Data processing method and related device

Abstract

The embodiment of the invention discloses a data processing method and a related device in the field of artificial intelligence, and the method comprises the steps: obtaining a to-be-detected image which is a game screenshot of a target object; whether a suspicious element exists in the to-be-detected image or not is detected through the target detection model, the target position of the suspicious element in the to-be-detected image and the element type of the suspicious element are determined, the suspicious element is an element possibly corresponding to a cheating trace, and the cheating trace is visual content generated by a game plug-in and used for assisting cheating; determining an element cheating detection result corresponding to the suspicious element according to a local image corresponding to a target position in the to-be-detected image through an image processing model corresponding to the element type under the condition that the suspicious element exists in the to-be-detected image; and determining whether the target object uses the game plug-in or not according to the element cheating detection result. According to the method, the detection accuracy of cheating behaviors in the game application can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a data processing method and related devices. Background Art

[0002] With the rapid development of computer and internet technologies, gaming applications are now widely popular, but this has also led to the problem of game cheats. Game cheats are third-party software programs that can modify normal in-game data and logic. The use of game cheats can disrupt the balance of the gaming environment and seriously affect the gaming experience of normal players.

[0003] To address the issue of game cheating, gaming applications' backend devices employ cheat detection methods to detect whether players are using cheats and punish those who engage in cheating. Currently, a common cheating detection method uses visual models to detect cheating patterns in player screenshots of the game interface. However, screenshots typically contain a wealth of information, making it difficult to accurately capture information related to cheating using visual models. Consequently, these methods have limited cheating detection accuracy. Summary of the Invention

[0004] The embodiments of the present application provide a data processing method and related devices that can improve the accuracy of detecting cheating behavior in game applications.

[0005] In a first aspect, the present application provides a data processing method, the method comprising:

[0006] Acquire an image to be detected; the image to be detected is a screenshot of a game of a target object;

[0007] Using a target detection model, detecting whether a suspicious element exists in the image to be detected, and determining the target position of the suspicious element in the image to be detected and the element type of the suspicious element; the suspicious element is an element that may correspond to cheating traces, which are visual content generated by game plug-ins to assist cheating;

[0008] When the suspicious element exists in the image to be detected, determining an element cheating detection result corresponding to the suspicious element based on a local image corresponding to the target position in the image to be detected by using an image processing model corresponding to the element type;

[0009] According to the element cheating detection result, it is determined whether the target object uses the game plug-in.

[0010] A second aspect of the present application provides a data processing device, the device comprising:

[0011] An acquisition module for acquiring an image to be detected; the image to be detected is a game screenshot of a target object;

[0012] A first detection module for detecting, through a target detection model, whether there is a suspicious element in the image to be detected, and determining the target position of the suspicious element in the image to be detected and the element type of the suspicious element; the suspicious element is an element that may correspond to a cheating trace, and the cheating trace is a visual content generated by a game cheat for assisting cheating;

[0013] A second detection module for, in the case where the suspicious element exists in the image to be detected, determining, through an image processing model corresponding to the element type, an element cheating detection result corresponding to the suspicious element according to a local image corresponding to the target position in the image to be detected;

[0014] An anti-cheat detection module for determining whether the target object uses the game cheat according to the element cheating detection result.

[0015] The third aspect of the present application provides a computer device, the device includes a processor and a memory:

[0016] The memory is used for storing a computer program;

[0017] The processor is used for executing the steps of the data processing method as described in the first aspect above according to the computer program.

[0018] The fourth aspect of the present application provides a computer-readable storage medium, the computer-readable storage medium is used for storing a computer program, and the computer program is used for executing the steps of the data processing method as described in the first aspect above.

[0019] The fifth aspect of the present application provides a computer program product or a computer program, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps of the data processing method as described in the first aspect above.

[0020] From the above technical solutions, it can be seen that the embodiments of the present application have the following advantages:

[0021] An embodiment of the present application provides a data processing method. This method innovatively proposes a two-stage detection mechanism for game screenshots. According to the detection idea of first overall and then local, it first detects whether there are visible cheating elements generated by game cheats in the game screenshot, and then detects cheating behaviors in the game application. In this method, first, an image to be detected is obtained, and the image to be detected is a game screenshot of a target object. Then, through a target detection model, it is determined whether there are suspicious elements in the image to be detected that may correspond to cheating traces. And when it is determined that there are suspicious elements, the target position of the suspicious elements in the image to be detected is located, and the element type to which the suspicious elements belong is determined, so as to achieve the first-stage overall detection of the image to be detected. Furthermore, when it is determined through the first-stage overall detection that there are suspicious elements in the image to be detected, a second-stage local detection is further started for this image to be detected, that is, using an image processing model corresponding to the element type of the suspicious elements to process the local image corresponding to the target position in the image to be detected to determine whether the suspicious elements are real cheating traces; on the one hand, the image processing model adopted in this stage corresponds to the element type of the suspicious elements, and it is a model dedicated to processing images related to this element type. Therefore, it can ensure that the determined element cheating detection result has high accuracy; on the other hand, the image that the image processing model needs to process is the local image corresponding to the target position where the suspicious elements are located. This local image mainly reflects the content related to the suspicious elements and has less interference information. Therefore, it can reduce the interference of irrelevant information on the image processing model and is conducive to improving the accuracy of the determined element cheating detection result. Finally, it can be determined whether the target object uses a game cheat according to the element cheating detection result; since the determined element cheating detection result has high accuracy, it can ensure that the detection result of the game cheat determined accordingly has high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Schematic diagram of the application scenario of the data processing method provided by the embodiment of the present application;

[0023] Figure 2 Schematic diagram of the flow of the data processing method provided by the embodiment of the present application;

[0024] Figure 3a Schematic diagram of the detection result output by the target detection model provided by the embodiment of the present application;

[0025] Figure 3b Schematic diagram of another detection result output by the target detection model provided by the embodiment of the present application;

[0026] Figure 4a Schematic diagram of the local image of the data processing method provided by the embodiment of the present application;

[0027] Figure 4b Schematic diagram of another partial image for the data processing method provided by the embodiment of the present application;

[0028] Figure 5 Structural diagram of the image classification model provided by the embodiment of the present application;

[0029] Figure 6 Structural schematic diagram of the data processing method provided by the embodiment of the present application;

[0030] Figure 7 Schematic diagram of another data processing method provided by the embodiment of the present application;

[0031] Figure 8a Schematic diagram of constructing the first training black sample provided by the embodiment of the present application;

[0032] Figure 8b Schematic diagram of another method for constructing the first training black sample provided by the embodiment of the present application;

[0033] Figure 9 Schematic diagram of iteratively optimizing the object detection model provided by the embodiment of the present application;

[0034] Figure 10 Schematic diagram of constructing the second training white sample provided by the embodiment of the present application;

[0035] Figure 11 Structural schematic diagram of the data processing device provided by the embodiment of the present application;

[0036] Figure 12 Structural schematic diagram of the terminal device provided by the embodiment of the present application;

[0037] Figure 13 Structural schematic diagram of the server provided by the embodiment of the present application. Detailed implementation manners

[0038] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present application.

[0039] The terms "first", "second", "third", "fourth", etc. (if any) in the description, claims, and the above-mentioned drawings of this application are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that comprises a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0040] Artificial Intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines to enable the machines to have the functions of perception, reasoning, and decision-making.

[0041] Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, pre-trained model technology, operation / interaction systems, mechatronics, etc. Among them, the pre-trained model, also known as the large model or the foundation model, can be widely applied to downstream tasks in various major directions of artificial intelligence after fine-tuning. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0042] Computer Vision (CV) technology is a science that studies how to enable machines to "see". More specifically, it refers to using cameras and computers to replace human eyes for tasks such as object recognition and measurement in machine vision, and further performing graphics processing to make the computer-processed images more suitable for human eyes to observe or be transmitted to instruments for detection. As a scientific discipline, computer vision researches related theories and technologies, and attempts to build artificial intelligence systems that can obtain information from images or multi-dimensional data. Large model technology has brought important changes to the development of computer vision technology. Pretrained models in the visual field such as swin-transformer, ViT, V-MOE, and MAE can be quickly and widely applied to downstream specific tasks after fine-tuning. Computer vision technology usually includes technologies such as image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content / behavior recognition, 3D object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, etc., and also includes common biometric recognition technologies such as face recognition and fingerprint recognition.

[0043] The solution provided in the embodiments of this application relates to the computer vision technology of artificial intelligence, and is specifically described through the following embodiments:

[0044] The data processing method provided in the embodiments of this application can be executed by a computer device, which can be a terminal device or a server. Among them, terminal devices include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, aircraft, etc. A server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server.

[0045] It should be noted that the information (including but not limited to relevant information of accounts, etc.), data (including but not limited to game screenshot data, etc.) and signals involved in the embodiments of this application are all authorized by relevant parties or fully authorized by all parties, and the collection, use and processing of relevant data all comply with the relevant laws, regulations and standards of relevant countries and regions.

[0046] To facilitate the understanding of the data processing method provided in the embodiments of this application, the application scenario of this data processing method will be exemplarily introduced below taking the execution subject of this data processing method as a server.

[0047] See Figure 1 , Figure 1 which is a schematic diagram of the application scenario of the data processing method provided in the embodiments of this application. As Figure 1As shown, the application scenario includes a terminal device 110 and a server 120, and the terminal device 110 and the server 120 can communicate directly or indirectly through a wired network or a wireless network. Among them, a game application runs on the terminal device 110, and the current object of the game application is the target object. Through the terminal device 110, a screenshot of the game interface of the target object can be taken to obtain a to-be-detected image. The server 120 is used to execute the data processing method provided in the embodiments of the present application, and perform cheating detection based on the to-be-detected image provided by the terminal device 110 to detect whether the target object uses a game cheat.

[0048] In practical applications, after obtaining the to-be-detected image, the terminal device 110 can transmit the to-be-detected image to the server 120 through the network. The server 120 can detect whether there are suspicious elements in the to-be-detected image through a target detection model, and when it is determined that there are suspicious elements in the to-be-detected image, it can locate the target position of the suspicious element in the to-be-detected image and determine the element type of the suspicious element, so as to achieve the overall detection of the first stage of the to-be-detected image.

[0049] When the server 120 determines that there are suspicious elements in the to-be-detected image through the above-mentioned overall detection in the first stage, it enters the second stage. That is, the server 120 can process the local image corresponding to the target position in the to-be-detected image through an image processing model corresponding to the element type of the suspicious element, so as to determine whether the suspicious element is a real cheating trace and obtain an element cheating detection result corresponding to the suspicious element. Since the image processing model corresponds to the element type of the suspicious element and is a neural network model dedicated to processing local images related to this element type, it can ensure that the determined element cheating detection result has high accuracy. At the same time, the processing object of the image processing model is the local image corresponding to the target position where the suspicious element is located, and this local image mainly reflects the content related to the suspicious element and has less interference information. Therefore, it can reduce the interference of irrelevant information on the image processing model and is beneficial to improving the accuracy of the determined element cheating detection result.

[0050] Finally, the server 120 can determine whether the target object uses a game cheat according to the element cheating detection result corresponding to the suspicious element. Since the element cheating detection result determined through the above steps has high accuracy, determining whether the target object uses a game cheat according to this element cheating detection result can obtain a relatively accurate detection result of the game cheat.

[0051] It should be understood that Figure 1The application scenarios shown are only examples. In actual applications, the data processing method provided by the embodiments of this application can also be applied to other scenarios, and no limitations are imposed on the application scenarios of the data processing method provided by the embodiments of this application.

[0052] The data processing method provided by this application will be introduced in detail below through method embodiments.

[0053] See Figure 2 , Figure 2 which is a schematic flowchart of the data processing method provided by the embodiments of this application. For ease of description, the execution subject of this data processing method is taken as an example of a server in the following. As Figure 2 shown, the data processing method includes the following steps:

[0054] Step 201: Obtain an image to be detected, where the image to be detected is a game screenshot of a target object.

[0055] In the embodiments of this application, the target object is the object to be detected for whether it uses game cheats, that is, the player operating the game application. A game cheat is a third-party software program used during the game, which can help players cheat by modifying game files; as an example, a game cheat can be reflected in the game in the following forms. In a confrontation game, a game cheat can be manifested as automatic shooting or automatic operation, perspective (even if the opponent player is outside the current player's field of vision or behind an obstacle, the position of the opponent player can still be obtained), and speed increase, etc.

[0056] The image to be detected is a screenshot of the entire game interface of the target object; in the embodiments of this application, the image to be detected is used to assist in detecting whether the target object uses game cheats. In some cases, after a player uses game cheats, cheating traces will correspondingly be displayed in the game interface, such as displaying a cheat function directory (indicating the cheat functions available to the player), displaying markers indicating the positions of opponent players or items, etc.; based on this, the embodiments of this application can detect whether the target object uses game cheats by detecting whether the above cheating traces exist in the game screenshot. It should be understood that the above game screenshot refers to a screenshot of the game interface during the game, and this game screenshot can be, for example, a screenshot of a first-person shooting game (FPS), a screenshot of a multiplayer online battle arena (MOBA) game, etc. The embodiments of this application do not impose any limitations on the game to which the game screenshot belongs.

[0057] As an example, the server can obtain the image to be detected in the following ways: during the game, the terminal device running the game application can periodically capture screenshots of the entire game interface and send the game screenshots to the server for cheating detection. Alternatively, the terminal device running the game application can periodically capture screenshots of the entire game interface and store the game screenshots in a specific database, and then the server can obtain the game screenshots from the database. This application does not specifically limit the implementation method of obtaining the image to be detected.

[0058] Step 202: Use the target detection model to detect whether there are suspicious elements in the image to be detected, and determine the target position of the suspicious elements in the image to be detected and the element type of the suspicious elements. The suspicious elements are elements that may correspond to cheating traces, and the cheating traces are visual contents generated by game cheats for assisting in cheating.

[0059] The suspicious elements refer to the elements in the game screenshot that may correspond to cheating traces; specifically, the suspicious elements can be real cheating traces in the game screenshot, or the suspicious elements can also not be real cheating traces in the game screenshot, but elements misdetected as cheating traces. The cheating traces refer to the visual contents generated by game cheats for assisting in cheating. For example, they can be function directories for indicating cheating functions, boxes or skeletons for marking the positions of opponent players (opponent players not visible in the normal field of view), blood bars for marking the remaining health of opponent players, etc.

[0060] The target detection model is used to detect whether there are suspicious elements in the game screenshot, and in the case of detecting that there are suspicious elements in the game screenshot, it can also determine the target position of the suspicious elements in the game screenshot and the element type of the suspicious elements. As an example, the types of suspicious elements can be divided into text types and marker types. The text type refers to text-based cheating traces, such as function directories in the game interface for indicating cheating functions, and the marker type refers to marker-based cheating traces for assisting in cheating, such as boxes, skeletons, and blood volumes in the game interface for marking opponent players.

[0061] In addition, the target detection model can also output the confidence level of the suspicious elements. The confidence level of the suspicious elements is used to indicate the probability that the suspicious elements are real cheating traces. In practical applications, for an image to be detected, the target detection model can detect one suspicious element or multiple suspicious elements; in the case of detecting multiple suspicious elements, the top k (k is an integer greater than or equal to 1, for example, equal to 3) suspicious elements with the highest confidence levels can be selected according to the confidence levels of each suspicious element as the basis for subsequent detection of cheating behaviors.

[0062] Exemplarily, see Figure 3a , Figure 3aSchematic diagram of the detection result output by the target detection model provided in the embodiment of the present application; the obtained game screenshot is input into the target detection model, and the target detection model can detect whether the game screenshot includes suspicious elements. When the detection result is that there are no suspicious elements in the game screenshot, there is no need to perform subsequent detection on the image to be detected. See Figure 3b , Figure 3b Schematic diagram of another detection result output by the target detection model provided in the embodiment of the present application. If the target detection model detects that there are suspicious elements in the game screenshot, the detection result will simultaneously include the target position of the suspicious elements in the game screenshot and the type of the suspicious elements (such as the bone type).

[0063] As an example, the target detection model can be a neural network model based on YOLOv7. YOLOv7 is a common deep learning model in the field of target detection, and its model structure is mainly composed of deep convolutional layers and residual modules. Of course, in practical applications, the target detection model can also be a model with other structures, and the embodiments of the present application do not make any limitations on this.

[0064] Step 203: In the case where there are suspicious elements in the image to be detected, through the image processing model corresponding to the element type, determine the element cheating detection result corresponding to the suspicious elements according to the local image corresponding to the target position in the image to be detected.

[0065] Generally, when people view an image, they will first focus on which area of the entire image the interesting object is in at the first glance, and then carefully view the specific area where the interesting object is located, so as to be able to distinguish the interesting object more accurately. Based on this, in the embodiment of the present application, when detecting the entire game screenshot based on the target detection model and determining that there are suspicious elements in the game screenshot, refined detection can be further performed according to the target position of the suspicious elements in the game screenshot, that is, the local image corresponding to the target position in the game screenshot can be further refined for detection, so as to more accurately determine the element cheating detection result corresponding to the suspicious elements, in line with the law of people observing small objects in the entire image.

[0066] In the case where there are suspicious elements in the image to be detected, the target detection model can output the target position of the suspicious elements in the image to be detected. Therefore, the corresponding local image can be intercepted based on the target position of the suspicious elements in the image to be detected. See Figure 4a , Figure 4aSchematic diagram of a partial image of the data processing method provided in the embodiments of the present application. As an example, the target position of a suspicious element output by a target detection model can be marked by a prediction box, and then the center coordinate point of the prediction box can be extended up, down, left, and right to obtain a local area of 300*300. Finally, the local area can be cropped by taking a screenshot to obtain the partial image corresponding to the target position. See also Figure 4b , Figure 4b Schematic diagram of another partial image of the data processing method provided in the embodiments of the present application. As an example, when the element type is a text type, the prediction box output by the target detection model can also be directly cropped by taking a screenshot, and the cropped image can be used as the partial image corresponding to the target position.

[0067] The image processing model corresponding to the element type is used to detect the partial image to which a suspicious element belonging to the element type belongs in the game screenshot, so as to obtain the element cheating detection result of the suspicious element. The element cheating detection result is used to indicate whether the suspicious element is a real cheating trace, and specifically can indicate the probability that the suspicious element is a real cheating trace.

[0068] As an example, the image processing model corresponding to the element type can be an image text recognition model, such as an Optical Character Recognition (OCR) model, which can detect suspicious elements involving text. The image processing model corresponding to the element type can also be an image classification model, which can detect suspicious elements involving markers. Among them, the image classification model can be any model structure such as a LeNet model, an AlexNet model, a VGGNet model, a ResNet model, etc. The embodiments of the present application do not make any limitations on the structure of the image classification model here.

[0069] In a possible implementation manner, when the element type of the suspicious element is a text type, the target text included in the partial image can be recognized through the image text recognition model corresponding to the text type, and then the element cheating detection result of the suspicious element can be determined according to the recognized target text.

[0070] It should be noted that the element type of the suspicious element being a text type means that the suspicious element is embodied in the partial image in the form of text; the target text is the text included in the partial image, which may be the text used to indicate the cheating function in the game screenshot, such as the function directory of the game cheating software, which is used to indicate different cheating functions, such as alt+2 for switching protective equipment, alt+3 for switching medicine, and alt+4 for switching accessories, etc.

[0071] As an example, the image text recognition model can be an OCR model, and based on the OCR model, the target text in the partial image can be recognized. Since the OCR model only needs to recognize the text included in the partial image, compared with recognizing the whole image, the performance overhead based on the OCR model is reduced; in addition, when recognizing the whole image, it may happen that some non-cheating texts involving cheating keywords (such as the conversation text involving cheating information in the chat window) are misrecognized as cheating texts. However, in the embodiments of the present application, the recognition is performed based on the partial image of the text-based suspicious element, which can effectively avoid the above misrecognition situation and improve the recognition accuracy.

[0072] In a possible implementation manner, the above-mentioned element cheating detection result of the suspicious element determined according to the recognized target text is mainly determined based on the cheating keyword matching strategy, and specifically may include: detecting the number of cheating matching keywords included in the target text, and then determining the element cheating detection result of the suspicious element in the partial image according to the relationship between the number of cheating matching keywords in the target text and the preset matching word number threshold.

[0073] The cheating keyword refers to a keyword used to indicate the cheating function, which can be mined in advance from relevant cheating scenarios or set manually according to experience; as an example, the preset cheating keywords may include automatic aiming, perspective, viewing the health bar of the opponent player, and viewing the distance of the opponent player, etc. The cheating matching keyword refers to the word in the target text that matches the preset cheating keyword. The cheating matching keyword can be the same as the preset cheating keyword, or a word with similar or the same semantics as the preset cheating keyword. Here, the present application does not specifically limit the cheating matching keyword.

[0074] It should be noted that the preset matching word number threshold is used to indicate the value range of the element cheating detection result corresponding to the number of cheating matching keywords. As an example, the element cheating detection result can be divided into a low-probability cheating trace, a medium-probability cheating trace, and a high-probability cheating trace according to the possibility that the suspicious element belongs to the real cheating trace; correspondingly, the preset matching word number threshold corresponding to the low-probability cheating trace can be 0-2. When the number of cheating matching keywords is within the range of 0-2, it means that the element cheating detection result of the suspicious element is a low-probability cheating trace; the preset matching word number threshold corresponding to the medium-probability cheating trace is 3-5. When the number of cheating matching keywords is within the range of 3-5, it means that the element cheating detection result of the suspicious element is a medium-probability cheating trace; the preset matching word number threshold corresponding to the high-probability cheating trace is 6-10. When the number of cheating matching keywords is within the range of 6-10, it means that the element cheating detection result of the suspicious element is a high-probability cheating trace.

[0075] Therefore, after determining the number of cheating matching keywords in the target text through the above method, according to the relationship between the preset matching word number threshold and the number of cheating matching keywords in the target text, the element cheating detection result of the text type suspicious element in the local image can be determined. In this way, through the corresponding relationship between the preset matching word number threshold and the number of cheating matching keywords, the cheating detection of the text type suspicious element can be quickly completed, and at the same time, a relatively accurate element cheating detection result can be obtained.

[0076] When the element type of the suspicious element is a marker type, the local image can be classified through the image classification model corresponding to the marker type, so that the corresponding element cheating detection result can be obtained.

[0077] It should be noted that the element type of the suspicious element being a marker type means that the suspicious element is embodied in the local image in the form of a marker. As an example, the marker type may include a square, a skeleton, and blood volume, etc. The image classification model is used to classify the local image including the suspicious element, so as to determine whether the local image is a cheating image, such as determining the probability that the local image belongs to a cheating image, and further determining whether the suspicious element of the marker type in it is a real cheating trace, such as determining the probability that the suspicious element in it is a real cheating trace.

[0078] As an example, the image classification model can be a multi-classification model, which is used to identify the possibility that the local image is a cheating image of multiple types. Here, the cheating images of multiple types correspond one-to-one with multiple marker types. For example, assuming that the multiple marker types include square type, skeleton type, and blood volume type, then this image classification model can be used to identify the probability that the input local image belongs to a square type cheating image, the probability that it belongs to a skeleton type cheating image, and the probability that it belongs to a blood volume type cheating image. Furthermore, the cheating image type with the highest probability is determined as the cheating image type to which the local image belongs. Correspondingly, the suspicious element included in the local image is the marker type corresponding to the cheating image type. In this way, using an image classification model that performs a multi-classification task to identify the cheating image type to which the local image belongs can reduce the models to be deployed, reduce the model deployment cost, and reduce the occupation of relevant resources.

[0079] As another example, the image classification model can be a binary classification model. In the case of including multiple types of markers, it is correspondingly necessary to deploy the image classification models corresponding to these multiple types of markers respectively, that is, there is a one-to-one correspondence between the marker types and the image classification models, and different marker types correspond to different image classification models. Correspondingly, the local image can be classified by the image classification model corresponding to the marker type to which the suspicious element belongs, and the probability that the local image belongs to a cheating image can be obtained. Thus, the element cheating detection result can be determined according to the relationship between the probability that the local image belongs to a cheating image and the preset probability threshold.

[0080] That is, the image classification model (binary classification model) corresponding to the marker type to which the suspicious element belongs can be called to analyze and process the local image corresponding to the target position, so as to determine the probability that the local image belongs to a cheating image; it should be understood that the probability that the local image belongs to a cheating image here specifically represents the probability that the local image belongs to the cheating image corresponding to the marker type, that is, it represents the probability that the suspicious element in the local image belongs to the cheating element of the marker type.

[0081] Exemplarily, the above-mentioned multiple types of markers can be divided into box types, skeleton types, blood volume types, etc. Correspondingly, the image classification models corresponding to the multiple types of markers can be divided into box-type image classification models, skeleton-type classification models, and blood volume-type image classification models. The box-type image classification model is used to classify and process the local image including box-type suspicious elements to determine whether the local image to which the box-type suspicious element belongs is a cheating image. The output result after the classification and processing of the box-type image classification model includes the probability that the local image belongs to a cheating image, that is, the output result after the classification and processing includes the probability that the box-type suspicious element in the local image belongs to the true cheating trace of the box type; the skeleton-type image classification model is used to classify and process the local image including skeleton-type suspicious elements to determine whether the local image to which the skeleton-type suspicious element belongs is a cheating image. The output result after the classification and processing of the skeleton-type image classification model includes the probability that the local image belongs to a cheating image, that is, the output result after the classification and processing includes the probability that the skeleton-type suspicious element in the local image belongs to the true cheating trace of the skeleton type; the blood volume-type image classification model is used to classify and process the local image including blood volume-type suspicious elements to determine whether the local image to which the blood volume-type suspicious element belongs is a cheating image. The output result after the classification and processing of the blood volume-type image classification model includes the probability that the local image belongs to a cheating image, that is, the output result after the classification and processing includes the probability that the blood volume-type suspicious element in the local image belongs to the true cheating trace of the blood volume type.

[0082] It should be noted that the preset probability threshold is a probability threshold used to measure whether a local image belongs to a cheating image, that is, the value range of the corresponding element cheating detection result. As an example, the element cheating detection result can be divided into low-probability cheating traces, medium-probability cheating traces, and high-probability cheating traces; correspondingly, the preset probability threshold corresponding to the low-probability cheating trace is 0-0.2. When the probability that the local image belongs to the cheating image is within the range of 0-0.2, it means that the element cheating detection result of the suspicious element in the local image is a low-probability cheating trace; the preset probability threshold corresponding to the medium-probability cheating trace is 0.3-0.5. When the probability that the local image belongs to the cheating image is within the range of 0.3-0.5, it means that the element cheating detection result of the suspicious element in the local image is a medium-probability cheating trace; the preset probability threshold corresponding to the high-probability cheating trace is 0.6-1. When the probability that the local image belongs to the cheating image is within the range of 0.6-1, it means that the element cheating detection result of the suspicious element in the local image is a high-probability cheating trace.

[0083] Thus, by using the image classification model dedicated to the marker type to which the suspicious element belongs to classify the local image, the accuracy of cheating detection is improved. At the same time, the method for determining the element cheating detection result of the suspicious element of the marker type in the local image based on the relationship between the probability that the local image belongs to the cheating image and the preset probability threshold is more efficient and accurate.

[0084] Exemplarily, the image classification model used in the above embodiment can be a model with a residual structure. Refer to Figure 5 , Figure 5 which is the structure diagram of the image classification model provided by the embodiment of the present application. The image classification model is a model with a residual structure. As an example, the model structure of the image classification model can adopt a residual structure block similar to the Resnet model architecture, and the number of model layers can be 80 layers.

[0085] Specifically, the model structure can include three layers of residual structure. First, the local image containing the suspicious element can be used as the input intput(X) of the model. Then, after the input local image passes through three convolutions with different dimensions, it is input into the coarse-grained and fine-grained attention module for feature extraction. Finally, the output result output(X”) of the coarse-grained and fine-grained attention module and the input local image can be input into the gating mechanism for weighted fusion, so as to obtain the output result output(X’), that is, the probability that the local image belongs to the cheating image.

[0086] Among them, the three-layer convolutional structure can exemplarily include that the convolutional kernel size in the first convolutional layer (conv(1*1, s = 1)) is 1*1, and the stride S of the convolutional layer is 1; the convolutional kernel size in the second convolutional layer (conv(3*3, s = 1)) is 3*3, and the stride S of the convolutional layer is 1; the convolutional kernel size in the third convolutional layer (conv(1*1, s = 1)) is 1*1, and the stride S of the convolutional layer is 1.

[0087] The coarse-grained and fine-grained attention module can specifically include that the feature image after three-layer convolution of the input intput(X) of the image classification model can be used as the input intput(X’) of the coarse-grained and fine-grained attention module. Then, based on the coarse-grained attention mechanism and the fine-grained attention mechanism respectively, feature extraction can be performed on the convolved feature image (C, H, W), where C represents the number of channels of the feature image, H represents the height of the feature image, and W represents the width of the feature image. Among them, the coarse-grained attention mechanism mainly performs feature extraction based on the image aspect, including an average pooling layer and a linear fully connected layer. The fine-grained attention mechanism mainly performs feature extraction based on the pixel aspect, including a 1*1 convolutional layer and a 3*3 convolutional layer. The coarse-grained attention mechanism is used to extract the information at the whole-image level of the feature image, and the fine-grained attention mechanism is used to extract the feature information at the pixel level of the feature image. In this way, feature extraction in two different dimensions at the whole-image level and the pixel level is realized, ensuring that the information extracted from the feature image has high reference value and no valid information is omitted.

[0088] After that, the output result (C, 1, 1) of the coarse-grained attention mechanism and the output result (C, H, W) of the fine-grained attention mechanism can be fused and input into the loss function sigomod to calculate the loss value. Finally, after multiplying the loss value by the input intput(X’) of the coarse-grained and fine-grained attention module, the output feature map of the coarse-grained and fine-grained mechanism can be obtained. Then, after adding the output feature map of the coarse-grained and fine-grained mechanism and the input intput(X’) of the coarse-grained and fine-grained attention module, the output result output(X”) of the coarse-grained and fine-grained attention module can be obtained.

[0089] Through the above image classification model with a residual structure, classifying the local image can comprehensively consider the coarse-grained image information at the whole-image level and the fine-grained image information at the pixel level. Thus, it is ensured that the information contained in the local image is comprehensively considered during the image classification process, and further, it is ensured that the determined classification result has high accuracy.

[0090] It should be understood that the above image classification model can specifically be the image classification model for performing a multi-classification task mentioned above, or the image classification model for performing a binary-classification task mentioned above.

[0091] See Figure 6 , Figure 6 which is a schematic architecture diagram of the data processing method provided in the embodiment of the present application. Specifically, it may include: First, the entire game screenshot can be output to the target detection model, so that the target detection model can detect whether there are suspicious elements in the game screenshot that may correspond to cheating traces. When there are suspicious elements in the game screenshot, the target detection model can locate the target position of the suspicious elements in the game screenshot and the element type of the suspicious elements. Taking the detection of the function directory and bone markers of the game cheating software as an example, the detection results output by the target detection model include the position of the function directory of the game cheating software and its element type belonging to the text category, and also include the position of the bone markers and its element type belonging to the bone category.

[0092] Then, operations such as expansion, interception, and cropping can be performed on the positions of the detected suspicious elements to form local images of the suspicious elements. For text-type suspicious elements, their corresponding local images can be input into the image text recognition model for processing to obtain the target text included therein, and the corresponding element cheating detection result can be determined according to the number of cheating matching keywords included in the target text. For bone-type suspicious elements, their corresponding local images can be input into the bone image classification model for processing to determine the probability that the bone markers belong to real cheating traces.

[0093] Thus, by distinguishing the image processing models corresponding to the text type and the marker type respectively, more adaptable and refined detection can be performed on the local images of different types of suspicious elements to obtain more accurate element cheating detection results, which helps to improve the accuracy of subsequent cheating detection.

[0094] Step 204: Determine whether the target object uses game cheating software according to the element cheating detection result.

[0095] The element cheating detection result is used to indicate whether the suspicious elements in the local image are real cheating traces. When the element cheating detection result indicates that the suspicious elements in the local image are real cheating traces (such as medium-probability cheating traces and high-probability cheating traces in the above text), it can be considered that the target object uses game cheating software, and then corresponding automatic penalty measures can be taken to punish the target object. For example, the account of the target object can be banned. When the element cheating detection result indicates that the suspicious elements in the local image are not real cheating traces, it can be considered that the target object does not use game cheating software and no treatment is required.

[0096] In a possible implementation, when the image to be detected is a single image, it is possible to determine whether the target object has used a game cheat based on the number of cheating elements among multiple suspicious elements in the image to be detected. Specifically, it may include: when there are multiple suspicious elements in the image to be detected, the number of cheating elements among the multiple suspicious elements can be determined according to the element cheating detection results corresponding to each of the multiple suspicious elements; then, based on the relationship between the number of cheating elements and a preset element quantity threshold, it can be determined whether the target object has used a game cheat.

[0097] The multiple suspicious elements in the image to be detected can be suspicious elements of the same element type or not of the same element type. The present application does not specifically limit the multiple suspicious elements in the image to be detected. The element cheating detection results corresponding to each of the multiple suspicious elements are obtained through the above step 203. Exemplarily, the element cheating detection results can be divided into low-probability cheating traces, medium-probability cheating traces, and high-probability cheating traces; here, the suspicious elements determined to belong to medium-probability cheating traces and high-probability cheating traces can be regarded as cheating elements, while the suspicious elements determined to belong to low-probability cheating traces can be regarded as non-cheating elements.

[0098] Correspondingly, when determining whether the target object has used a game cheat based on the relationship between the number of cheating elements and a preset element quantity threshold, an element quantity threshold can be set in advance. For example, 2 is set as the element quantity threshold. When the number of cheating elements in the image to be detected reaches or exceeds 2, it can be considered that the target object has used a game cheat. Or, several element quantity thresholds can also be set to divide different intervals. For example, element quantity thresholds 1 and 2 can be set. When the number of cheating elements in the image to be detected reaches or exceeds 2, it is considered that the target object has probably used a game cheat. When there is 1 cheating element in the image to be detected, it is considered that the target object has used a game cheat with medium probability. For the above different situations of using game cheats, different penalty measures can be adopted to penalize the target object.

[0099] When the images to be detected are multiple images, it is possible to determine whether the target object has used a game cheat based on the number of suspected cheating images among the multiple images to be detected. Specifically, it may include: when obtaining multiple images to be detected, the number of suspected cheating images among the multiple images to be detected can be determined based on the element cheating detection results corresponding to the suspicious elements in the multiple images to be detected; then, based on the relationship between the number of suspected cheating images and a preset image quantity threshold, it can be determined whether the target object has used a game cheat.

[0100] The element cheating detection results corresponding to the suspicious elements in multiple images to be detected refer to the element cheating detection results in multiple images to be detected obtained by detecting the multiple images to be detected through the above-mentioned steps 202 and 203 respectively. As an example, the element cheating detection results can be divided into low-probability cheating traces, medium-probability cheating traces, and high-probability cheating traces. When the element cheating detection results of the suspicious elements in the image to be detected only contain low-probability cheating traces, it can be considered that the image to be detected is not a suspected cheating image; when the element cheating detection results of the suspicious elements in the image to be detected contain one or more medium-probability cheating traces or high-probability cheating traces, it can be considered that the image to be detected is a suspected cheating image; of course, in practical applications, other mechanisms can also be adopted to determine whether the image to be detected is a suspected cheating image according to the element cheating detection results corresponding to the suspicious elements included in the image to be detected, and the embodiments of the present application do not make any limitations on this.

[0101] Correspondingly, when determining whether the target object uses a game cheat according to the relationship between the number of suspected cheating images and the preset image number threshold, the image number threshold can be set in advance. For example, the preset image number threshold is set to 2. When the number of suspected cheating images in multiple images to be detected is 2 or more than 2, it can be determined that the target object uses a game cheat; on the contrary, when the number of suspected cheating images in multiple images to be detected is less than 2, it can be determined that the target object does not use a game cheat. Or, several image number thresholds can also be set to divide different intervals. For example, the image number thresholds can be 1 and 2. When the number of suspected cheating images in multiple images to be detected is 1, it can be considered that the target object uses the game cheat with medium probability. When the number of suspected cheating images in multiple images to be detected is greater than or equal to 2, it is considered that the target object uses the game cheat with high probability. For the above different situations of using game cheats, different punishment measures can be taken to punish the target object.

[0102] In addition, when making a determination according to the above-set element number threshold interval or according to the above-set image number threshold interval, and it is considered that the target object uses the game cheat with high probability and medium probability, the target object can be automatically punished to increase the proportion of automatic punishment, realize timely response to cheating behaviors, and at the same time reduce the manual review cost; when it is considered that the target object uses the game cheat with low probability, the image to be detected can be manually reviewed again to determine whether the target object uses the game cheat, and when it is determined that the target object uses the game cheat, the punishment measures for the target object are configured manually. For example, the account of the target object is banned for one week. The manual review and punishment of this situation ensure the coverage of cheating detection and avoid the occurrence of mis-punishment.

[0103] Thus, based on the number of cheating elements in a single image to be detected and / or the number of suspected cheating images among multiple images to be detected, determining whether the target object uses game cheats enriches the application scenarios of the element cheating detection results and improves the coverage of cheating detection.

[0104] In the data processing method provided in the embodiments of the present application, a detection mechanism for cheating detection of game screenshots based on two stages is provided. According to the detection architecture of first overall and then local, it is detected whether there are visual cheating elements generated by game cheats in the game screenshots. First, obtain the image to be detected, and the image to be detected is a game screenshot of the target object. Then, through the target detection model, it is detected whether there are suspicious elements in the image to be detected, and when it is determined that there are suspicious elements in the image to be detected, the position of the suspicious elements and the element type of the suspicious elements in the image to be detected can be located, thus realizing the first-stage overall detection of the image to be detected. Furthermore, when it is determined that there are suspicious elements in the image to be detected through the first-stage overall detection, a second-stage local detection is further started for the image to be detected, that is, the local image where the suspicious elements in the image to be detected belong can be processed through the image processing model corresponding to the element type of the suspicious elements, so as to determine whether the suspicious elements are real cheating traces. Since the image processing model corresponds to the element type of the suspicious elements, by using the image processing model dedicated to processing this element type to detect the local image, the obtained element cheating detection result has higher accuracy. At the same time, the image processing model processes the local image corresponding to the target position where the suspicious elements are located, and the local image is determined based on the target position of the suspicious elements in the image to be detected, with less interference information. Therefore, the influence of irrelevant information on the image processing model can be reduced, thereby improving the accuracy of the determined element cheating detection result. Finally, based on the relatively accurate element cheating detection result, it can be accurately determined whether the target object uses game cheats, improving the accuracy of the detection result of game cheats.

[0105] In a possible implementation, in order to further enhance the detection intensity for visual visibility cheating, the present application can also use the overall image classification model to detect the image to be detected. That is, the method provided in the embodiments of the present application further includes: processing the image to be detected through the overall image classification model to determine the overall image cheating detection result corresponding to the image to be detected; correspondingly, step 204 above can specifically include: determining whether the target object uses game cheats according to the element cheating detection result and the overall image cheating detection result.

[0106] The overall image classification model can be a full-type classification model that does not distinguish specific cheating types and is only used to distinguish whether the image to be detected is a cheating image. That is, the overall image classification model can divide the input image into two categories for the input image; one category is cheating images, which may include any type of cheating traces, such as text-based cheating traces, box-based cheating traces, bone-based cheating traces, and blood volume-based cheating traces, etc.; the other category is non-cheating images, that is, normal images without cheating traces. The overall image classification model can output the probability that the image to be detected belongs to a cheating image.

[0107] It should be noted that the overall image classification model can also be a model with a residual structure, that is, the overall image classification model can also adopt the Figure 5 model structure shown. Of course, the overall image classification model can also be of other structures, and the embodiments of the present application do not make any limitations in this regard. In addition, since the proportion of general cheating traces in the whole image is usually small, before inputting the image to be detected into the overall image classification model, the resolution of the image to be detected can be increased, that is, a larger input resolution can be adopted to increase the coverage of fine cheating traces and improve the detection accuracy of the overall image classification model.

[0108] As an example, when training the above overall image classification model, first, a general image classification model can be trained based on historical cheating game images of multiple different game applications to obtain a basic model for general cheating detection in games; then, based on a small amount of sample data of a specific game application, such as historical cheating game images of the game application used by the target object, the basic model can be fine-tuned to obtain an overall image classification model for identifying whether a game screenshot in a specific game application is a cheating image. Since the labels of the training samples of the overall image classification model only need to indicate whether it is a cheating image and do not need to indicate the cheating type, the sample acquisition cost is relatively low. That is, the audit data reviewed by historical manual audits can be directly used as training samples, reducing the labeling cost while enabling the overall image classification model to be trained and applied quickly.

[0109] In addition, during the training process of the overall image classification model, it is also necessary to test the trained overall image classification model to test the current performance of the overall image classification model. For the test samples used in the test process, they can be obtained based on the detection results of other cheating detection methods. Exemplarily, in practical applications, in addition to using the cheating detection method provided by the embodiments of the present application for cheating detection, other cheating detection methods can also be used for cheating detection. On this basis, according to the detection results determined by other cheating detection methods, the corresponding game screenshots can be obtained as the test samples of the above overall image classification model.

[0110] For example, during the game process of the target object, the operation data of the target object can be captured. Then, the captured operation data can be compared with the operation data under normal circumstances. According to the comparison result, it can be determined whether the target object uses game cheats. If it is determined that the target object uses game cheats, the corresponding game interface can be captured, and the captured game screenshot is the cheating image. Thus, the captured cheating image can be used as the test black sample of the overall image classification model. Correspondingly, if it is determined that the target object does not use game cheats, the game interface in the current game application can be captured, and the captured game screenshot this time is a non-cheating image, that is, it can be used as the test white sample of the overall image classification model.

[0111] Then, the test sample can be input into the overall image classification model for detection, so that the test detection result corresponding to the test sample can be determined. The test detection result corresponding to the test sample is used to represent whether the test sample is a cheating image. If the overall image classification model cannot accurately classify a certain test sample, for example, identifying the test black sample as a non-cheating image, or identifying the test white sample as a cheating image, then this test sample can be used as the training sample for subsequent optimizing and training of the overall image classification model. In this way, obtaining the test sample based on the cheating detection result of other cheating detection methods can reduce the acquisition cost of the test sample and utilize the performance of the overall image classification model more accurately.

[0112] To avoid the occurrence of misjudgment and wrong punishment, on the basis of determining whether the target object uses the cheating mechanism based on the element cheating detection result, the whole-image cheating detection result can also be combined to determine whether the target object uses the cheating mechanism, which not only takes into account the cheating traces in the local area but also further determines the cheating situation in the image from the whole-image scenario. Since there may be situations where there are similar cheating traces but non-cheating traces in the local image, the cheating situation in the whole image can be combined for discrimination and punishment, and at the same time, the conditions for discrimination and punishment based on the element cheating detection result are also expanded, ensuring the coverage of discrimination and punishment.

[0113] As an example, the whole-image cheating detection result can be divided into cheating images and non-cheating images (for example, it can be determined according to the relationship between the image to be detected represented by the whole-image cheating detection result and the preset probability threshold), and the element cheating detection result can be divided into real cheating traces (such as the medium-probability cheating traces and high-probability cheating traces mentioned above) and false cheating traces (such as the low-probability cheating traces mentioned above). Therefore, when the whole-image cheating detection result is a cheating image and the element cheating detection result is a real cheating trace, it can be considered that the target object uses game cheats; when the whole-image cheating detection result is a non-cheating image and the element cheating detection result is a false cheating trace, it can be considered that the target object does not use game cheats.

[0114] Specifically, based on the element cheating detection result and the whole-image cheating detection result, determining whether the target object uses game cheats may include: determining whether the target object uses game cheats based on the relationship between the probability that the suspicious element characterized by the element cheating detection result is a cheating element and the preset cheating element information threshold, and the relationship between the probability that the image to be detected characterized by the whole-image cheating detection result is a cheating image and the preset cheating whole-image probability threshold.

[0115] In a possible implementation manner, the probability that the suspicious element characterized by the element cheating detection result is a cheating element, that is, the probability that the element cheating detection result indicates that the suspicious element is a cheating element. The preset cheating element information threshold is used to indicate the probability thresholds corresponding to different probability value ranges in the element cheating detection result, and different probability value ranges may correspond to different element cheating detection results. For example, the preset cheating element information threshold can be divided into 3 levels: low-probability cheating traces, medium-probability cheating traces, and high-probability cheating traces. Among them, the probability value range corresponding to the low-probability cheating traces can be 0 - 0.2. Correspondingly, when the probability value characterized by the element cheating detection result is within this interval, it indicates that the corresponding suspicious element is a low-probability cheating trace; the probability value range corresponding to the medium-probability cheating traces can be 0.3 - 0.6. Correspondingly, when the probability value characterized by the element cheating detection result is within this interval, it indicates that the corresponding suspicious element is a medium-probability cheating trace; the probability value range corresponding to the high-probability cheating traces can be 0.7 - 1. Correspondingly, when the probability value characterized by the element cheating detection result is within this interval, it indicates that the corresponding suspicious element is a high-probability cheating trace.

[0116] Correspondingly, the preset cheating whole-image probability threshold is used to indicate the probability thresholds corresponding to different probability value ranges in the whole-image cheating detection result, and different probability value ranges may correspond to different whole-image cheating detection results. For example, the preset cheating whole-image probability threshold can be divided into 3 levels: low-probability cheating images, medium-probability cheating images, and high-probability cheating images. The probability value range corresponding to the low-probability cheating images can be 0 - 0.2. Correspondingly, when the probability value characterized by the whole-image cheating detection result is within this interval, it indicates that the corresponding whole-image cheating detection result is a low-probability cheating image; the probability value range corresponding to the medium-probability cheating images can be 0.3 - 0.6. Correspondingly, when the probability value characterized by the whole-image cheating detection result is within this interval, it indicates that the corresponding whole-image cheating detection result is a medium-probability cheating image; the probability value range corresponding to the high-probability cheating images can be 0.7 - 1. Correspondingly, when the probability value characterized by the whole-image cheating detection result is within this interval, it indicates that the corresponding whole-image cheating detection result is a high-probability cheating image.

[0117] As an example, when the whole-image cheating detection result is a low-probability cheating image and the element cheating detection result is a low-probability cheating trace, it can be considered that the target object does not use game cheats. When the whole-image cheating detection result is either a medium-probability cheating image or a high-probability cheating image, and the element cheating detection result is either a medium-probability cheating trace or a high-probability cheating trace, it can be considered that the target object uses game cheats, and corresponding penalty measures can be taken against the target object. For example, the account of the target object can be banned for one month. Thus, real-time automatic penalty for cheating is achieved, the automatic penalty rate is increased, manual review is not required, the timeliness of cheat penalty is satisfied, and the possibility of the target object cheating is suppressed.

[0118] In addition, for some cases, manual review can also be carried out. For example, when the whole-image cheating detection result is a low-probability cheating image, or the element cheating detection result is a low-probability cheating trace, the whole-image cheating detection result and the element cheating detection result can be manually reviewed again to determine whether the target object uses game cheats. When it is determined that the target object uses game cheats, penalty measures can be taken manually. For example, the account of the target object can be banned for one week. At the same time, a small part is manually reviewed and punished to ensure the cheating detection and thus the coverage of punishing the target object.

[0119] Thus, based on the relationship between the element cheating detection result and the preset cheating element information threshold, and the relationship between the whole-image cheating detection result and the preset cheating whole-image probability threshold, it can be determined whether the target object uses game cheats. In this way, the judgment conditions for determining whether the target object uses game cheats are increased, the accuracy of cheating detection is ensured, and the timeliness of punishment is guaranteed.

[0120] See Figure 7 , Figure 7 which is a schematic diagram of another data processing method provided by the embodiment of the present application. Specifically, it introduces the method for cheating detection of games with visually visible cheats in the embodiment of the present application, which can be divided into two branches. One branch is for cheating detection based on the overall image classification model, and the other branch is for cheating detection based on the object detection model, image text recognition model, box-type image classification model, bone-type image classification model, and blood volume-type image classification model.

[0121] As an example, first, the entire game screenshot can be obtained, and then the entire game screenshot can be input into the overall image classification model for processing, and the whole-image cheating detection result corresponding to the game screenshot can be obtained. This whole-image cheating detection result is used to indicate the probability that the game screenshot belongs to a cheating image.

[0122] Meanwhile, the entire game screenshot can also be input into the object detection model for detection. The object detection model can detect whether there are suspicious elements in the game screenshot, and when it is determined that there are suspicious elements in the game screenshot, it can determine the position of the suspicious elements in the game interface and the element type of the suspicious elements.

[0123] When the element type of the suspicious element is a text type, a local image containing the text type suspicious element can be cropped according to the position of the text type suspicious element in the game screenshot, that is, the local image of the text type suspicious element. Then, the local image of the text type suspicious element can be input into the image text recognition model for processing to obtain the target text in the local image. Furthermore, according to the recognized target text, the element cheating detection result corresponding to the local image of the text type suspicious element can be determined. When the element type of the suspicious element is a box type, a bone type, or a blood volume type, a local image containing the suspicious element can be cropped according to the position of the suspicious element in the game screenshot. Then, the local image can be input into the image classification model corresponding to the element type of the suspicious element for classification processing, and the corresponding element cheating detection result can be obtained.

[0124] Finally, the overall detection result of the first branch and the element cheating detection result of the second branch can be combined to determine whether the target object uses game cheats. And according to the corresponding determination strategy, the punishment method for the target object can also be determined. For example, when the overall cheating detection result is a medium probability cheating image and the element cheating detection result is a medium probability cheating trace, manual secondary review can be performed based on the overall cheating detection result and the element cheating detection result to determine whether the target object uses game cheats. Or, when the overall cheating detection result is a high probability cheating image and the element cheating detection result is a high probability cheating trace, it is determined that the target object uses game cheats, and punishment measures can be automatically implemented for the target object.

[0125] In a possible implementation manner, the object detection model in step 202 above can be trained through the following steps:

[0126] Step 301: Obtain a game screenshot as the first training sample; the first training sample is a first training black sample or a first training white sample, and the label corresponding to the first training black sample is used to indicate the position and element type of the cheating elements in the first training black sample, and the label corresponding to the first training white sample is used to indicate that there are no cheating elements in the first training white sample.

[0127] In practical applications, before the server trains an object detection model, a large number of first training samples for training the object detection model need to be obtained. The first training samples are entire game screenshots. Exemplarily, the server can use historical game images as the above-mentioned first training samples, and the historical game images are game screenshots generated historically.

[0128] Among the obtained first training samples, there may be first training black samples and first training white samples. Among them, the first training black samples are game screenshots belonging to cheating images, which include visible cheating elements generated by game cheats. Correspondingly, the labels corresponding to the first training black samples are used to indicate the positions and element types of the cheating elements in the first training black samples. The first training white samples are game screenshots belonging to non-cheating images, that is, the first training white samples are game screenshots belonging to normal game images, which do not include visible cheating elements generated by game cheats. Correspondingly, the labels corresponding to the first training white samples are used to indicate that there are no cheating elements in the first training white samples.

[0129] In addition, the first training samples can be obtained by retrieving from a database or from the server. As an example, the first training black samples may include text-based cheating elements and marker-based cheating elements. Among them, the marker-based cheating elements can further include box-based cheating elements, skeleton-based cheating elements, and health-based cheating elements.

[0130] In a possible implementation manner, the embodiments of the present application can obtain the above-mentioned first training black samples through at least one of the following methods:

[0131] The first method: Perform clustering processing on multiple historical cheating game images to obtain multiple clusters, and then n historical cheating game images can be extracted from each cluster, where n is an integer greater than or equal to 1, as the first training black samples. At the same time, the positions and element types of the cheating elements in the extracted historical cheating images can be marked, so that the labels corresponding to the first training black samples can be obtained.

[0132] Clustering processing refers to dividing a data set into different clusters according to a specific standard, so that the similarity of data objects within the same cluster is as large as possible, and the difference between data objects not in the same cluster is also as large as possible. It is often used in the image processing process.

[0133] As an example, the above-mentioned historical cheating game images can be cheating images accumulated after manual review. In the embodiment of the present application, first, an image feature extraction model (such as a convolutional neural network model) can be used to perform feature extraction processing on multiple acquired historical cheating game images respectively to obtain image feature vectors corresponding to each of the multiple historical cheating game images; then, a specific clustering algorithm (such as the K-Means algorithm, the KNN algorithm, etc.) can be used to perform clustering processing based on the image feature vectors corresponding to each of the multiple historical cheating game images to obtain multiple clustering clusters; furthermore, n historical cheating game images can be randomly selected from each clustering cluster as the first training black samples, or n historical cheating game images with a relatively close distance to the cluster center can be selected from each clustering cluster as the first training black samples. In this way, the first training black samples with a wider coverage and more comprehensive and rich categories can be obtained.

[0134] Since the first training black samples contain cheating elements, the positions and element types of the cheating elements in the first training black samples can be marked through a labeling program or manual labeling as the labels of the first training black samples.

[0135] The second method: Add template cheating elements to the historical normal game images to obtain the first training black samples, and use the added positions and element types of the template cheating elements as the labels corresponding to the first training black samples.

[0136] In addition, an automatic generation method of cheating elements can also be adopted to generate more first training black samples. Refer to Figure 8a , Figure 8a which is a schematic diagram for constructing the first training black samples provided by the embodiment of the present application. Cheating elements of various colors and sizes can be randomly generated, such as solid-line box-type cheating marks, dotted-line box-type cheating marks, bone-type cheating marks, text-type cheating marks, and blood volume-type cheating marks, etc., and randomly superimposed on the historical normal game images to obtain the first training black samples, and mark the positions and element types of the self-constructed cheating elements as the labels of the first training black samples. For example, box-type cheating elements can be added to the historical normal game images, so that the first training black samples with the positions of the box-type cheating elements marked and the element type marked as box-type can be obtained. Thus, the efficiency of constructing training samples is improved, and the labeling cost is reduced at the same time.

[0137] It should be noted that when randomly generating text-type cheating marks, different fonts and colors can be used, and the randomly generated cheating marks can be randomly deployed in the historical normal game images. The present application does not specifically limit the positions of the randomly generated cheating marks, which can be located on the left, right, or middle of the historical normal game images.

[0138] In addition, for some cheating traces that are difficult to reproduce through program simulation, or to ensure that the randomly generated cheating traces are more consistent with the cheating traces in the real environment, cheating elements can be extracted from historical cheating images as template cheating elements. Furthermore, the above template cheating elements can be randomly superimposed on historical normal game images to generate the first training black samples.

[0139] The template cheating element refers to a template composed of cheating elements, which is used to generate the first training black samples, so as to ensure the diversity and effectiveness of the first black samples.

[0140] As an example, reference can be made to Figure 8b , Figure 8b which is another schematic diagram for constructing the first training black samples provided by the embodiments of this application. First, an image containing cheating elements can be intercepted from historical cheating game images. Then, the intercepted image of the cheating elements can be cropped, and the cropped cheating elements can be used as template cheating elements. For example, the cropped cheating traces of the skeletal type can be used as template cheating elements. After that, the template cheating elements can be randomly superimposed on historical normal game images, and at the same time, the positions where the template cheating elements are added and the corresponding element types are marked as labels for distinguishing different first training black samples. Thus, the first training black samples with the positions of the skeletal type cheating elements marked and the element type marked as skeletal can be obtained.

[0141] It should be noted that after adding the template cheating elements to the historical normal game images, the template cheating elements can also be adjusted, such as adjusting the size, position, etc. of the template cheating elements. When the template cheating element is a blocky text type cheating element or a randomly generated blocky text type cheating element, it can also be cropped first and then superimposed on the historical normal game images.

[0142] Thus, through the above method, the first training black samples with diversity and containing effective cheating elements can be constructed efficiently, so that the target detection model can more accurately learn the characteristics of the cheating elements in the first training black samples, as well as the correlation between the positions and element types of the cheating elements, thereby improving the model performance of the target detection model.

[0143] Step 302: Determine the training detection result corresponding to the first training sample through the target detection model to be trained; the training detection result is used to represent whether the first training sample includes cheating elements, and in the case of including the cheating elements, it also represents the predicted position and predicted type of the predicted cheating elements in the first training sample.

[0144] After obtaining the first training sample, the server can input the first training sample into the target detection model to be trained. After analyzing and processing the input first training sample, the target detection model can output the training detection result corresponding to the first training sample. The training detection result corresponding to the first training sample can be used to represent whether the first training sample includes cheating elements. In the case where the first training sample includes cheating elements, the training detection result is also used to represent the predicted position and predicted element type of the cheating elements in the first training detection sample. It should be understood that the predicted position is the position of the cheating elements determined by the trained target detection model in the first training sample, and the predicted element type is the element type to which the cheating elements determined by the trained target detection model belong.

[0145] It should be understood that the target detection model here can be any neural network model capable of performing image-based target detection tasks. For example, it can be the YOLOV7 model. The embodiments of the present application do not make any limitations on the model structure of the target detection model here.

[0146] Step 303: Train the target detection model according to the difference between the label corresponding to the first training sample and the training detection result.

[0147] After obtaining the training detection result corresponding to the first training sample, the difference between the label corresponding to the first training sample and the training detection result can be calculated through a loss function. Thus, based on the calculated loss value, the target detection model can be trained, that is, with the goal of reducing the loss value, by continuously adjusting the parameters of the target detection model, the detection performance of the target detection model is optimized. The present application does not specifically limit the loss function in the training process of the target detection model.

[0148] It should be understood that in practical applications, a training end condition can be preset in advance. When the training of the target detection model reaches this training end condition, it can be considered that the training of the target detection model ends. The training end condition can be, for example, that the number of training rounds of the target detection model reaches a preset round threshold, or for example, the performance of the target detection model is tested and it is found that the performance of the target detection model reaches a preset performance standard (such as reaching a preset detection accuracy, etc.), or for example, the performance of the target detection model is tested and it is found that the performance of the target detection model no longer improves significantly as the training progresses. The embodiments of the present application do not make any limitations on this training end condition here.

[0149] Thus, through the above training steps, the final object detection model can be obtained, enabling the trained object detection model to accurately detect cheating elements in the entire game screenshot, that is, it can accurately determine whether there are cheating elements in the entire game screenshot, and can also accurately determine the position and element type of the cheating elements when there are cheating elements in the entire game screenshot. In this way, the preliminary cheating detection is completed, ensuring the coverage of cheating detection.

[0150] During the process of model training, it is also necessary to test the trained object detection model to test the current performance of the object detection model, so as to ensure the detection accuracy of the object detection model. In one possible implementation, the above object detection model can be tested in the following way:

[0151] First, a game screenshot serving as the first test sample can be obtained. The first test sample is the first test black sample or the first test white sample. The first test black sample is a cheating game image, that is, the first test black sample contains cheating elements, and the first test white sample is a normal game image, that is, the first test white sample does not contain cheating elements. In addition, the acquisition method of the first test sample includes obtaining it from a database or from a server.

[0152] As an example, the first test sample can be composed of a large number of game screenshots only labeled with black and white labels. That is, the labels in the first test sample are only used to indicate whether the first test sample is a cheating image. When the game screenshot is a non-cheating image, it is labeled with a white label, that is, the first test white sample; when the game screenshot is a cheating image, it is labeled with a black label, that is, the first test black sample. It should be noted that for the first test black sample, its corresponding label only needs to indicate that it is a cheating image, and there is no need to specifically indicate the position and element type of the cheating elements included therein. The first test black sample has a lower acquisition cost compared to the first training black sample in the above text.

[0153] As an example, the first test sample can be obtained from historical game images. For example, historical game cheating images are obtained as the first test black sample, and historical normal game images are obtained as the first test white sample. More specifically, the training black sample of the above overall image classification model can be used as the first test black sample, and the training white sample of the above overall image classification model can be used as the first test white sample. In this way, the training samples of the overall image classification model are reused, reducing the acquisition cost of the first test sample.

[0154] After obtaining the object detection model, the first test sample can be input into the object detection model for detection, so as to determine the test detection result corresponding to the first test sample.

[0155] If the first test sample is the first test black sample, and the test detection result corresponding to the first test black sample indicates that the first test black sample does not include cheating elements, it means that the target detection model fails to successfully detect cheating images. Then, the first test black sample can be verified to confirm whether it contains cheating elements. After confirming that it contains cheating elements, the first test black sample can be used as a misjudged sample that is missed, that is, the first test black sample can be used as the first training black sample used when optimizing and training the target detection model, and the positions and element types of the cheating elements in the first test black sample can be marked as the labels corresponding to the first training black sample.

[0156] If the first test sample is the first test white sample, and the test detection result corresponding to the first test white sample indicates that the first test white sample includes cheating elements, it means that the target detection model wrongly detects cheating images. Then, the first test white sample can be used as a misjudged sample, that is, the first test white sample can be used as the first training white sample used when optimizing and training the target detection model; or, the first test white sample can be marked as a white sample class, so that the target detection model can precisely learn the features in the white sample class and improve the detection accuracy of the target detection model. That is, the elements in the first test white sample that are misidentified as cheating elements can be marked as white sample elements.

[0157] Thus, the newly marked misjudged samples that are missed and misjudged samples can be added to the training set of the target detection model, so that the target detection model is re-finely tuned and trained based on the updated training set until the target detection model with the best test detection effect is output as the final target detection model.

[0158] See Figure 9 , Figure 9 For the schematic diagram of the iterative target detection model provided by the embodiment of the present application, specifically, it can include: First, train the target detection model to be trained based on the first training sample to obtain the trained target detection model. Then, the trained target detection model can be tested through the first test sample. If the current target detection model fails in prediction, that is, it detects that the first test black sample does not include cheating elements or detects that the first test white sample includes cheating elements, the first test black sample can be manually marked as a misjudged sample that is missed or a misjudged sample, and the first test white sample can be manually marked as a misjudged sample that is missed or a misjudged sample. Thus, the newly marked misjudged samples that are missed and misjudged samples can be used as the training set to re-finely tune and train the target detection model.

[0159] Thus, the target detection model can be tested in the above manner, and the test samples that the target detection model fails to accurately detect can be added to the training sample set as new training samples. The target detection model is fine-tuned using the updated training sample set, thereby improving the detection accuracy of the target detection model, enabling the target detection model to have better performance, and further learning more information from the images it fails to accurately detect.

[0160] In a possible implementation manner, the image classification model in step 203 can be trained through the following steps:

[0161] Step 401: Obtain second training samples; the second training samples are local images in game screenshots, and the second training samples are second training black samples or second training white samples. The label corresponding to the second training black sample is used to indicate that the second training black sample includes cheating elements, and the label corresponding to the second training white sample is used to indicate that the second training white sample does not include cheating elements.

[0162] In practical applications, before the server trains the image classification model, a large number of second training samples for training the image classification model need to be obtained. The second training samples are local images in game screenshots. As an example, the server can use the local images in historical game images as the second training samples, and the historical game images are game screenshots generated historically.

[0163] Since the image classification model is used to classify local images of suspicious elements of the marker type, the second training samples can include marker-type cheating elements, such as box-type cheating elements, bone-type cheating elements, and blood volume-type cheating elements, etc. It should be noted that when the trained image classification model is a multi-classification model, the obtained second training samples include local images of multiple marker types. When the trained image classification model is a binary classification model, the obtained second training samples only include local images of the marker type corresponding to the image classification model.

[0164] For example, when the trained image classification model is a binary classification model, such as the trained image classification model is a box-type image classification model, the obtained second training samples only include box-type local images; or when the trained image classification model is a bone-type image classification model, the obtained second training samples only include bone-type local images; or when the trained image classification model is a blood volume-type image classification model, the obtained second training samples only include blood volume-type local images.

[0165] Each of the obtained second training samples may include second training black samples and second training white samples. The labels corresponding to the second training black samples are used to indicate that the second training black samples include cheating elements, that is, the second training black samples are partial images of cheating images. Specifically, when the trained image classification model is a multi-classification model, the labels of the second training black samples are also used to indicate the element types of the cheating elements. For example, box-type cheating traces, bone-type cheating traces, and blood volume-type cheating traces; when the trained image classification model is a binary classification model, the labels of the second training black samples are used to indicate that the partial images of the second training black samples belong to cheating images.

[0166] The labels corresponding to the second training white samples are used to indicate that the second training white samples do not include cheating elements, that is, the second training white samples are partial images of non-cheating images. Therefore, by constructing the second training black samples and the second training white samples, the image classification model to be trained can successfully distinguish whether the partial image belongs to a cheating image or a non-cheating image during the classification process of the second training samples, and then determine whether the suspicious elements of the marker type in the partial image are real cheating traces.

[0167] Since the image classification model is mainly used for cheating detection of partial images, the accuracy of cheating detection needs to be ensured. Therefore, during the training process of the image classification model, there are high-quality requirements for the training samples.

[0168] In a possible implementation manner, the embodiments of the present application can construct the second training black samples in the following way: According to the positions of the cheating elements in the historical cheating game images, partial images of a preset size are cropped out as the second training black samples, and the second training black samples include complete cheating elements.

[0169] It should be noted that the historical cheating game images refer to the historically determined cheating images, and the cheating game images include real cheating traces, such as box-type cheating traces, bone-type cheating traces, and blood volume-type cheating traces, etc. The historical cheating game images can be stored in a database, or can also be stored in the historical records of the server.

[0170] As an example, compared with cheating detection of the entire image, the partial image contains less interference information, and the accuracy of cheating detection of the partial image is higher. Therefore, first, the entire historical cheating game image can be obtained from the database, and then, at the positions of the cheating elements in the entire historical cheating game image, partial images of a preset size can be cropped out. For example, if the cheating element in the historical cheating image is a bone-type cheating trace, a partial image with a length and width of 300*300 can be cropped out based on the position of the bone-type cheating trace as the second training black sample. Thus, the second training black sample contains a complete bone-type cheating trace.

[0171] Correspondingly, a second training white sample can also be constructed in a similar manner. As an example, first, the entire historical cheating game image can be obtained from the database, and then a partial image of a preset size can be cropped and intercepted at the position of the cheating element in the entire historical cheating game image. Different from the construction of the second training black sample above, the partial image in the second training white sample only contains partial cheating elements. Specifically, the same preset size as when constructing the second training black sample can be adopted, but in a way with different coverage areas, to crop and intercept a partial image that only contains partial cheating traces, that is, the cropped partial image does not include a complete cheating element.

[0172] That is to say, partial gray samples can be intercepted from the historical cheating game image, that is, the second training white sample constructed above. The gray sample can be a partial image containing partial cheating traces, or a partial image containing similar cheating traces, such as brightly colored lines with similar box-like cheating traces. It should be noted that the above gray samples can be obtained by means of periodic automatic interception, can be obtained by means of program synthesis, and can also be obtained by extracting cheating elements from a similar cheating template and then synthesizing the extracted cheating elements. Thus, by constructing the second training black sample and correspondingly constructing the second training white sample, the learning ability of the image classification model is enhanced to improve the performance of the image classification model.

[0173] As another example, in order for the image classification model to accurately learn the cheating traces of different marker types, after cropping partial images of different marker types according to the preset size, manual secondary review can be performed to mark the partial images containing obvious and clear cheating traces as the second training black sample, and label the second training black sample, such as labeling the label corresponding to the marker type of the second training black sample, or labeling the partial image as a cheating image. In addition, through the above method, a more comprehensive and effective second training black sample can be obtained, and in the training process of the image classification model, when the magnitude of the second training black sample is several thousand, a better training effect can be obtained. Therefore, when the second training sample is more than 100,000, the loss function of the image classification model can adopt weighted cross-entropy to increase the weight of the second training black sample, so that the image classification model can more accurately learn the feature information of the cheating traces.

[0174] Thus, based on the position of the cheating traces in the historical cheating game images and the preset size, a partial image serving as the second training black sample can be obtained, enabling the image classification model to more accurately learn the features of the cheating traces for the partial image. Additionally, manual secondary review can be performed on the intercepted partial images, ensuring that the image classification model can more clearly and specifically learn the correlation between the labels and marker types in the cheating traces.

[0175] Furthermore, to enable the image classification model to accurately learn the cheating traces rather than the information features in other non-cheating areas of the cheating images, a second training white sample can also be constructed. In one possible implementation, the method of constructing the second training white sample can also include: intercepting a partial image of a preset size from the historical normal game images as the second training white sample.

[0176] It should be noted that the historical normal game images refer to the game images that have been historically determined to have no cheating traces. The historical normal game images can be stored in a database or in the historical records of the server.

[0177] As an example, historical normal game images can be obtained from the database, and then partial images can be cropped and intercepted from the historical normal game images according to the preset size as the second training white sample.

[0178] As another example, to further construct a normal white sample that only differs from the second training black sample in the position of the cheating elements, the method of constructing the second training white sample can also include: replacing the cheating elements in the second training black sample with non-cheating content to obtain the second training white sample.

[0179] See Figure 10 , Figure 10 which is a schematic diagram of constructing the second training white sample provided by the embodiments of this application. First, an image serving as the second training black sample can be obtained. For example, a second training black sample containing bone-type cheating traces. Then, the second training black sample can be input into the target detection model to identify the position of the cheating elements in the second training black sample, that is, the position of the bone-type cheating traces in the image. After that, the cheating elements in the image can be replaced with other normal scenes, that is, the image information at the positions of the cheating elements in the image can be replaced with non-cheating content. Specifically, the cropping position can be randomly selected in the same image or other historical normal game images, and an image of the same size can be cropped and extracted to fill the positions of the cheating elements; or the pixel values near the cheating elements in the same image can also be directly selected to fill the positions of the cheating elements. Finally, the filled image can be used as the second training white sample.

[0180] As an example, based on each second training black sample, in the same way as constructing the second training white sample described above, 2-3 images with only different positions of cheating elements can be generated. Then, these images can be secondarily reviewed manually to avoid the similarity between the filled part in the synthesized image and the cheating elements of other marker types being too high, which may affect the training effect of the image classification model.

[0181] Thus, through the above-mentioned way of constructing the second training white sample, by comparing with the second training black sample, the image classification model can more accurately and meticulously learn the features of cheating traces in the local image during the training process, reduce the interference of other non-cheating information on the training of the image classification model, and improve the processing performance of the image classification model.

[0182] Step 402: Classify the second training sample through the image classification model to be trained, and obtain the training detection result corresponding to the second training sample; the training detection result is used to represent whether the second training sample includes cheating elements.

[0183] After the server obtains the second training sample, it can input the second training sample into the image classification model to be trained. After classifying the input second training sample, the image classification model can output the training detection result corresponding to the second training sample. The training detection result corresponding to the second training sample can be used to represent whether the second training sample includes cheating elements, that is, it can be determined whether the second training sample is a cheating image.

[0184] It should be understood that the model structure in the image classification model here can be a residual structure. Specifically, reference can be made to Figure 5 the model structure shown.

[0185] Step 403: Train the image classification model according to the difference between the label corresponding to the second training sample and the training detection result.

[0186] After obtaining the training detection result corresponding to the second training sample, the difference between the label corresponding to the second training sample and the training detection result can be calculated through a loss function. Thus, the image classification model can be trained through the calculated loss value. That is, with the goal of reducing the loss value, the parameters of the image classification model are continuously adjusted to optimize the performance of the image classification model. The loss function in the training process of the image classification model is not specifically limited in this application.

[0187] It should be understood that in actual applications, a training end condition can be pre-set. When the training of the image classification model reaches the training end condition, the training of the image classification model can be considered to be terminated. The training end condition can be, for example, that the number of training rounds for the image classification model reaches a preset round threshold. Another example can be that the performance of the image classification model is tested and it is found that the performance of the image classification model reaches a preset performance standard (such as reaching a preset detection accuracy, etc.). Another example can be that the performance of the image classification model is tested and it is found that the performance of the image classification model no longer improves significantly with the progress of training. The embodiment of the present application does not perform any training on the training end condition.

[0188] Thus, the final image classification model can be obtained through the above training steps, so that the trained image classification model can perform cheating detection on local images. Based on the image classification model, the probability that the local image is a cheating image can be output, and the probability that the suspicious element of the marker type in the local image is a real cheating trace can be determined, thereby ensuring the accuracy of cheating detection.

[0189] Based on the data processing method provided in the above embodiment, this application also provides a data processing device. Figure 11 Provide explanation. Figure 11 This is a schematic diagram of the structure of a data processing device provided in an embodiment of the present application, the device comprising:

[0190] An acquisition module 1101 is configured to acquire an image to be detected; the image to be detected is a screenshot of a game of a target object;

[0191] The first detection module 1102 is configured to detect whether a suspicious element exists in the image to be detected using a target detection model, and to determine the target position of the suspicious element in the image to be detected and the element type of the suspicious element; the suspicious element is an element that may correspond to a cheating trace, which is visual content generated by a game plug-in to assist cheating;

[0192] The second detection module 1103 is configured to, if the suspicious element exists in the image to be detected, determine an element cheating detection result corresponding to the suspicious element based on a local image corresponding to the target position in the image to be detected using an image processing model corresponding to the element type;

[0193] The cheat detection module 1104 is used to determine whether the target object uses the game cheat according to the element cheat detection result.

[0194] Optionally, the second detection module 1103 includes:

[0195] A text recognition unit, configured to, when the element type is a text type, recognize the target text included in the partial image through an image text recognition model; and determine the element cheating detection result according to the target text;

[0196] An image classification unit, configured to, when the element type is a marker type, perform classification processing on the partial image through an image classification model to obtain the element cheating detection result.

[0197] Optionally, the text recognition unit includes:

[0198] A quantity detection unit, configured to detect the quantity of cheating matching keywords included in the target text; the cheating matching keywords are words that match the preset cheating keywords;

[0199] A first determination unit, configured to determine the element cheating detection result according to the relationship between the quantity of the cheating matching keywords included in the target text and a preset matching word quantity threshold.

[0200] Optionally, there are multiple types of the marker type, and different marker types correspond to different image classification models; the image classification unit includes:

[0201] A first classification processing unit, configured to perform classification processing on the partial image through the image classification model corresponding to the marker type to which the suspicious element belongs to obtain the probability that the partial image belongs to a cheating image;

[0202] A second determination unit, configured to determine the element cheating detection result according to the relationship between the probability that the partial image belongs to a cheating image and a preset probability threshold.

[0203] Optionally, the external plug-in detection module 1104 includes at least one of the following:

[0204] A third determination unit, configured to, when there are multiple suspicious elements in the image to be detected, determine the quantity of cheating elements among the multiple suspicious elements according to the respective element cheating detection results corresponding to the multiple suspicious elements; and determine whether the target object uses the game external plug-in according to the relationship between the quantity of the cheating elements and a preset element quantity threshold;

[0205] A fourth determination unit, configured to, when multiple images to be detected are obtained, determine the quantity of suspected cheating images among the multiple images to be detected according to the element cheating detection results corresponding to the suspicious elements in the multiple images to be detected; and determine whether the target object uses the game external plug-in according to the relationship between the quantity of the suspected cheating images and a preset image quantity threshold.

[0206] Optionally, the device further includes:

[0207] A fifth determination unit, configured to determine, by an overall image classification model, an overall image cheating detection result corresponding to the image to be detected according to the image to be detected;

[0208] The cheating detection module 1104 includes:

[0209] A sixth determination unit, configured to determine whether the target object uses the game cheating software according to the element cheating detection result and the overall image cheating detection result.

[0210] Optionally, the sixth determination unit includes:

[0211] A seventh determination unit, configured to determine whether the target object uses the game cheating software according to the relationship between the probability that the suspicious element characterized by the element cheating detection result is a cheating element and a preset cheating element information threshold, and the relationship between the probability that the image to be detected characterized by the overall image cheating detection result is a cheating image and a preset cheating overall image probability threshold.

[0212] Optionally, the target detection model is trained in the following manner:

[0213] A first acquisition unit, configured to acquire a game screenshot as a first training sample; the first training sample is a first training black sample or a first training white sample, the label corresponding to the first training black sample is used to indicate the position and element type of the cheating element in the first training black sample, and the label corresponding to the first training white sample is used to indicate that the first training white sample does not include a cheating element;

[0214] A first training unit, configured to determine, by the target detection model to be trained, a training detection result corresponding to the first training sample; the training detection result is used to characterize whether the first training sample includes a cheating element, and in the case of including the cheating element, further characterize the predicted position and predicted type of the predicted cheating element in the first training sample;

[0215] A second training unit, configured to train the target detection model according to the difference between the label corresponding to the first training sample and the training detection result.

[0216] Optionally, the first training black sample is obtained by at least one of the following methods:

[0217] A clustering processing unit is configured to perform clustering processing on multiple historical cheating game images to obtain multiple clustering clusters; extract n historical cheating game images from each of the clustering clusters as the first training black samples, where n is an integer greater than or equal to 1; and mark the positions and element types of the cheating elements in the extracted historical cheating game images to obtain the labels corresponding to the first training black samples.

[0218] An adding unit is configured to add template cheating elements to historical normal game images to obtain the first training black samples; and use the adding positions and element types of the template cheating elements as the labels corresponding to the first training black samples.

[0219] Optionally, the device further includes:

[0220] A second obtaining unit is configured to obtain a game screenshot as a first test sample; the first test sample is a first test black sample or a first test white sample, the first test black sample is a cheating game image, and the first test white sample is a normal game image.

[0221] A third training unit is configured to determine a test detection result corresponding to the first test sample through the object detection model.

[0222] An optimization training unit is configured to, if the test detection result corresponding to the first test black sample indicates that the first test black sample does not include cheating elements, use the first test black sample as the first training black sample for optimizing the training of the object detection model, and mark the positions and element types of the cheating elements in the first test black sample as the labels corresponding to the first training black sample; if the test detection result corresponding to the first test white sample indicates that the first test white sample includes cheating elements, use the first test white sample as the first training white sample for optimizing the training of the object detection model.

[0223] Optionally, the image classification model is trained in the following manner:

[0224] A third obtaining unit is configured to obtain second training samples; the second training samples are local images in game screenshots, the second training samples are second training black samples or second training white samples, the labels corresponding to the second training black samples are used to indicate that the second training black samples include cheating elements, and the labels corresponding to the second training white samples are used to indicate that the second training white samples do not include cheating elements.

[0225] A second classification processing unit, configured to classify the second training samples through the image classification model to be trained, and obtain a training detection result corresponding to the second training samples; the training detection result is used to characterize whether the second training samples include cheating elements;

[0226] A fourth training unit, configured to train the image classification model according to the difference between the label corresponding to the second training samples and the training detection result.

[0227] Optionally, the second training black samples are obtained through the following method:

[0228] A first intercepting unit, configured to intercept a local image with a preset size based on the position of the cheating elements in the historical cheating game images as the second training black samples; the second training black samples include the complete cheating elements.

[0229] Optionally, the second training white samples are obtained through at least one of the following methods:

[0230] A second intercepting unit, configured to intercept a local image with a preset size in the historical normal game images as the second training white samples;

[0231] A third intercepting unit, configured to intercept the local image with the preset size based on the position of the cheating elements in the historical cheating game images as the second training white samples; the second training white samples include partial cheating elements;

[0232] A replacement unit, configured to replace the cheating elements in the second training black samples with non-cheating content to obtain the second training white samples.

[0233] An embodiment of the present application further provides a computer device, which may specifically be a terminal device or a server. The terminal device and the server provided by the embodiment of the present application will be introduced from the perspective of hardware implementation below.

[0234] See Figure 12 [[]] Figure 12 is a schematic structural diagram of the terminal device provided by the embodiment of the present application. As Figure 12 shown, for the sake of convenience of description, only parts related to the embodiment of the present application are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present application. The terminal may be any terminal device including a mobile phone, a tablet computer, a personal digital assistant (PDA), a point of sales (POS), an in-vehicle computer, etc. Taking the terminal as a computer as an example:

[0235] Figure 12 ​The block diagram of a part of the structure of a computer related to the terminal provided in the embodiments of the present application is shown. Refer to Figure 12 , the computer includes: a Radio Frequency (RF) circuit 1210, a memory 1220, an input unit 1230 (including a touch panel 1231 and other input devices 1232), a display unit 1240 (including a display panel 1241), a sensor 1250, an audio circuit 1260 (connected with a speaker 1261 and a microphone 1262), a wireless fidelity (WiFi) module 1270, a processor 1280, and a power supply 1290 and other components. Those skilled in the art can understand that Figure 12 the computer structure shown in

[0236] does not constitute a limitation to the computer, and it may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0237] The processor 1280 is the control center of the computer, connecting various parts of the entire computer through various interfaces and lines, and by running or executing the software programs and / or modules stored in the memory 1220, and calling the data stored in the memory 1220, it executes various functions of the computer and processes data. Optionally, the processor 1280 may include one or more processing units; preferably, the processor 1280 may integrate an application processor and a modem processor, where the application processor mainly processes the operating system, user interface, and application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor may not be integrated into the processor 1280 either.

[0238] In the embodiments of the present application, the processor 1280 included in the terminal is used to execute the steps in the data processing methods described in the foregoing respective embodiments.

[0239] See Figure 13 , Figure 13Schematic diagram of a server 1300 provided by an embodiment of the present application. The server 1300 may vary greatly due to configuration or performance differences, and may include one or more central processing units (CPUs) 1322 (e.g., one or more processors) and a memory 1332, and one or more storage media 1330 (e.g., one or more mass storage devices) for storing application programs 1342 or data 1344. Among them, the memory 1332 and the storage media 1330 may be transient storage or persistent storage. The programs stored in the storage media 1330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Further, the central processing unit 1322 may be configured to communicate with the storage media 1330 and execute a series of instruction operations in the storage media 1330 on the server 1300.

[0240] The server 1300 may further include one or more power supplies 1326, one or more wired or wireless network interfaces 1350, one or more input / output interfaces 1358, and / or, one or more operating systems, such as Windows Server TM , Mac OS X TM , Unix TM , Linux TM , FreeBSD TM and so on.

[0241] The steps performed by the server in the above embodiments may be based on the Figure 13 server structure shown.

[0242] Among them, the CPU 1322 is used to execute the steps in the data processing methods described in the foregoing embodiments.

[0243] An embodiment of the present application further provides a computer-readable storage medium for storing a computer program, and the computer program is used to execute the steps in the data processing methods described in the foregoing embodiments.

[0244] An embodiment of the present application further provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the data processing methods described in the foregoing embodiments.

[0245] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0246] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0247] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0248] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0249] If the above-mentioned integrated units are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store computer programs.

[0250] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one (one) of the following" or its similar expression refers to any combination of these items, including any combination of single items (ones) or plural items (ones). For example, at least one (one) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0251] As described above, the above embodiments are only used to illustrate the technical solutions of this application, rather than to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of this application.

Claims

1. A data processing method, characterized in that, The method includes: Obtain an image to be detected; the image to be detected is a game screenshot of a target object; Through a target detection model, detect whether there are suspicious elements in the image to be detected, and determine the target position of the suspicious elements in the image to be detected and the element type of the suspicious elements; the suspicious elements are elements that may correspond to cheating traces, and the cheating traces are visual contents generated by game cheats for assisting in cheating; In the case where the suspicious elements exist in the image to be detected, through the image processing model corresponding to the element type, according to the local image corresponding to the target position in the image to be detected, determine the element cheating detection result corresponding to the suspicious elements; According to the element cheating detection result, determine whether the target object uses the game cheat.

2. The method according to claim 1, characterized in that, The step of through the image processing model corresponding to the element type, according to the local image corresponding to the target position in the image to be detected, determine the element cheating detection result corresponding to the suspicious elements includes: When the element type is a text type, through an image text recognition model, recognize the target text included in the local image; according to the target text, determine the element cheating detection result; When the element type is a marker type, through an image classification model, perform classification processing on the local image to obtain the element cheating detection result.

3. The method according to claim 2, wherein The step of according to the target text, determine the element cheating detection result includes: Detect the number of cheating matching keywords included in the target text; the cheating matching keywords are words that match the preset cheating keywords; According to the relationship between the number of the cheating matching keywords included in the target text and a preset matching word number threshold, determine the element cheating detection result.

4. The method according to claim 2, characterized in that There are multiple types of the marker type, and different marker types correspond to different image classification models; The step of through the image classification model, perform classification processing on the local image to obtain the element cheating detection result includes: Through the image classification model corresponding to the marker type to which the suspicious elements belong, perform classification processing on the local image to obtain the probability that the local image belongs to a cheating image; According to the relationship between the probability that the local image belongs to a cheating image and a preset probability threshold, determine the element cheating detection result.

5. The method according to any one of claims 1 to 4, characterized in that, The step of according to the element cheating detection result, determine whether the target object uses the game cheat includes at least one of the following: In the case where there are multiple suspicious elements in the image to be detected, according to the element cheating detection results corresponding to the multiple suspicious elements respectively, determine the number of cheating elements among the multiple suspicious elements; according to the relationship between the number of the cheating elements and a preset element number threshold, determine whether the target object uses the game cheat; In the case of obtaining multiple images to be detected, determine the number of suspected cheating images among the multiple images to be detected according to the element cheating detection results corresponding to the suspicious elements in the multiple images to be detected; determine whether the target object uses the game cheat according to the relationship between the number of suspected cheating images and a preset image number threshold.

6. The method according to any one of claims 1 to 5, characterized in that The method further includes: Through an overall image classification model, determine the overall image cheating detection result corresponding to the image to be detected according to the image to be detected; The step of determining whether the target object uses the game cheat according to the element cheating detection result includes: Determine whether the target object uses the game cheat according to the element cheating detection result and the overall image cheating detection result.

7. The method according to claim 6, wherein The step of determining whether the target object uses the game cheat according to the element cheating detection result and the overall image cheating detection result includes: Determine whether the target object uses the game cheat according to the relationship between the probability that the suspicious element represented by the element cheating detection result is a cheating element and a preset cheating element information threshold, and the relationship between the probability that the image to be detected represented by the overall image cheating detection result is a cheating image and a preset cheating overall image probability threshold.

8. The method according to any one of claims 1 to 7, characterized in that The target detection model is trained in the following manner: Obtain game screenshots as the first training samples; the first training samples are the first training black samples or the first training white samples, the label corresponding to the first training black sample is used to indicate the position and element type of the cheating element in the first training black sample, and the label corresponding to the first training white sample is used to indicate that the first training white sample does not include a cheating element; Through the target detection model to be trained, determine the training detection result corresponding to the first training sample; the training detection result is used to represent whether the first training sample includes a cheating element, and in the case of including the cheating element, it also represents the predicted position and predicted type of the predicted cheating element in the first training sample; Train the target detection model according to the difference between the label corresponding to the first training sample and the training detection result.

9. The method according to claim 8, wherein The first training black samples are obtained through at least one of the following methods: Perform clustering processing on multiple historical cheating game images to obtain multiple clustering clusters; extract n historical cheating game images from each clustering cluster as the first training black samples, where n is an integer greater than or equal to 1; And mark the position and element type of the cheating element in the extracted historical cheating game images to obtain the label corresponding to the first training black sample; Add template cheating elements to historical normal game images to obtain the first training black samples; Use the adding position and element type of the template cheating element as the label corresponding to the first training black sample.

10. The method according to claim 8, wherein The method further includes: Obtain game screenshots as the first test samples; the first test samples are the first test black samples or the first test white samples, the first test black samples are cheating game images, and the first test white samples are normal game images; Determine the test detection result corresponding to the first test sample through the target detection model; If the test detection result corresponding to the first test black sample indicates that the first test black sample does not include cheating elements, then use the first test black sample as the first training black sample used for optimizing the training of the target detection model, and mark the position and element type of the cheating elements in the first test black sample as the label corresponding to the first training black sample; if the test detection result corresponding to the first test white sample indicates that the first test white sample includes cheating elements, then use the first test white sample as the first training white sample used for optimizing the training of the target detection model.

11. The method according to any one of claims 2 to 4, characterized in that, The image classification model is trained in the following manner: Obtain a second training sample; the second training sample is a local image in a game screenshot, and the second training sample is a second training black sample or a second training white sample. The label corresponding to the second training black sample is used to indicate that the second training black sample includes cheating elements, and the label corresponding to the second training white sample is used to indicate that the second training white sample does not include cheating elements; Perform classification processing on the second training sample through the image classification model to be trained, and obtain the training detection result corresponding to the second training sample; The training detection result is used to characterize whether the second training sample includes cheating elements; Train the image classification model according to the difference between the label corresponding to the second training sample and the training detection result.

12. The method according to claim 11, wherein The second training black sample is obtained in the following manner: Intercept a local image of a preset size based on the position of the cheating elements in the historical cheating game image as the second training black sample; the second training black sample includes the complete cheating elements.

13. The method according to claim 11 or 12, characterized in that, The second training white sample is obtained through at least one of the following methods: Intercept a local image of a preset size in the historical normal game image as the second training white sample; Intercept the local image of the preset size based on the position of the cheating elements in the historical cheating game image as the second training white sample; the second training white sample includes partial cheating elements; Replace the cheating elements in the second training black sample with non-cheating content to obtain the second training white sample.

14. A data processing device, characterized in that, The device includes: An acquisition module, configured to acquire an image to be detected; the image to be detected is a game screenshot of a target object; A first detection module, configured to detect whether there are suspicious elements in the image to be detected through a target detection model, and determine the target position of the suspicious elements in the image to be detected and the element type of the suspicious elements; the suspicious elements are elements that may correspond to cheating traces, and the cheating traces are visual contents generated by game cheats for assisting cheating; A second detection module, configured to, when there are suspicious elements in the image to be detected, determine the element cheating detection result corresponding to the suspicious elements according to the local image corresponding to the target position in the image to be detected through the image processing model corresponding to the element type. An external cheat detection module, configured to determine whether the target object uses the game external cheat according to the element cheat detection result.

15. A computer device, characterized in that, The device includes a processor and a memory; The memory is configured to store a computer program; The processor is configured to execute the data processing method according to any one of claims 1 to 13 based on the computer program.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium is configured to store a computer program, and when the computer program is executed by an electronic device, the data processing method according to any one of claims 1 to 13 is implemented.

17. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, the data processing method according to any one of claims 1 to 13 is implemented.

Images