Verification confrontation method and system for social robot

Abstract

The invention discloses a verification confrontation method and system for a social robot. The method comprises the following steps: collecting interaction data of a plurality of social platforms in registration, login and verification processes, and identifying a verification type as an image type or a sliding block type; aiming at the sliding block type verification code, extracting a target area and an initial area by utilizing a target detection model, generating an initial sliding block track route by utilizing a Bezier curve, and comparing and optimizing the initial sliding block track route with a historical track library; for image verification codes, the recognition accuracy is improved by adopting an image processing technology; receiving a short message verification code through an API in butt joint with a virtual mobile phone service platform, and extracting a mailbox verification code by using a semantic target extraction model; and storing the registration data of each account. According to the method, a dynamic closed-loop verification confrontation system is formed, so that the adaptability and the operation concealment of the social robot in a complex verification environment are improved.

Description

TECHNICAL FIELD

[0001] The application relates to a verification confrontation method for a social robot and also relates to a corresponding verification confrontation system, and belongs to the technical field of data identification. BACKGROUND

[0002] Under the impetus of artificial intelligence technology, social robots have gradually become key roles in various online services. Especially in the scenes of automatic registration, information interaction, public opinion monitoring and data collection, the application prospect of social robots is broad. It not only can improve the efficiency of information flow, but also can provide support for business intelligence analysis, and plays an important role in optimizing user services, and has significant social value.

[0003] However, the operation of social robots faces some challenges. In order to prevent malicious operations and false registrations and other behaviors, various social platforms have strengthened the identity verification mechanism. They commonly use multiple verification methods, such as image-based verification codes, sliding block-based verification codes, short message verification and email verification, to limit user behavior. Although these verification mechanisms have played a positive role in enhancing platform security, they have also hindered the normal operation of social robots. Especially in the legal automated application scene, these verification mechanisms have increased the difficulty of interaction and formed a high barrier.

[0004] At present, most verification confrontation systems of social robots have obvious limitations. On the one hand, these systems can only deal with a single type of verification, such as only being able to deal with image-based verification codes or sliding block-based verification codes, and lack unified identification and processing capabilities for multiple verification types, resulting in poor adaptability. On the other hand, when dealing with sliding block-based verification codes, some bypass methods fail to combine image recognition models to accurately locate target areas, or only use simple trajectory simulation methods. This makes the generated trajectory lack the characteristics of real operation and is easily identified and screened by the platform. In addition, when dealing with short message verification and email verification information, existing technologies mostly rely on fixed rules or manual configuration, lack intelligent information extraction mechanisms, and are difficult to adapt to the diversification of verification code content and form. SUMMARY

[0005] The primary technical problem to be solved by the present application is to provide a verification confrontation method for a social robot.

[0006] Another technical problem to be solved by the present application is to provide a verification confrontation system for a social robot.

[0007] To achieve the above technical purposes, the technical scheme adopted by the present application is as follows:

[0008] According to a first aspect of an embodiment of the present application, a verification confrontation method for a social robot is provided, comprising the following steps:

[0009] S100: Collecting interaction data of a plurality of social platforms in a registration, login and verification process, wherein the interaction data comprises a verification code triggering condition and a verification type;

[0010] S200: Identifying the verification type, wherein the verification type comprises a slider type;

[0011] S300: When the verification type is the slider type, extracting a target region and a starting region by using a target detection model, generating an initial slider track route by using a Bezier curve according to coordinates of the target region and the starting region, comparing the initial slider track route with a historical route dataset, determining whether to adjust the initial slider track route according to a comparison result, and determining a final slider track route after adjustment;

[0012] S400: Obtaining information in the verification code;

[0013] S500: Storing registration data of each account.

[0014] Preferably, the step of extracting the target region and the starting region by using the target detection model comprises the following steps:

[0015] Obtaining historical slider verification data according to the interaction data, extracting a historical slider background image and a historical slider block image, and establishing a historical dataset;

[0016] Labeling the historical dataset, wherein the labeling target comprises a gap region and a slider block;

[0017] Sampling the historical dataset according to a preset proportion to obtain a training subset and a test subset, obtaining an initial target detection model, iteratively training the initial target detection model according to the training subset, and evaluating the target detection model after iterative training according to the test subset;

[0018] If a confidence degree of the target detection model after current iterative training is less than a confidence degree of the target detection model after previous iterative training, the target detection model after previous iterative training is used as the target detection model; if the confidence degree of the target detection model after current iterative training is greater than or equal to the confidence degree of the target detection model after previous iterative training, a variation amplitude of the target detection model in a gradient direction is increased, and the iterative training is continued until a preset iteration number is reached or the confidence degree of the target detection model after iterative training is less than the confidence degree of the target detection model after previous iterative training, and finally the target detection model is obtained;

[0019] Inputting current slider graphical data into the target detection model, and extracting a first detection box and a second detection box from an output result;

[0020] The target region is determined according to the first detection frame, and the starting region is determined according to the second detection frame.

[0021] In the running stage, the slider image to be recognized is input into the trained target detection model; two detection frames are marked in the model output result: the first detection frame corresponds to the gap region of the verification code, serving as the target region of the slider; the second detection frame corresponds to the initial position of the slider, serving as the starting region; the center coordinates of the two regions are taken as the key reference points for trajectory generation;

[0022] The center coordinates of the starting region are taken as the starting point of the trajectory, a horizontal line is drawn from the starting point, and a vertical line is drawn from the center point of the target region, and the intersection point of the horizontal line and the vertical line is taken as the end point; two offset points are randomly generated between the starting point and the end point, and the offset points are offset by a predetermined number of pixels in the vertical direction to simulate the slight jitter that may occur in human hand operation;

[0023] In combination with the starting point, the two offset points and the end point, a smooth trajectory line with real operation characteristics is fitted using a Bezier curve;

[0024] The final initial slider trajectory route is synthesized based on all the trajectory points, which is used for subsequent comparison and adjustment.

[0025] More preferably, the initial slider trajectory route is compared with the historical route dataset, and whether to adjust the initial slider trajectory route is determined according to the comparison result, which specifically includes:

[0026] A first average coincidence degree of the initial slider trajectory route and the historical successful verification trajectory route set is obtained; a second average coincidence degree of the initial slider trajectory route and the historical failed verification trajectory route set is obtained;

[0027] When the first average coincidence degree is greater than or equal to the second average coincidence degree, it is determined that the initial slider trajectory route is not adjusted, and the initial slider trajectory route is taken as the final slider trajectory route; when the first average coincidence degree is less than the second average coincidence degree, it is determined that the initial slider trajectory route needs to be adjusted, and the final slider trajectory route is determined after adjustment.

[0028] More preferably, after the final slider trajectory route is determined, it further includes:

[0029] Based on a nonlinear speed model, the sliding speed between each adjacent trajectory point is determined;

[0030] According to the positions of the adjacent trajectory points in the final slider trajectory route, a maximum sliding speed set at the corresponding positions in the historical successful verification trajectory route set is determined;

[0031] According to the final slider track route, the sliding speed between each adjacent track point, and the maximum sliding speed, a slider verification operation is performed; if the sliding speed exceeds the maximum sliding speed, the operation is performed at the maximum sliding speed.

[0032] According to a second aspect of the embodiments of the present application, another verification confrontation method for a social robot is provided, comprising the following steps:

[0033] S100: Collecting interactive data of a plurality of social platforms in a registration, login and verification process, wherein the interactive data comprises a verification code triggering condition and a verification type;

[0034] S200': Identifying the verification type, wherein the verification type comprises an image type;

[0035] S300': When the verification type is the image type, identifying the verification content based on image processing, comprising the following sub-steps: S301: converting the obtained color verification code image into a gray-scale image; S302: taking any point in the gray-scale image as an origin, performing Gaussian weighted average processing on the surrounding pixel region, thereby dynamically calculating the pixel threshold value of the region to obtain an optimized gray-scale image; S303: analyzing the optimized gray-scale image based on an edge detection algorithm to identify the obvious structural edges in the image, thereby determining the contour region where the verification code may exist; S304: when the identified contour region is a text verification code, further performing image correction processing; S305: if the verification code contour region is a graphic type, comparing the extracted region features with the standard images in a graphic database to further determine the graphic type or matching content of the current verification code;

[0036] S400: Obtaining information in the verification code;

[0037] S500: Storing the registration data of each account.

[0038] Preferably, the step S304 specifically comprises:

[0039] The overall inclination angle of the verification code region is calculated by identifying the minimum circumscribed rectangle of each character block, and a rotation matrix is constructed for rotation correction based on the inclination angle, thereby ensuring that the characters are arranged in a normal horizontal direction;

[0040] The center coordinates of each character region are extracted, and all the center points are connected;

[0041] If the connection line presents a nonlinear bending trend, it indicates that the character arrangement is distorted;

[0042] The connection line is modeled based on polynomial fitting, and the character positions are aligned and translated according to the fitting result.

[0043] More preferably, in the step S400, the API is used to access multiple virtual mobile phone service platforms to receive and extract the short message to obtain the information in the verification code.

[0044] More preferably, in the step S400, the mailbox account used for registration is accessed, and a semantic target extraction model is used to extract the content of the mail body or title to obtain the information in the verification code.

[0045] According to a third aspect of the embodiments of the present application, a verification confrontation system for a social robot is provided for executing the verification confrontation method for a social robot as described above, comprising:

[0046] The collection unit is configured to collect interactive data of multiple social platforms in the registration, login and verification processes, wherein the interactive data includes a verification code trigger condition and a verification type.

[0047] The verification code recognition unit is configured to recognize the verification type, wherein the verification type includes an image type and a sliding block type.

[0048] The short message and mailbox processing unit is configured to access multiple virtual mobile phone service platforms based on an API to receive and extract a short message verification code, or access a mailbox account used for registration, and use a semantic target extraction model to extract the content of the mail body or title to obtain the information in the verification code.

[0049] The account management unit is configured to store the registration data of each account.

[0050] Compared with the prior art, the present application has the following technical effects:

[0051] (1) In the processing of the sliding block type verification code, the target detection model is used to accurately locate the target area and the starting area, the initial sliding block trajectory is generated combined with the Bezier curve, and the historical trajectory library is compared and optimized to make the sliding block trajectory more close to the real user operation.

[0052] (2) In the processing of the image type verification code, adaptive gray processing, Gaussian weighting, edge detection, rotation correction and polynomial fitting image processing techniques are used to improve the accuracy and scope of the verification code recognition.

[0053] (3) In the processing of the short message and mailbox verification code, the virtual mobile phone service platform is accessed based on the API, and the semantic target extraction model is used to extract the verification code to adapt to the diversification of the verification code format. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 A structural block diagram of a verification confrontation system for a social robot is provided for the first embodiment of the present application.

[0055] Figure 2 A flowchart of a verification confrontation method for a social robot according to a second embodiment of the present application is provided.

[0056] Figure 3 A flowchart of a verification confrontation method for a social robot according to a third embodiment of the present application is provided. DETAILED DESCRIPTION

[0057] The technical content of the present application will be described in detail below in combination with the drawings and specific embodiments.

[0058] The technical concept in the embodiments of the present application is to collect interactive data in the registration, login and verification process of multiple social platforms, including the verification code trigger condition and verification type, to solve the single type limitation problem of social robots when facing different verification types. In the processing of the slider type verification code, the target detection model is used to accurately locate the target area and the starting area, and then the Bezier curve is used to generate a high simulation track, and the dynamic track optimization is combined to improve the simulation degree of the slider. At the same time, the semantic target extraction model is used to extract text verification information, which is suitable for its format diversity, so as to form a dynamic closed-loop verification confrontation system, and improve the technical effects of the automatic adaptation ability and operation authenticity of the social robot.

[0059] First embodiment

[0060] As shown in Figure 1 The first embodiment of the present application provides a verification confrontation system for a social robot, which at least includes an acquisition unit, a verification code recognition unit, an SMS mailbox processing unit and an account management unit. Among them,

[0061] The acquisition unit is responsible for collecting all interactive data generated in the registration, login and verification process of multiple social platforms, which explicitly contains the trigger condition of the verification code and the type information of the verification code; through the system collection and analysis of these interactive data, key data support is provided for subsequent identification and response to different verification code types, and dynamic adaptation to the platform verification mechanism is realized.

[0062] The verification code recognition unit is configured to identify the verification type, and in an embodiment of the present application, the verification type includes image type and slider type. When the verification type is image type, the verification code recognition unit identifies the verification content based on image processing, and adopts an image processing algorithm to identify the character, graphic or jigsaw content in the verification code; when the verification type is slider type, the verification code recognition unit extracts the target region and the starting region by using a target detection model, determines the initial slider track route according to the target region and the starting region, compares the initial slider track route with the historical route data set, determines whether to adjust the initial slider track route according to the comparison result, and determines the final slider track route after adjustment. Specifically, the verification code recognition unit accurately identifies the target region and the starting region involved in the slider type verification code by the target detection model, then constructs the initial sliding track based on the Bezier curve, compares and analyzes the initial track with the typical track data in the historical track database, optimizes the track path according to the comparison result if the difference is large or does not have "naturalness", and thus outputs the final slider track conforming to the real user operation characteristics.

[0063] The SMS mailbox processing unit is configured to connect multiple virtual mobile service platforms based on API, receive and extract the SMS verification code; the SMS mailbox processing unit is also configured to access the mailbox account used for registration, and extract the content of the mail body or title by using a semantic target extraction model; specifically, the SMS mailbox processing unit has the automatic receiving and extracting capability of the verification code information, specifically connects multiple virtual mobile service platforms by API to realize real-time acquisition of the SMS verification code, and simultaneously, the unit accesses the mailbox account used for the registration process, intelligently extracts the content of the mail title or body by using a semantic target extraction model, and accurately extracts the verification code data therefrom. Compared with the traditional method based on fixed rules, the SMS mailbox processing unit has stronger universality and content adaptability, and can cope with the variability of the verification code format.

[0064] The account management unit is configured to store the registration data of each account; specifically, the account management unit is used to uniformly store and manage the registration information of various social platform accounts, including registration time, use platform, verification code processing record, etc., to provide a data basis for subsequent maintenance, use and management of the accounts.

[0065] Compared with the prior art, the embodiment of the application improves the adaptability and operation concealment of social robots in a complex verification environment by constructing a verification confrontation system for data acquisition, multi-type verification code recognition processing, verification code content extraction, and account unified management. The acquisition unit obtains the conditions and verification types of the verification codes triggered by different social platforms during the registration, login, and verification processes, providing data support for subsequent identification and simulation. In the slider type verification processing, the verification code recognition unit introduces a target detection model to accurately extract the starting and target areas of the slider, and generates a high-simulation trajectory path by combining the Bezier curve and historical trajectory data, thereby enhancing the naturalness and robustness of the slider simulation operation and reducing the risk of being identified as abnormal. The SMS email processing unit automatically receives the verification codes by connecting multiple virtual mobile service platforms through the API, and flexibly extracts the email content using a semantic target extraction model, effectively adapting to the diversity of verification code formats. The account management unit stores and maintains the registered accounts, facilitating subsequent tracking and batch management.

[0066] Second embodiment

[0067] As shown in Figure 2 The second embodiment of the application provides a verification confrontation method for a social robot, which is implemented based on the verification confrontation system for a social robot provided in the first embodiment, and at least includes the following steps:

[0068] S100: Acquire interaction data of multiple social platforms during the registration, login, and verification processes, wherein the interaction data includes verification code trigger conditions and verification types.

[0069] S200: Identify the verification types, wherein the verification types include slider types.

[0070] S300: When the verification type is a slider type, use a target detection model to extract the target area and the starting area, generate an initial slider trajectory route based on the coordinates of the target area and the starting area using a Bezier curve, compare the initial slider trajectory route with historical route data sets, determine whether to adjust the initial slider trajectory route based on the comparison result, and determine the slider trajectory route as the final route.

[0071] S400: Obtain information in the verification code.

[0072] S500: Store the registration data of each account.

[0073] Further, in step S300, when the verification type is a slider type, the following sub-steps are included:

[0074] S311: The verification code recognition unit uses a target detection model to extract the target area and the starting area.

[0075] Specifically, the verification code recognition unit obtains historical slider verification data according to the interaction data, extracts a historical slider background image and a historical slider block image, and establishes a historical data set.

[0076] The historical data set is labeled, and the labeling targets include a gap region and a slider block.

[0077] The historical data set is sampled according to a preset proportion to obtain a training subset and a test subset. The verification code recognition unit obtains an initial target detection model, and iteratively trains the initial target detection model according to the training subset. The target detection model after the iterative training is evaluated according to the test subset.

[0078] If the confidence of the target detection model after the current iterative training is less than the confidence of the target detection model after the previous iterative training, the target detection model after the previous iterative training is used as the target detection model.

[0079] If the confidence of the target detection model after the current iterative training is greater than or equal to the confidence of the target detection model after the previous iterative training, the amplitude of the change of the target detection model in the gradient direction is increased, and the iterative training is continued until a preset number of iterations is reached or the confidence of the target detection model after the iterative training is less than the confidence of the target detection model after the previous iterative training. Finally, the target detection model is obtained.

[0080] The current slider image data is input into the target detection model, and a first detection box and a second detection box are extracted from the output result. The first detection box is used to detect a gap region (slot). The target detection model identifies the position of the gap region in the verification code image and labels it with a bounding box (detection box). The coordinate information (such as the coordinates of the upper left corner and the lower right corner) of this bounding box is used to determine the position of the gap in the image. The second detection box is used to detect a slider (slider). The target detection model identifies the position of the draggable slider and labels it with another bounding box. The coordinate information of this bounding box is used to determine the initial position of the slider in the image.

[0081] The verification code recognition unit determines a target region according to the first detection box and determines a starting region according to the second detection box.

[0082] S312: The verification code recognition unit generates an initial slider trajectory route using a Bezier curve according to the coordinates of the target region and the starting region.

[0083] Specifically, the verification code recognition unit takes the center of the starting region as a starting point and takes the vertically corresponding point of the center of the target region as a target point.

[0084] The verification code recognition unit randomly generates two offset points between the starting point and the target point, and the offset points are vertically offset by ±5-15 pixels in the vertical coordinate.

[0085] The verification code recognition unit determines the trajectory points with the Bezier curve, and obtains the initial slider trajectory route according to all the trajectory points.

[0086] Specifically, a historical data set containing a large number of samples is established by extracting a historical slider background image and a historical slider block image from historical interaction data; the historical data set is target-labeled to clearly indicate the positions of the "gap region" and the "slider block" in the image, which are marked as slot and slider, respectively; the data set is divided into a training subset and a test subset according to a preset ratio; an initial target detection model is obtained, and the training set is used for multiple rounds of iterative training, and the test set is used to evaluate the performance of the training result; if the performance of the target detection model decreases (such as a decrease in confidence) after training, the last target detection model with better performance is used; if the performance improves, the adjustment range of the target detection model in the gradient direction is further increased to speed up the convergence speed until the number of training times or the performance stops improving; and finally, the target detection model with the best performance is formed.

[0087] The Bezier curve generates a trajectory with varying acceleration (initially slow, fast in the middle, and slow to stop at the end) through control points, perfectly reproduces human operation characteristics, realizes physical-level simulation, and has strong anti-detection ability.

[0088] In the running phase, the slider image to be recognized is input into the trained target detection model. Two detection boxes are marked in the model output result: the first detection box corresponds to the gap region of the verification code, which is the target region of the slider; and the second detection box corresponds to the initial position of the slider, which is the starting region; the center coordinates of these two regions are used as key reference points for trajectory generation. The center coordinate of the starting region is set as the starting point of the trajectory, a horizontal line is drawn from the starting point, a vertical line is drawn from the center point of the target region, and the intersection of the horizontal line and the vertical line is set as the end point; two offset points are randomly generated between the starting point and the end point, and the offset points are offset by a predetermined number of pixels (for example, ±5-15 pixels) in the vertical direction, which introduces a control point random offset mechanism to make each trajectory unique; a smooth trajectory line with real operation characteristics is fitted using the Bezier curve based on the four key points (the starting point, the two offset points, and the end point); and all the trajectory points are combined to form the final initial slider trajectory route, which is used for subsequent comparison and adjustment.

[0089] S313: The verification code recognition unit compares the initial slider trajectory route with the historical route data set, determines whether to adjust the initial slider trajectory route based on the comparison result, and realizes continuous self-optimization.

[0090] It should be noted that the historical route data set includes a historical successful verification trajectory route set and a historical failed verification trajectory route set.

[0091] The verification code recognition unit obtains a first average coincidence degree of the initial sliding block trajectory route and the set of historical successful verification trajectory routes, and obtains a second average coincidence degree of the initial sliding block trajectory route and the set of historical failed verification trajectory routes.

[0092] When the first average coincidence degree is greater than or equal to the second average coincidence degree, the verification code recognition unit determines not to adjust the initial sliding block trajectory route, and takes the initial sliding block trajectory route as the final sliding block trajectory route. When the first average coincidence degree is less than the second average coincidence degree, the verification code recognition unit determines that the initial sliding block trajectory route needs to be adjusted, and determines the final sliding block trajectory route after adjustment.

[0093] S314: If the verification code recognition unit determines that the initial sliding block trajectory route needs to be adjusted, specifically includes:

[0094] The verification code recognition unit determines an adjustment region of the initial sliding block trajectory route, and the adjustment region is a region in the initial sliding block trajectory route that has a coincidence degree higher than a coincidence degree threshold with the set of historical failed verification trajectory routes. The verification code recognition unit determines an adjustment starting point and an adjustment ending point of each adjustment region, traverses the set of historical successful verification trajectory routes according to the adjustment starting point and the adjustment ending point, and intercepts a route in the set of historical successful verification trajectory routes to replace the adjustment region.

[0095] Specifically, the set of historical successful verification trajectory routes (i.e., sliding trajectories that have previously successfully bypassed verification). The set of historical failed verification trajectory routes (i.e., sliding trajectories that have failed to successfully bypass verification). The average coincidence degrees of two dimensions are calculated, which are referred to as the first average coincidence degree and the second average coincidence degree. The coincidence degree can be measured based on the degree of spatial overlap of the trajectory point positions, such as the average Euclidean distance of the trajectory points, the proportion of similar path segments, etc. If the first average coincidence degree (i.e., the similarity to the successful trajectories) is higher than or equal to the second average coincidence degree (i.e., the similarity to the failed trajectories), it indicates that the current trajectory is closer to the successful samples, and the initial sliding block trajectory is directly adopted as the final route; if the first average coincidence degree is lower than the second average coincidence degree, it indicates that the trajectory is more similar to the failed samples, and there is a risk of being identified and intercepted by the platform, triggering the trajectory optimization mechanism to adjust the current trajectory. The regions in the initial sliding block trajectory that have a coincidence degree higher than a preset threshold with the historical failed verification trajectory are identified as adjustment regions, because these regions may have obvious abnormal features or repetitive features; for each adjustment region, the starting point and the ending point position thereof are determined, i.e., the range of the trajectory segment that needs to be replaced; then, a trajectory segment between similar starting and ending points is searched out from the set of historical successful verification trajectories, for replacing the current adjustment region, and if there is no exactly the same starting and ending point position, a trajectory segment corresponding to the highest similarity value is adopted based on the similarity. The adjusted trajectory is closer to the real user behavior, thereby avoiding platform risk identification.

[0096] S315: The verification code recognition unit determines the final slider track route.

[0097] S316: The verification code recognition unit determines the sliding speed between each adjacent track point based on the nonlinear speed model. The sliding speed is calculated by the following formula:

[0098]

[0099] wherein Vt represents the sliding speed of t points between each adjacent track point, t0 represents the midpoint of adjacent track points, Vmax represents the maximum sliding speed, and k represents the speed variation rate.

[0100] S317: The verification code recognition unit determines the maximum sliding speed set at the corresponding position in the historical successful verification track route set according to the position of the adjacent track point in the final slider track route.

[0101] wherein the maximum sliding speed is calculated by the following formula:

[0102]

[0103] wherein N represents the number of tracks in the historical successful verification track route set, and Vi represents the maximum sliding speed in the i-th historical successful verification track route.

[0104] S318: Perform the slider type verification code operation according to the final slider track route, the sliding speed between each adjacent track point, and the maximum sliding speed obtained in steps S315 to S317. If the sliding speed generated in step S316 exceeds the maximum sliding speed determined in step S317, the operation is performed at the maximum sliding speed.

[0105] That is, a complete behavior simulation needs to meet both path authenticity and speed characteristic authenticity. This embodiment determines the final track through step S315, generates each speed through step S316, obtains the maximum sliding speed through step S317, and performs the specific operation through step S318. This process ensures that the track meets the requirements in terms of path simulation (step S315) and speed simulation (steps S316-S317) to achieve the core goal of "humanized operation".

[0106] If only the path is authentic (Bezier curve) and the speed is not authentic (e.g., uniform sliding), it is easy to be identified by the speed detection mechanism. The nonlinear speed model (step S316) simulates the acceleration characteristics of "slow-fast-slow" of humans, and then step S317 constrains the upper limit of the speed through historical data to avoid outliers.

[0107] In an embodiment of the present application, step S400 is completed by two ways to obtain the verification code information: first, use the preset API interface to connect multiple virtual mobile phone service platforms at the same time, real-time monitor and receive the short message sent to the specified virtual number, then parse the short message content by field, extract the numerical or alphabetical verification code therein. Alternatively, log in the email account used in the registration process, use the semantic target extraction model based on Transformer to scan and match the email body and title character by character, automatically identify and extract the verification code field, thereby completely covering the two scenarios of short message verification code and email verification code.

[0108] Compared with the prior art, in the simulation of the slider type verification code, the embodiment of the present application combines a target detection model (preferably YOLOv8, but not limited thereto) with a Bezier curve and introduces a trajectory comparison optimization mechanism, effectively enhancing the naturalness of the trajectory and reducing the risk of being identified; a semantic target extraction model (preferably Transformer, but not limited thereto) is used for email content extraction, improving the intelligentization and generalization ability of the verification code recognition. The verification countermeasures system has enhanced the ability to counter the variable verification mechanism, and also provides protection for the stable implementation of legal automated operations in scenarios such as public opinion analysis, information extraction, platform testing, etc.

[0109] The embodiment of the present application realizes high-precision automatic identification of the key area in the slider type verification code by constructing and training a target detection model, which has stronger robustness and adaptability than traditional image template matching methods; at the same time, a trajectory generation mechanism based on Bezier curve is introduced, which simulates a sliding trajectory closer to the natural operation habits of human beings by reasonably setting the offset points and path curves, improving the authenticity and concealment of the simulation operation. On the basis of ensuring technical precision, the operation simulation degree is improved, and the breakthrough ability of the slider verification mechanism is enhanced.

[0110] In addition, the embodiment of the present application establishes a closed-loop optimization structure between the slider trajectory generation process and historical data feedback through coincidence degree calculation and trajectory segment replacement. Compared with the traditional method of static use after single trajectory generation, it can dynamically learn and integrate historical verification experience, balance between trajectory authenticity and success rate, and is suitable for platform environments with constantly upgraded protection mechanisms. The camouflage ability and verification pass rate of the slider trajectory are improved.

[0111] The embodiment of the present application realizes highly personalized simulation of the slider behavior in the time dimension by fusing nonlinear speed modeling and historical successful trajectory data analysis, avoiding the exposure of "mechanical features" such as uniform speed and smooth trajectory to the platform verification system. Especially in various social platforms with trajectory detection and behavior analysis mechanisms, the concealment and pass rate of the sliding trajectory are enhanced, further improving the stability and intelligence of the entire verification countermeasures system.

[0112] Third Embodiment

[0113] like Figure 3 As shown, the third embodiment of the present invention provides a verification adversarial method for social robots, which is implemented based on the verification adversarial system for social robots provided in the first embodiment, and includes at least the following steps:

[0114] S100: Collects interaction data from multiple social platforms during the registration, login, and verification processes. The interaction data includes verification code triggering conditions and verification types.

[0115] S200': Identify the verification type, which includes image types.

[0116] S300': When the verification type is image type, the verification content is identified based on image processing.

[0117] If you click to select a text-based verification method (interactive), then the following sub-steps will be taken:

[0118] a. Recognize all clickable text / icons using OCR;

[0119] b. Parse the verification command (e.g., "Click the Xth character");

[0120] c. Generate a random click sequence (coordinates + time sequence) that conforms to human operation characteristics, where coordinates represent the location information of the click and time sequence represents the time order of the click, in order to simulate human operation and avoid detection by anti-robot systems (random delays or paths can be added);

[0121] d. Simulate the click operation by following this click sequence to complete the verification.

[0122] S400: Retrieve information from the verification code.

[0123] S500: Stores registration data for each account.

[0124] Furthermore, in step S300', when the verification type is an image, the following sub-steps are included:

[0125] S301: Convert the acquired color verification code image into a grayscale image to reduce the computational complexity of image processing.

[0126] S302: Taking any point in the grayscale image as the origin, perform Gaussian weighted average processing on the surrounding pixel area to dynamically calculate the pixel threshold of the area, so as to adapt to the image characteristics of different lighting, contrast and other interference factors, and obtain an optimized grayscale image.

[0127] S303: Based on the edge detection algorithm (such as Canny edge detection), the processed image is analyzed to identify the obvious structural edges in the image, thereby determining the possible outline area of the verification code.

[0128] S304: When the identified outline area is a text verification code, further image correction processing is performed. Specifically, it includes the following sub-steps: calculate the overall tilt angle of the verification code area by identifying the minimum circumscribed rectangle of each character block, and construct a rotation matrix for rotation correction based on this, to ensure that the character arrangement is in the normal horizontal direction; extract the center coordinates of each character area, and connect all the center points; if the connection line shows a nonlinear bending trend, it means that the character arrangement is distorted. Model the connection line based on polynomial fitting, and align the character positions according to the fitting results, so that the characters are as neat as possible in structure.

[0129] S305: If the outline area of the verification code is a graphical type (such as graphical selection, pattern matching, etc.), compare the extracted area features with the standard images in the graphical database to determine the graphical type or matching content of the current verification code.

[0130] In an embodiment of the present application, step S400 completes the acquisition of verification code information in two ways: first, use a preset API interface to connect multiple virtual mobile service platforms at the same time, real-time monitor and receive the short message sent to the specified virtual number, then parse the short message content by field, extract the numerical or alphabetical verification code therein. Alternatively, log in to the email account used during the registration process, and use a semantic target extraction model based on Transformer to scan and match the email body and title character by character, automatically identify and extract the verification code field, thereby covering both short message verification code and email verification code scenarios.

[0131] Compared with the prior art, in the image type verification code processing, the embodiment of the application directly calls image processing and geometric modeling means such as adaptive grayscale processing, Gaussian weighting, edge detection, rotation correction and polynomial fitting: first, the color verification code image is converted into a grayscale image, then the local pixels are averaged by Gaussian weighting to dynamically calculate the threshold, and then the character or graphic contour is located by edge detection; if the text verification code is detected, the overall tilt angle is calculated by the minimum enclosing rectangle and the rotation matrix is constructed to complete the rotation correction, and then the character center coordinates are extracted and the non-linear distortion is corrected by polynomial fitting to restore the character arrangement to be horizontal; if the verification code is a graphic type, the extracted region features are compared with the standard images in the graphic database to judge the graphic type or match the content. The process has higher robustness in image distortion correction and structure restoration than the traditional character recognition method, can effectively cope with complex interference such as tilt, distortion, nonlinear arrangement, and at the same time, the graphic matching mechanism expands the recognition ability of the verification confrontation system for non-text verification codes, improves the recognition accuracy, the application range and the general intelligent level.

[0132] It should be noted that the above embodiments are only illustrative. The technical solutions of each embodiment can be combined, and the order of each step can be changed, all of which are within the protection scope of the application.

[0133] The verification confrontation method and system for social robots provided by the application are described in detail above. Any obvious modification made by a person skilled in the art without departing from the essential content of the application will constitute an infringement of the patent right of the application and will bear the corresponding legal responsibility.

Claims

1. A verification adversarial method for a social robot, characterized in that The method comprises the following steps: S100: collecting interaction data of a plurality of social platforms in a registration, login and verification process, wherein the interaction data comprises a verification code triggering condition and a verification type; S200: identifying the verification type, wherein the verification type comprises a sliding block type; S300: when the verification type is the sliding block type, using a target detection model to extract a target region and a starting region, using a Bezier curve to generate an initial sliding block trajectory route according to coordinates of the target region and the starting region, comparing the initial sliding block trajectory route with historical route data sets, determining whether to adjust the initial sliding block trajectory route according to a comparison result, and determining a final sliding block trajectory route after adjustment; S400: obtaining information in the verification code; S500: storing registration data of each account. 2.The authentication adversarial method for social robots of claim 1, wherein The step of using the target detection model to extract the target region and the starting region specifically comprises: obtaining historical sliding block verification data according to the interaction data, extracting a historical sliding block background image and a historical sliding block block image, and establishing a historical data set; annotating the historical data set, wherein the annotation targets comprise a gap region and a sliding block block; sampling the historical data set according to a preset proportion to obtain a training subset and a test subset, obtaining an initial target detection model, and iteratively training the initial target detection model according to the training subset, and evaluating the target detection model after the iterative training according to the test subset; if a confidence degree of the target detection model after the current iterative training is less than a confidence degree of the target detection model after the previous iterative training, using the target detection model after the previous iterative training as the target detection model; if the confidence degree of the target detection model after the current iterative training is greater than or equal to the confidence degree of the target detection model after the previous iterative training, increasing a variation amplitude of the target detection model in a gradient direction, continuing the iterative training, and finally obtaining the target detection model until a preset iteration number is reached or the confidence degree of the target detection model after the iterative training is less than the confidence degree of the target detection model after the previous iterative training; inputting current sliding block image data into the target detection model, and extracting a first detection box and a second detection box from an output result; determining the target region according to the first detection box, and determining the starting region according to the second detection box.

3. The verification confrontation method for a social robot according to claim 2, wherein: in a running stage, inputting a sliding block image to be identified into the trained target detection model; marking two detection boxes in a model output result: a first detection box corresponds to a gap region of a verification code, and is used as a target region of the sliding block; a second detection box corresponds to an initial position of the sliding block, and is used as a starting region; using center coordinates of the two regions as key reference points for trajectory generation; setting the center coordinate of the starting region as a starting point of the trajectory, drawing a horizontal line from the starting point, drawing a vertical line from a center point of the target region, and setting an intersection point of the horizontal line and the vertical line as an ending point; randomly generating two offset points between the starting point and the ending point, and offsetting the two offset points by a predetermined pixel in a vertical coordinate direction to simulate slight shaking that may occur in human hand operation; combining the starting point, the two offset points and the ending point, and using a Bezier curve to fit a smooth trajectory line with real operation characteristics. Synthesizing a final initial slider track route based on all the track points, for subsequent comparison and adjustment. 4.The authentication adversarial method for social robots of claim 3, wherein Comparing the initial slider track route with the historical route dataset, and determining whether to adjust the initial slider track route according to the comparison result, specifically including: Obtaining a first average coincidence degree of the initial slider track route and the historical successful verification track route set; obtaining a second average coincidence degree of the initial slider track route and the historical failed verification track route set; When the first average coincidence degree is greater than or equal to the second average coincidence degree, it is determined that the initial slider track route does not need to be adjusted, and the initial slider track route is taken as the final slider track route; when the first average coincidence degree is less than the second average coincidence degree, it is determined that the initial slider track route needs to be adjusted, and the final slider track route is determined after adjustment. 5.The authentication adversarial method for social robots of claim 4, wherein After determining the final slider track route, it further includes: Determining the sliding speed between each adjacent track point based on a nonlinear speed model; Determining a maximum sliding speed set at a corresponding position in the historical successful verification track route set according to the position of the adjacent track points in the final slider track route; Performing a slider verification operation according to the final slider track route, the sliding speed between each adjacent track point, and the maximum sliding speed; if the sliding speed exceeds the maximum sliding speed, the operation is performed according to the maximum sliding speed.

6. A verification adversarial method for a social robot, characterized in that The steps include: S100: Collecting interaction data of multiple social platforms in the registration, login and verification process, the interaction data including verification code trigger conditions and verification types; S200': Identifying the verification type, the verification type including image type; S300': When the verification type is image type, identifying the verification content based on image processing; which includes the following sub-steps: S301: converting the obtained color verification code image into a grayscale image; S302: taking any point in the grayscale image as the origin, performing Gaussian weighted average processing on the surrounding pixel region, thereby dynamically calculating the pixel threshold of the region to obtain an optimized grayscale image; S303: analyzing the optimized grayscale image based on an edge detection algorithm to identify the obvious structural edges in the image, thereby determining the outline area where the verification code may exist; S304: when the identified outline area is a text verification code, further performing image correction processing; S305: if the verification code outline area is a graphic type, comparing the extracted region features with the standard images in the graphic database to determine the graphic type or matching content of the current verification code; S400: Obtaining information in the verification code; S500: Storing the registration data of each account. 7.The authentication adversarial method for social robots of claim 6, wherein The step S304 specifically includes: Calculate the overall inclination angle of the verification code area by identifying the minimum circumscribed rectangle of each character block, and construct a rotation matrix for rotation correction to ensure that the character arrangement is in the normal horizontal direction; Extract the center coordinates of each character region and connect all the center points; If the connection line shows a nonlinear bending trend, it indicates that the character arrangement is distorted; Model the connection line based on polynomial fitting, and align and translate the character positions according to the fitting result. 8.The authentication adversarial method for social robots according to claim 1 or 6, wherein In the step S400, multiple virtual mobile phone service platforms are accessed based on API to receive and extract short messages to obtain information in the verification code. 9.The authentication adversarial method for social robots according to claim 1 or 6, wherein In the step S400, the mailbox account used for registration is accessed, and a semantic target extraction model is used to extract the content of the mail body or title to obtain information in the verification code.

10. A verification adversarial system for a social robot, for implementing the verification adversarial method for a social robot according to any one of claims 1-7, characterized in that Comprise: The acquisition unit is configured to acquire interactive data of multiple social platforms in the registration, login and verification process, and the interactive data includes verification code trigger condition and verification type; The verification code recognition unit is configured to recognize the verification type, and the verification type includes image type and sliding block type; The short message mailbox processing unit is configured to access multiple virtual mobile phone service platforms based on API to receive and extract short message verification codes, or access the mailbox account used for registration, and use a semantic target extraction model to extract the content of the mail body or title to obtain information in the verification code; The account management unit is configured to store the registration data of each account.

Images