled by the fourth-party payment platform, and the fourth-party payment platform uses a large number of such fake merchant accounts on mobile phones to form a collection account pool, and uses a random collection account each time it collects money, in order to achieve the purpose of hiding the real fund transfer process. When the user completes the payment to the fake merchant account through the payment platform and the payment platform settles with the fake merchant account, the fourth-party payment platform then controls the fake merchant account to recover the funds and settle the funds with the actual risk application operator. Figure 3 As shown, Figure 3 It is a schematic diagram of the risk platform payment replacement method provided by the present invention.

[0042] A large number of fraudulent payment accounts are a key source of actual payment revenue for risky applications on fourth-party payment platforms. This embodiment proposes a highly flexible and intelligent automated payment account extraction solution, aiming to generalize its application to the management of as many risky applications as possible. By simulating the registration and top-up processes for risky applications through a variety of AI-powered automated algorithms, various modules can be combined to automate different processes across different websites.

[0043] The processing of different pages can be divided into the following three stages: accessing the registration page from the homepage via the pathfinding module, registering and verifying via the detection module, and extracting payment information from the top-up page via the extraction module. These three stages are composed of multiple modules, each handling a specific operation with a single, independent function. These modules include finding a specific button, clicking on it, identifying distracting elements, closing distracting elements, and entering registration information. Each module can be used independently and combined as needed. Figure 4 This is the second flow chart of the method for extracting the payment account provided by the present invention, combined with Figure 4 The implementation process of these three stages is described below.

[0044] First, identify the risky application from which to extract payment account information and use it as the target application for subsequent processes. This example can also be applied to normal applications.

[0045] Next, an HTTP request is generated to access the target application's registration portal, using the HTTP request to access the target application's registration portal URL. The "registration portal URL" is a specific web address used to access or register for a system, platform, or service. This address is typically included in the user interface, allowing users to click it to access the registration page and complete the registration process. Therefore, by accessing the registration portal URL, you can automatically navigate to the target application's registration page.

[0046] However, different apps may have different settings. For ease of use, some apps have a very clear registration entry URL that directly navigates to the registration page, which will immediately load the registration form, allowing users to complete registration by filling in their registration information. Some apps, for anti-crawl mechanisms, do not directly navigate to the registration page when accessing the registration entry. Instead, users navigate to another page, then find and click the registration button on that page to navigate to the registration page, load the registration form, and complete registration by filling in their registration information.

[0047] Therefore, after accessing the registration entry URL of the target application through an HTTP request, it is determined whether the page currently routed to is the registration page by identifying the unique elements of the registration page.

[0048] If yes, proceed to the next registration process; if not, search for the registration button on the current page, eliminate interference elements during the search process, click the registration button, navigate to the actual registration page, and then proceed to the next registration process.

[0049] After entering the registration page, look for input box elements and buttons that can submit the form.

[0050] First, enter the registration information in the input box element. The registration information usually includes the account number and password. Then, click the button to submit the form.

[0051] If you are directed to the target app's homepage after submitting your application, your account has been successfully registered. However, after submitting your registration information, there may be errors or a verification code may be required. In this case, a registration prompt will be displayed, indicating whether there are errors or a verification code is required.

[0052] Identify the prompt content in the registration prompt information. If the prompt content is about incorrect registration information, modify the registration information and resubmit it; if the prompt content is about the need to verify the verification code, verify the verification code based on the registration prompt information until the verification is successful.

[0053] When the registration information is entered correctly and the verification code is successfully verified, it means that the target application account has been successfully registered and enters the target application's homepage.

[0054] After entering the homepage of the target application, select a recharge order, find the button to submit the order on the recharge order page, and click the button to submit the recharge order. It will automatically navigate to the recharge page of the target application and extract the payment account information on the recharge page.

[0055] The payment account extraction method provided by the present invention first automatically navigates to the target application's registration page based on page element feature recognition, effectively avoiding the problem of different registration page routing methods for different applications and achieving precise positioning of the registration page. It then completes the full registration process of submitting registration information and verifying the verification code by simulating real user behavior, ensuring the validity of the registered account. Finally, it automatically triggers the recharge order during the payment process and parses the recharge page structure to extract the embedded payment account information. This series of automated operations breaks through the efficiency bottleneck of traditional manual operations and achieves end-to-end automation from registration to account information acquisition. Especially in the case of rapidly growing risky applications, it can efficiently and accurately obtain the target application's payment account, while saving a considerable amount of manpower and time costs.

[0056] In some embodiments, based on step 101, navigating to the registration page of the target application by identifying the unique elements of the registration page includes: Navigate to the target application's registration page to be identified; If the registration page to be identified includes a registration page-specific element, determining that the registration page to be identified is the registration page of the target application; If the registration page to be identified does not contain a registration page-specific element, a registration button is searched for in the registration page to be identified, and interference elements are eliminated during the search process. Based on the registration button, the registration page of the target application is navigated to.

[0057] The following Figure 4 The pathfinding module describes the process of navigating to the registration page and ensuring that the page is operational.

[0058] Specifically, an HTTP request is made to the target application's registration entry URL, accessing the target application's registration page to be identified. The Document Object Model (DOM) structure of the registration page to be identified is also loaded. The DOM structure represents a web document as a tree. In this tree, each Hypertext Markup Language (HTML) element is considered a node, and nodes have parent-child and sibling relationships.

[0059] When loading the registration entry URL, set an implicit wait time (such as driver.implicitly_wait(30)) and use an explicit wait (WebDriverWait) to monitor the DOM's document.readyState to ensure the page is fully loaded. If the page load times out or returns an error status code (such as 404), log the error and terminate the process.

[0060] Combined with the page DOM structure, identify the registration page-specific elements in the registration page to be identified and determine whether the registration page to be identified is the actual registration page. Three methods can be used to jointly determine the current page status, of course not limited to these three methods: (1) based on URL regular expressions (such as r' / register / ?$'); (2) page title keywords (such as "Register" and "Sign Up"); (3) parse whether there are registration page-specific elements in the DOM (such as <form id="register">or the text "User Agreement").

[0061] If a registration page-specific element is detected, the registration page to be identified is determined to be the registration page of the target application, and the process proceeds to the next step.

[0062] If no unique elements of the registration page are detected, a registration button is searched for in the registration page to be identified, and interference elements are eliminated during the search process. By clicking the registration button, the user navigates to the registration page of the target application.

[0063] The step of searching for a registration button on the registration page to be identified, eliminating interference elements during the search, and navigating to the registration page of the target application by clicking the registration button specifically includes the following steps: Traversing and clicking valid buttons in the registration page to be identified; the valid buttons include at least a registration button; After each click is detected, the page after the click is determined to be the registration page of the target application by identifying the unique elements of the registration page; If the page after the click is not the registration page of the target application, a close button is searched for on the page after the click, and the interference element in the page after the click is eliminated by clicking the close button.

[0064] Specifically, this example defines the mark on the web page that realizes a specific function by clicking as a button. 、 <button>After finding all the specific buttons that meet the requirements, simulate clicking them until you find a button that works. If you find a button that does not work after simulating clicking all the specific buttons, exit the program and end the process.

[0065] Figure 5 This is a schematic diagram of searching for an effective button provided by the present invention. Figure 5 As shown, buttons usually include image buttons, non-image buttons, submit buttons, etc. For image buttons, find all For label elements, you can search by keywords contained in class or scr parameters, then output the xpath paths of elements that meet the requirements, traverse these paths, and retain clickable buttons. For non-image buttons, find all leaf nodes, you can search by keywords contained in class or text, then output the xpath paths of elements that meet the requirements, traverse these paths, and retain clickable buttons. For submit buttons, find all leaf nodes, you can search by element type indicating submittable or class or attribute name containing submit or sub, then output the xpath paths of elements that meet the requirements, traverse these paths, and retain clickable buttons.

[0066] When an effective button is clicked, whether the page has changed can be determined by comparing the node lists of the page before the click and the page after the click.

[0067] To build a node list for a page, you first need to parse the HTML text of the page and then, through depth-first traversal, build a node list for all nodes in the form of [node label, object corresponding to the node itself, [list of child nodes]]. For example, there is currently a node A, and the object containing all its information is<Element A> , the node label of A is a, A has child nodes B and C, where B has no child nodes and C has child node D, and the label and object representation are the same as A. Then the output data format is [a,<Element A> ,[[b,<Element B> ,[]],[c,<Element C> ,[d,<Element D> ,[]]]].

[0068] When comparing pages before and after a click, a breadth-first traversal is used to compare each level, finding nodes that have changed, such as newly added or deleted nodes, and outputting them. In other words, if the number of nodes increases or decreases before and after the operation, the page is considered to have changed.

[0069] If the page has not changed, the adaptive retry mechanism can be triggered. After the first failure, it will fall back to the previous page (driver.back()) and re-execute the navigation process. If the cumulative number of failures is ≥ 2, the process will be terminated and an error code (HTTP 503) will be returned.

[0070] If the page changes, it could be a simulated click on the registration button or another distracting button. Therefore, it's necessary to first identify unique elements on the registration page to determine whether the page after the click is the target app's registration page.

[0071] If a registration page-specific element is detected, indicating a click on the registration button, the page after the click is determined to be the target app's registration page and the next step is taken. If no registration page-specific element is detected, indicating a click on another distracting button, the page will display the distracting element or hide the distracting layer, typically in the form of a pop-up window. A close button is then searched for on the page. Clicking the close button determines whether the distracting element has been successfully removed by determining whether the page has changed. If the page element changes after the click (e.g., whether the DOM node has been removed), the distracting element is successfully removed. The method for determining page changes is the same as described above. If no close button is found on the page, the page has not changed, and the distracting element was not successfully removed, an adaptive retry mechanism is triggered. After the first failure, the page will fall back to the previous page (driver.back()) and re-execute the navigation process. If the cumulative number of failures is ≥ 2, the process is terminated and an error code (HTTP 503) is returned.

[0072] Among them, interference elements generally include occlusion interference elements and non-occlusion interference elements.

[0073] For blocking interference elements, this example builds an exception handling module based on the fact that ignoring blocking interference elements will return an exception. If the program detects an exception return, it will parse the exception. The exception information usually contains relevant information about the blocking interference element, which can be used to help locate the interference element. After determining the location of the interference element, you can search for a specific button. For example, for a dynamically generated pop-up window on the page (such as the data-random-id attribute), iterate through all< / button> <button>Element, filters candidate buttons containing "Close" or "Cancel" text or class name, and performs simulated click tasks.

[0074] The difficulty with non-blocking distracting elements is that they don't receive the same feedback as blocking distracting elements, meaning it's impossible to determine whether they appear on the page. By default, these distracting elements are not present. Only when all current operations fail to take effect during program execution does the program enter the non-blocking distracting element processing module. This works similarly to the blocking distracting element processing module, but the non-blocking distracting element processing module lacks exception information to locate the distracting element. Therefore, this module searches for a close button across the entire page and simulates clicking to close the distracting element.

[0075] Figure 6 Schematic diagram of interference element detection provided by the present invention. Figure 6 As shown, this interference elimination strategy differs from common active detection solutions in that it adopts a passive detection and processing strategy, requiring no human intervention. This module is designed for pre-startup mode and is not normally activated. It is enabled only when processing mandatory interactive dialog boxes set up by risky applications to ensure the smooth execution of automated processes. It is optimized for the layout characteristics of such websites, aiming to conserve resources and improve operational efficiency. Furthermore, this solution has the ability to automatically restore the original operation, including automatic triggering, automatic exclusion, and automatic recovery, without the need for human intervention, thereby minimizing the interference of abnormal situations on the automated processes of illegal gambling applications.

[0076] This embodiment of the present invention detects whether the current page contains elements specific to a registration page. If not, it then traverses and clicks on valid buttons. It also verifies that the redirected page is indeed the registration page by identifying unique elements. If the redirect is incorrect, it automatically searches for and clicks on a close button to eliminate interference, ensuring precise targeting of the target registration page. This process significantly improves the accuracy and automation efficiency of registration page recognition, reducing the need for manual intervention.

[0077] In some embodiments, based on step 102, registering an account on the registration page by submitting registration information and verifying the verification code includes: Submit registration information on the registration page; If registration prompt information is displayed after submitting the registration information, identifying the prompt category of the registration prompt information; If the prompt type is a verification code prompt, the verification code is successfully verified based on the registration prompt information; If the prompt type is a prompt indicating that the registration information is incorrect, the registration information is modified, and the step of submitting the registration information on the registration page is iteratively executed until the verification code is successfully verified.

[0078] The following Figure 4 The process of the detection module obtaining the information required for registration and performing verification is described in detail.

[0079] Specifically, first, find all input box elements and buttons that can submit the form on the registration page, then enter the registration information in the input box, and submit the registration information by clicking the button that can submit the form. The account and password used to enter the registration information for the first time can be pre-set.

[0080] If you directly enter the target application's homepage after submitting the registration information, it means that the target application's account has been successfully registered.

[0081] If a registration prompt is displayed after submitting the registration information, identify the prompt type of the registration prompt. Registration prompts are usually presented in the form of pop-ups. Since pop-ups themselves do not have any features that can be used for positioning, the DOM structure mentioned above for determining page changes is used to locate pop-ups.

[0082] If the prompt type of the registration prompt information is a verification code prompt, return to the step of entering the registration information, modify the registration information, and continue the subsequent process.

[0083] There are two ways to modify registration information, specifically your account and password. One is to parse the registration prompt, which typically indicates why the current information doesn't meet the registration requirements. By analyzing this information, you can regenerate the account and password. Another approach is to build an account and password database based on a dynamic priority adjustment mechanism. This database stores accounts and passwords in a specific order, with the most likely accounts and passwords receiving the highest priority. If the current information doesn't meet the registration requirements, the database immediately switches to the next highest priority account and password and tries again.

[0084] For the first method, after clicking the submit button, the page change determination module is called to make a judgment. If a change occurs, the newly added nodes in the DOM structure are retrieved and parsed in sequence. The parsing process includes the following steps: First, all the text information in all nodes is found and divided into sentences, and only the Chinese part is used; second, a simple screening step is performed using registration keywords, such as user name, password, letters, numbers, etc.; third, regular expressions are used to further parse the sentences containing these keywords, and the account password is regenerated based on the parsing results. The pop-up information parsing step does not need to be very precise, it only needs to confirm that the newly added node does contain the registration prompt information.

[0085] For the second method, you first need to pre-build an account password library, set as P, where each account password combination Assigned an initial priority The initial priority can be set based on historical data or preset rules, such as The system can be configured to first try a six-digit string of random lowercase letters, then add two digits, then three digits, until it reaches eight digits. If that still doesn't work, add an uppercase letter. Based on the collected account and password requirements, the above possibilities can cover all possible illegal application accounts. Figure 7 This is a schematic diagram of the account combination provided by the present invention, such as Figure 7 An example of an account combination is given, in which the character strings in the column of lowercase English letters, the character strings in the column of lowercase English letters + numbers, and the character strings in the column of uppercase English letters + lowercase English letters + numbers are the contents in the account library.

[0086] During the automated registration process, real-time monitoring of the registration success rate of each account and password combination Based on the registration success rate, the priority of the account and password combination is dynamically adjusted, namely: ; Priority adjustment function It can be defined as follows: ; in, is the updated priority, is the current priority, Is the account password combination The registration success rate, Is the adjustment coefficient, which controls the magnitude of the priority adjustment. This is the target value or threshold for the success rate, used to measure the relative performance of the success rate. Furthermore, a random delay (time.sleep(random.uniform(0.1,0.3))) is introduced through the ActionChains(driver).send_keys_to_element(element,text).perform() chain operation to circumvent timing detection.

[0087] Specifically, if the registration prompt contains a verification code, the system retrieves the verification code image from the registration prompt, identifies the verification code type, and uses the appropriate verification method based on the verification code type. If verification succeeds, the system proceeds to the next step. If verification fails, the system re-verifies until verification succeeds and the next step is reached.

[0088] Common types of verification codes include character recognition verification codes, point-based verification codes, and slider verification codes. For character recognition verification codes, they can be directly identified by obtaining the verification code image in the registration prompt information, and then the recognized characters are entered into the input box. For point-based verification codes, a full-angle verification code recognition method based on a sampling attention mechanism is proposed. This method recognizes each character in the verification code image, and then clicks the corresponding character in the verification code image according to the point-based selection requirements in the registration prompt information. For slider verification codes, as an anti-crawl mechanism to combat risky applications, an adaptive mouse sliding trajectory generation method based on deep reinforcement learning is proposed. The mouse is controlled to slide along the mouse sliding trajectory.

[0089] This embodiment of the present invention automates the submission of registration information. If a verification code check is triggered, the verification code is automatically recognized and successfully verified. If the registration information is incorrect, the information is dynamically adjusted and resubmitted until the verification code check passes. This process significantly improves the registration success rate, reduces manual repetitive operations, and ensures the accuracy of the registration information and the reliability of the verification code check.

[0090] According to the above content, the successful verification of the verification code based on the registration prompt information includes: If the verification code image in the registration prompt information is a point-to-point verification code, segmenting the verification code image into a plurality of character image regions; Perform multi-angle rotation sampling on each character image area to obtain a rotation image set corresponding to each character image area; Performing character recognition based on each rotation angle image in each rotation image set to obtain a character recognition result for each character image area; Based on the requirement of clicking the verification code in the registration prompt information, the verification code is successfully verified by clicking on multiple target character recognition results.

[0091] Specifically, the verification code image I in the registration prompt information is obtained, and it is identified that the type of the verification code image is a click-type verification code.

[0092] Segment multiple character image regions from the verification code image , that is, the image area containing the entire single character, constitutes the character image set S, .

[0093] Create an empty candidate set C to store possible characters and their related information.

[0094] Each character image area is sampled by multi-angle rotation, for example, starting from 0°, with a step length of 10°, 36 equal-angle rotation samples are performed to rotate the character image area counterclockwise. To angle θ, generate the rotated image . Figure 8 This is a schematic diagram of the multi-angle rotation of the character image area provided by the present invention, such as Figure 8 As shown in FIG, the character image area starts from 0° and is sampled with 10° as a step, and is rotated 36 times with equal angles.

[0095] Then the rotated image Perform optical character recognition (OCR). The OCR engine samples the top three candidate characters and their confidence scores (c1, p1), (c2, p2), and (c3, p3). The combination of these candidate characters, rotation angle, and confidence score (c, θ, p) is added to the candidate set C. The probability distribution of the candidate characters for each rotated image is shown in Table 1 below.

[0096] Table 1

[0097] For each candidate character in the candidate set C, the confidence of the same character is weighted and summed to obtain the target confidence of multiple different candidate characters. The expression is as follows: in, is the target confidence; is the confidence of each candidate character; is the rotation angle of the image corresponding to the identified candidate character; is the optimal angle, that is, the rotation angle at which character c achieves the highest confidence.

[0098] The confidence function w(c) based on the sampling attention mechanism is the recognition probability of character c at each rotation angle under sampling. Based on this, its attention weight mainly considers the optimal angle Neighborhood attenuation factor.

[0099] Then the target confidence of each different candidate character is compared, and the candidate character corresponding to the highest target confidence is taken as the character image area. The character recognition results.

[0100] The character recognition result of each character image area can be identified in the above manner.

[0101] Based on the registration prompt's requirement to click on the verification code, the verification code was successfully verified by clicking on the required multiple target character recognition results in the verification code image. Considering the user behavior characteristics of risky apps, which often exhibit slight deviations and delays during operation, a simulated mouse micro-positioning offset and a random click interval of 1-2 seconds were set to closely simulate the operating habits of risky app users, thereby improving the pass rate of automatic verification code verification.

[0102] The embodiment of the present invention dynamically adjusts the distribution weights of candidate characters at each rotation angle through a sampling attention mechanism, thereby improving the accuracy and robustness of recognition. At the same time, the method of candidate character sampling and character recognition can reduce computational complexity.

[0103] According to the above content, the successful verification of the verification code based on the registration prompt information also includes: If the verification code image in the registration prompt information is a slider verification code, inputting the mouse state space into a pre-trained reinforcement learning model to obtain the acceleration variation range of the mouse during sliding output by the pre-trained reinforcement learning model; the mouse state space is determined based on the mouse position and the mouse sliding trajectory; The verification code is successfully verified by controlling the mouse to slide according to the mouse sliding trajectory and the acceleration variation range.

[0104] Specifically, the verification code image in the registration prompt information is obtained, and it is identified that the type of the verification code image is a click-type verification code.

[0105] In the slider environment, the mouse position and trajectory are fed into a pre-trained reinforcement learning model. The mouse, acting as an agent, interacts with the verification code environment through trial and error, continuously learning the optimal strategy and ultimately outputting the acceleration range of the mouse during sliding. The verification code is then successfully verified by controlling the mouse to slide according to the trajectory and acceleration range.

[0106] During the reinforcement learning model training process, first of all, it is necessary to set the basic components of reinforcement learning in risk application scenarios.

[0107] Round: Set the action of the user clicking the mouse button to start sliding the slider as the beginning of a round, and if the user releases the mouse button, the round is considered to be over.

[0108] State space: Introducing second-order kinematic features into state space modeling, that is, time, mouse position, velocity, and acceleration are used as states, namely: ; In order to express the cumulative movement time of the mouse at each time point, it is recorded as , The value of , Represents the final time point of the observation period, which is a parameter that changes according to specific circumstances. is the normalized time factor. Considering that when users of risky applications slide the verification code slider, the mouse movement path is usually not completely along the horizontal straight line, but will be offset in the vertical direction. Therefore, when defining the state space, the mouse position information is taken into account. Specifically, the position information is expressed as the pixel distance relative to the starting point, which is recorded as . Accordingly, Represents the mouse in The absolute velocity in the horizontal and vertical directions and the absolute acceleration in the horizontal and vertical directions at a point in time.

[0109] In addition, the result indicator will be finally verified by the slider. Added to the state space definition, 1 means verification is passed. A value of 0 indicates that the verification has not been passed. The final state space (as shown in the figure) is defined as: .

[0110] Figure 9 This is a schematic diagram of the slider verification code provided by the present invention, such as Figure 9 As shown, the slider verification process needs to slide from the starting point to the end point. Each time point in the mouse sliding process has a corresponding mouse position, speed and acceleration.

[0111] Action space: To simulate the sliding trajectory of users in risk applications when performing slider verification, the action space Set as the range of change of the mouse acceleration in the horizontal and vertical directions. Since the acceleration of the mouse movement will not be fixed at a specific value within a unit time interval, a variation range is set to represent this uncertainty. By collecting and analyzing the sliding trajectory data during the slider verification process in some risk applications, the acceleration is determined. The value range is [-500pixel / s², 500pixel / s²]. In order to simplify the model, a discretized acceleration control strategy is defined. ,in Indicates The discretization granularity is determined by statistical analysis of measured data. ,For example When , the action strategy satisfies: .

[0112] State transfer function: the state transfer function used by the agent It enables it to make decisions based on the current state and select corresponding actions. Subsequently, a state transition equation based on dynamics is established.

[0113] This process is repeated until the agent successfully controls the mouse to move the slider to the specified position.

[0114] Reward Function: For the risky application slider verification code scenario, the result of the agent passing or failing the verification is considered as direct feedback on its strategy. If the agent successfully passes the verification, it is rewarded with a high score of +1000; conversely, if the verification fails, the reward score is -1. In addition, considering that most users require a single slider verification to complete the time window of [0.5s, 2s], the reward score is set to a negative number for passing the verification outside this time window. Therefore, a composite reward function is proposed, which includes a first reward function, a second reward function, and a third reward function. The first reward function provides a positive reward when the agent successfully passes the slider verification code and the sliding completion time is within a preset optimal time interval. The second reward function provides a negative reward when the agent fails the slider verification code. The third reward function penalizes sudden acceleration changes when the agent slides.

[0115] The target reward function can be defined as: in, is the optimal time parameter (such as 1.25s), is the time penalty coefficient (such as 0.8), is the acceleration change modulus.

[0116] After setting up the basic components of reinforcement learning, we use the improved TD3 algorithm to build a dual-critic evaluation network: ; Furthermore, the Actor Policy Network uses a Gaussian policy. The decision action output by this neural network is the current input mouse position, velocity, and acceleration. The model takes the current user mouse state as input, calculates the optimal acceleration, predicts the possible reward, and updates the neural network parameters using the reward function and feedback to ensure that the expected reward is close to the actual reward.

[0117] During the reinforcement learning training process, the original loss of the TD3 algorithm Under this condition, the trajectory smoothness constraint term is innovatively introduced , the total loss function is: in, is the smoothing coefficient. In this way, the model not only learns the user's behavioral characteristics, but also understands the impact of these behaviors on the slider verification code results, thereby generating mouse trajectories that are closer to real user behavior in a real network environment.

[0118] This embodiment of the present invention incorporates second-order kinematic features into state-space modeling, establishes a composite reward function with time constraints, and proposes an action space discretization strategy based on acceleration constraints. By constructing a continuous action space reinforcement learning model with spatiotemporal constraints, it achieves adversarial imitation of real user behavior characteristics, resulting in a high success rate and stability of the reinforcement learning model during the sliding verification process.

[0119] In some embodiments, based on step 103, the recharge order is submitted, the recharge page of the target application is navigated to, and the payment account information is extracted from the recharge page, specifically including the following content.

[0120] The following Figure 4 The process of extracting the module to locate the recharge page and outputting the payment account information is described in detail.

[0121] Specifically, select a recharge order, search for the recharge button on the recharge order page, traverse all found button elements, and simulate clicks. If all button elements are simulated clicked but the buttons are invalid, exit the program and end the process.

[0122] If a valid button is clicked, the page after the click is determined to be a recharge page. The method for determining whether it is a recharge page is similar to the method for determining whether it is a registration page, namely: (1) URL regular expression (such as target_url + ' / recharge'); (2) page title keywords (such as containing "recharge"); (3) one or more features of DOM elements (such as XPath partial matching ( / / a[contains(@href,'recharge')]) combined with text similarity calculation (Levenshtein Distance ≤ 2) and fuzzy positioning) to comprehensively determine whether the current page is a recharge page.

[0123] If the current page is not the recharge page, the adaptive retry mechanism can be triggered. After the first failure, it will fall back to the previous page (driver.back()) and re-execute the process. If the cumulative number of failures is ≥ 2, the process will be terminated and an error code (HTTP 503) will be returned.

[0124] If the current page is a recharge page, parse the HTML elements of the web page, locate the bank card information block through CSS selectors (such as .bank-card>ul>li:nth-child(2)), and use regular expressions (such as r'\b(?:62[0-9]{14,17}|(?:4\d{3}|5[1-5]\d{2}|6011)[ -]?\d{4}[ -]?\d{4}[ -]?\d{3,5})\b') to find and output the payment account information on the recharge page.

[0125] In summary, all the above embodiments simulate the entire registration and recharge process of users of risk applications based on process automation, solving the problem of insufficient real-time adaptability caused by long manual simulation time and high human resource costs. It automatically obtains payment account information and locates specific accounts for subsequent interception or control, which is of great significance and value.

[0126] Furthermore, the full registration and top-up process covers the entire registration, verification, and top-up process. Compared to existing solutions that only address specific issues, this design provides a highly integrated solution, significantly expanding the breadth of operations. Loosely coupled communication between modules is achieved through status codes (such as HTTP 200 / 503) and exception flags, ensuring high fault tolerance and scalability.

[0127] In addition, based on the in-depth analysis and internal law extraction of information from various websites, this design uses advanced algorithms to deeply analyze data from multiple source websites and extract common element positioning rules from them. This method, which does not rely on fixed positions or attribute identifiers, enhances the adaptability and stability of the system, and can maintain efficient operation even in the face of frequently updated web page structures.

[0128] The payment account extraction device provided by the present invention is described below. The payment account extraction device described below and the payment account extraction method described above can be referenced to each other.

[0129] Reference Figure 10 , Figure 10 It is a structural diagram of the payment account extraction device provided by the present invention.

[0130] The payment account extraction device includes: The pathfinding module 1010 is used to navigate to the registration page of the target application by identifying unique elements of the registration page.

[0131] The detection module 1020 is used to register an account on the registration page by submitting registration information and verifying the verification code.

[0132] The extraction module 1030 is used to, when detecting that the account registration is successful, navigate to the recharge page of the target application by submitting a recharge order, and extract the payment account information from the recharge page.

[0133] The payment account extraction device provided by the present invention first automatically navigates to the target application's registration page based on page element feature recognition, effectively avoiding the problem of different registration page routing methods for different applications and achieving precise positioning of the registration page. It then completes the full registration process of submitting registration information and verifying the verification code by simulating real user behavior, ensuring the validity of the registered account. Finally, it automatically triggers the recharge order during the payment process and parses the recharge page structure to extract the payment account information embedded therein. This series of automated operations breaks through the efficiency bottleneck of traditional manual operations and achieves end-to-end automation from registration to account information acquisition. Especially in the case of rapidly growing risk applications, it can efficiently and accurately obtain the target application's payment account, while saving a lot of manpower and time costs.

[0134] Furthermore, the pathfinding module 1010 is further configured to: Navigate to the target application's registration page to be identified; If the registration page to be identified includes a registration page-specific element, determining that the registration page to be identified is the registration page of the target application; If the registration page to be identified does not contain a registration page-specific element, a registration button is searched for in the registration page to be identified, and interference elements are eliminated during the search process. Based on the registration button, the registration page of the target application is navigated to.

[0135] Furthermore, the pathfinding module 1010 is further configured to: Traversing and clicking valid buttons in the registration page to be identified; the valid buttons include at least a registration button; After each click is detected, the page after the click is determined to be the registration page of the target application by identifying the unique elements of the registration page; If the page after the click is not the registration page of the target application, a close button is searched for on the page after the click, and the interference element in the page after the click is eliminated by clicking the close button.

[0136] Furthermore, the detection module 1020 is further configured to: Submit registration information on the registration page; If registration prompt information is displayed after submitting the registration information, identifying the prompt category of the registration prompt information; If the prompt type is a verification code prompt, the verification code is successfully verified based on the registration prompt information; If the prompt type is a prompt indicating that the registration information is incorrect, the registration information is modified, and the step of submitting the registration information on the registration page is iteratively executed until the verification code is successfully verified.

[0137] Furthermore, the detection module 1020 is further configured to: If the verification code image in the registration prompt information is a point-to-point verification code, segmenting the verification code image into a plurality of character image regions; Perform multi-angle rotation sampling on each character image area to obtain a rotation image set corresponding to each character image area; Performing character recognition based on each rotation angle image in each rotation image set to obtain a character recognition result for each character image area; Based on the requirement of clicking the verification code in the registration prompt information, the verification code is successfully verified by clicking on multiple target character recognition results.

[0138] Furthermore, the detection module 1020 is further configured to: Performing character recognition on each rotation angle image in the rotation image set to obtain a plurality of candidate characters for each rotation angle image and a confidence level of each candidate character; The confidence scores of the same character in each candidate character of each rotation angle image are weighted and summed to obtain target confidence scores of multiple different candidate characters; The candidate character corresponding to the highest target confidence is determined as the character recognition result of the character image area.

[0139] Furthermore, the detection module 1020 is further configured to: If the verification code image in the registration prompt information is a slider verification code, inputting the mouse state space into a pre-trained reinforcement learning model to obtain the acceleration variation range of the mouse during sliding output by the pre-trained reinforcement learning model; the mouse state space is determined based on the mouse position and the mouse sliding trajectory; The verification code is successfully verified by controlling the mouse to slide according to the mouse sliding trajectory and the acceleration variation range.

[0140] It should be noted that the payment account extraction device provided by the present invention can execute the payment account extraction method described in any of the above embodiments during specific operation, which will not be described in detail in this embodiment.

[0141] Figure 11 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 11 As shown, the electronic device may include: a processor 1110, a communications interface 1120, a memory 1130, and a communications bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other via the communications bus 1140. The processor 1110 may invoke logic instructions in the memory 1130 to execute a method for extracting a payment account. The method includes: navigating to the registration page of a target application by identifying unique elements of the registration page; registering an account on the registration page by submitting registration information and verifying a verification code; and upon detecting successful account registration, navigating to the recharge page of the target application by submitting a recharge order, and extracting the payment account information on the recharge page.

[0142] Furthermore, the logic instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0143] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the payment account extraction method provided by the above-mentioned embodiments, the method including: navigating to the registration page of the target application by identifying the unique elements of the registration page; registering an account on the registration page by submitting registration information and verifying the verification code; when monitoring that the account registration is successful, navigating to the recharge page of the target application by submitting a recharge order, and extracting the payment account information on the recharge page.

[0144] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute the payment account extraction method provided in the above-mentioned embodiments, the method comprising: navigating to the registration page of the target application by identifying unique elements of the registration page; registering an account on the registration page by submitting registration information and verifying a verification code; when monitoring that the account registration is successful, navigating to the recharge page of the target application by submitting a recharge order, and extracting the payment account information on the recharge page.

[0145] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Those skilled in the art can understand and implement the present invention without inventive effort.

[0146] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.< / button> < / form>

Claims

1. A method for extracting a payment account, characterized in that: The method for extracting the payment account includes: Navigate to the registration page of the target application by identifying unique elements of the registration page; Register an account on the registration page by submitting registration information and verifying the verification code; When the account registration is detected to be successful, a recharge order is submitted to navigate to the recharge page of the target application, and the payment account information is extracted from the recharge page.

2. The method for extracting a payment account according to claim 1, wherein: The step of navigating to the registration page of the target application by identifying unique elements of the registration page includes: Navigate to the target application's registration page to be identified; If the registration page to be identified includes a registration page-specific element, determining that the registration page to be identified is the registration page of the target application; If the registration page to be identified does not contain a registration page-specific element, a registration button is searched for in the registration page to be identified, and interference elements are eliminated during the search process. Based on the registration button, the registration page of the target application is navigated to.

3. The method for extracting a payment account according to claim 2, wherein: The step of searching for a registration button in the registration page to be identified, eliminating interference elements during the search process, and navigating to the registration page of the target application based on the registration button includes: Traversing and clicking valid buttons in the registration page to be identified; the valid buttons include at least a registration button; After each click is detected, the page after the click is determined to be the registration page of the target application by identifying the unique elements of the registration page; If the page after the click is not the registration page of the target application, a close button is searched for on the page after the click, and the interference element in the page after the click is eliminated by clicking the close button.

4. The method for extracting a payment account according to claim 1, wherein: Registering an account on the registration page by submitting registration information and verifying the verification code includes: Submit registration information on the registration page; If registration prompt information is displayed after submitting the registration information, identifying the prompt category of the registration prompt information; If the prompt type is a verification code prompt, the verification code is successfully verified based on the registration prompt information; If the prompt type is a prompt indicating that the registration information is incorrect, the registration information is modified, and the step of submitting the registration information on the registration page is iteratively executed until the verification code is successfully verified.

5. The method for extracting a payment account according to claim 4, wherein: The successful verification of the verification code based on the registration prompt information includes: If the verification code image in the registration prompt information is a point-to-point verification code, segmenting the verification code image into a plurality of character image regions; Perform multi-angle rotation sampling on each character image area to obtain a rotation image set corresponding to each character image area; Performing character recognition based on each rotation angle image in each rotation image set to obtain a character recognition result for each character image area; Based on the requirement of clicking the verification code in the registration prompt information, the verification code is successfully verified by clicking on multiple target character recognition results.

6. The method for extracting a payment account according to claim 5, wherein: When character recognition is performed based on each rotation angle image in each rotation image set to obtain a character recognition result for each character image area, the following steps are performed for each character image area: Performing character recognition on each rotation angle image in the rotation image set to obtain a plurality of candidate characters for each rotation angle image and a confidence level of each candidate character; The confidence scores of the same character in each candidate character of each rotation angle image are weighted and summed to obtain target confidence scores of multiple different candidate characters; The candidate character corresponding to the highest target confidence is determined as the character recognition result of the character image area.

7. The method for extracting a payment account according to claim 4, wherein: The successful verification of the verification code based on the registration prompt information further includes: If the verification code image in the registration prompt information is a slider verification code, inputting the mouse state space into a pre-trained reinforcement learning model to obtain the acceleration variation range of the mouse during sliding output by the pre-trained reinforcement learning model; the mouse state space is determined based on the mouse position and the mouse sliding trajectory; The verification code is successfully verified by controlling the mouse to slide according to the mouse sliding trajectory and the acceleration variation range.

8. The method for extracting a payment account according to claim 7, wherein: The target reward function used by the reinforcement learning model in the pre-training process is composed of a first reward function, a second reward function and a third reward function; The first reward function is a function that gives a positive reward when the agent successfully passes the slider verification code and the sliding completion time is within a preset optimal time interval; The second reward function is a function that gives a negative reward when the agent fails to pass the slider verification code; The third reward function is a function that penalizes sudden acceleration changes when the agent slides.

9. A payment account extraction device, characterized in that: include: Pathfinding module, used to navigate to the registration page of the target application by identifying the unique elements of the registration page; A detection module is used to register an account on the registration page by submitting registration information and verifying the verification code; The extraction module is used to, when monitoring the successful registration of an account, navigate to the recharge page of the target application by submitting a recharge order, and extract the payment account information from the recharge page.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method for extracting the payment account number according to any one of claims 1 to 8 are implemented.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for extracting a payment account number according to any one of claims 1 to 8 are implemented.

12. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for extracting a payment account number according to any one of claims 1 to 8 are implemented.

Images