Object recognition method, apparatus, electronic device, and storage medium

Abstract

Embodiments of the present application provide an object identification method and device, electronic equipment and storage medium, and relate to the fields of artificial intelligence, cloud technology and big data processing. The method comprises: obtaining game data of a to-be-identified object in a target game application; determining parameter values of at least one first game ability evaluation parameter of the to-be-identified object according to the game data; based on the parameter values of the first game ability evaluation parameters, obtaining a preliminary identification result of an object type of the to-be-identified object by using a trained first object identification model; when the preliminary identification result is a target type, determining parameter values of at least one second game ability evaluation parameter of the to-be-identified object based on the game data; and determining whether the to-be-identified object is an object of the target type according to the parameter values of the second game ability evaluation parameters. The method provided in the present application can effectively improve the accuracy of the object identification result.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence, cloud technology, big data, and computer technology. Specifically, this application relates to an object recognition method, apparatus, electronic device, and storage medium. Background Technology

[0002] In recent years, with the rapid development of the Internet and information technology, people are interacting with the Internet in more and more scenarios, and there are more and more interactive applications. Diverse game applications are also emerging one after another. Game players can control the corresponding virtual objects in the game scene by issuing operation commands to obtain a good game experience.

[0003] In some games, especially competitive ones, some players resort to cheating, which severely impacts other players' perception of the game's true nature, leading to a decreased gaming experience and damaging the game's reputation. Therefore, identifying cheaters is a crucial aspect of many game applications. While existing technologies offer some identification methods, their accuracy is not ideal. Improving the accuracy of cheater identification and reducing false positives remains a technical challenge that those skilled in the art have been working to address. Summary of the Invention

[0004] The purpose of this application is to provide an object recognition method, apparatus, electronic device, and storage medium that can effectively improve recognition accuracy. To achieve the above objective, the technical solutions provided by this application are as follows:

[0005] On one hand, embodiments of this application provide an object recognition method, the method comprising:

[0006] Obtain game data of the object to be identified in the target game application;

[0007] Based on the game data mentioned above, determine the parameter value of at least one first game ability evaluation parameter of the object to be identified;

[0008] Based on the parameter values ​​of each first game ability evaluation parameter, the pre-trained first object recognition model predicts the preliminary recognition result of the object type of the object to be recognized. The preliminary recognition result is either the target type or the non-target type. The non-target type includes the first type and the second type.

[0009] When the initial identification result is the target type, based on the game data, determine the parameter value of at least one second game ability evaluation parameter of the object to be identified, wherein the at least one second game ability evaluation parameter is an evaluation parameter used to measure the difference between the target type object and the first type object;

[0010] Based on the parameter values ​​of each of the second game ability assessment parameters, determine whether the object to be identified is an object of the target type.

[0011] On the other hand, embodiments of this application provide an object recognition device, which includes:

[0012] The game data acquisition module is used to acquire game data of the object to be identified in the target game application;

[0013] The first discrimination module is used to determine the parameter value of at least one first game ability evaluation parameter of the object to be identified based on the game data mentioned above. Based on the parameter value of each first game ability evaluation parameter, the module predicts the preliminary identification result of the object type of the object to be identified through the trained first object recognition model. The preliminary identification result is either a target type or a non-target type. The non-target type includes the first type and the second type.

[0014] The second discrimination module is used to determine the parameter value of at least one second game ability evaluation parameter of the object to be identified based on game data when the preliminary identification result is the target type. Based on the parameter value of each second game ability evaluation parameter, it determines whether the object to be identified is an object of the target type. The at least one second game ability evaluation parameter is an evaluation parameter used to measure the difference between the target type object and the first type object.

[0015] Optionally, the aforementioned first object recognition model includes at least two classification models, wherein the at least two classification models are trained on their respective training datasets, and there are at least some different training samples between any two datasets in the at least two training datasets corresponding to the at least two classification models; the aforementioned first discrimination model can be used to: obtain a preliminary prediction result of the object type of the object to be identified by each of the at least two classification models based on the parameter values ​​of each first game ability evaluation parameter; and determine a preliminary recognition result of the object type of the object to be identified based on the at least two preliminary prediction results corresponding to the at least two classification models.

[0016] Optionally, the preliminary prediction results include the probability that the object type of the object to be identified is the target type; the first discrimination module can be used for:

[0017] If there is a probability greater than or equal to the first set threshold among at least two probabilities corresponding to at least two classification models, then the preliminary identification result is determined to be the target type;

[0018] If at least two probabilities corresponding to at least two classification models are both less than the first set threshold, then the at least two probabilities are fused to obtain the fused probability. If the fused probability is greater than or equal to the second set threshold, then the preliminary identification result is determined to be the target type.

[0019] If at least two probabilities are less than the first set threshold and the fusion probability is less than the second set threshold, then the preliminary identification result is determined to be a non-target type.

[0020] Optionally, the first object recognition model includes a first classification model and a second classification model. The first object recognition model can be trained by a model training device in the following way:

[0021] Obtain the first training dataset, which includes multiple labeled training samples. Each training sample includes the parameter values ​​of each first game ability evaluation parameter of a sample object. The label of each training sample represents whether the real object type of the sample object corresponding to the training sample is the target type or a non-target type.

[0022] The first training dataset is divided into a training set and a validation set;

[0023] The initial classification model is trained based on the training set to obtain a third classification model that meets the first preset condition, and the first classification model is obtained based on the third classification model.

[0024] The third classification model is used to predict the object type of each training sample in the validation set.

[0025] A second training dataset is constructed based on the third prediction results and the training samples whose labels do not match in the validation set.

[0026] The initial classification model is trained based on the second training dataset to obtain a second classification model that meets the second preset conditions.

[0027] Optionally, the model training device can be used for:

[0028] The first training dataset is divided into at least three datasets; for each of the at least three datasets, that dataset is used as the validation set, and the datasets other than that dataset in the at least three datasets are used as the training set, thus obtaining a sample combination;

[0029] The initial classification model is trained based on the training set of each sample combination to obtain the third classification model that satisfies the first preset condition for each sample combination.

[0030] Obtain the test set, and use the test samples in the test set to test the performance of the third classification model corresponding to each sample combination, and obtain the model test results corresponding to each sample combination;

[0031] Based on the model test results corresponding to each sample combination, the first classification model is selected from the third classification models corresponding to each sample combination.

[0032] For the validation set of each sample combination, the third prediction result of each training sample in the validation set of that sample combination is predicted using the third classification model corresponding to that sample combination.

[0033] The second training dataset is obtained based on the third prediction results in the validation set of each sample combination and the training samples with mismatched labels.

[0034] Optionally, the second discrimination model can be used to: obtain the parameter threshold corresponding to each second game ability assessment parameter; if the parameter value of each second game ability assessment parameter and its corresponding parameter threshold both satisfy the setting conditions corresponding to each parameter, then determine that the object to be identified is a non-target type object; otherwise, determine that the object to be identified is a target type object.

[0035] Optionally, the threshold values ​​for each of the second game ability evaluation parameters are determined in the following way:

[0036] Obtain multiple sample objects corresponding to the game replay files of the target game application. The multiple sample objects include multiple first objects and multiple second objects. The first objects are objects of the target type, and the second objects are objects of the first type.

[0037] The game replay file corresponding to each sample object is parsed to obtain the game data corresponding to each sample object;

[0038] For each sample object, the parameter values ​​corresponding to each second game ability evaluation parameter are determined based on the game data of that sample object.

[0039] For each second game ability evaluation parameter, a parameter threshold is determined based on the parameter values ​​of multiple first objects corresponding to the evaluation parameter and the parameter values ​​of multiple second objects corresponding to the evaluation parameter.

[0040] Optionally, the second discrimination module can be used to: predict the probability that the object type of the object to be identified belongs to the target type based on the parameter values ​​of each second game ability evaluation parameter and through the trained second object recognition model; determine whether the object to be identified is an object of the target type based on the probability of belonging to the target type; wherein, the second object recognition model is trained based on a third training dataset, which includes multiple first samples and multiple second samples, each first sample being the parameter value of each second game ability evaluation parameter of a target type sample object, and each second sample being the parameter value of each second game ability evaluation parameter of a first type sample object.

[0041] Optionally, the target game application is a shooting game application, and at least one of the above-mentioned second game capability evaluation parameters includes at least one of the following:

[0042] Game ability level assessment parameters;

[0043] The shooting aiming distance variation parameter includes at least one of a first parameter or a second parameter. The parameter value of the first parameter characterizes the fluctuation of the shooting aiming distance during the game, and the parameter value of the second parameter characterizes the sudden change of the shooting aiming distance during the game.

[0044] The number of times a specified action is performed in the game's virtual scene;

[0045] Parameters for evaluating the predictive ability of game situations in virtual game scenarios.

[0046] Optionally, the shooting aiming distance variation parameter includes a first parameter and a second parameter. When the second discrimination module determines the parameter value of at least one second game capability evaluation parameter of the object to be identified based on game data, it can be used for:

[0047] Based on game data, the shooting aiming information in the game virtual scene corresponding to the object to be identified is obtained. The shooting aiming information includes the aiming distance of multiple consecutive aiming actions corresponding to each shooting operation. The aiming distance is the distance between the virtual shooting prop in the game virtual scene and the target being aimed at.

[0048] For each shooting operation, determine the degree of dispersion between the aiming distances of multiple consecutive aiming actions corresponding to that shooting operation, and use the degree of dispersion as the first parameter, and the value of the degree of dispersion as the parameter value of the first parameter.

[0049] For each shooting operation, determine the target distance pair among the aiming distances of the consecutive aiming actions corresponding to that shooting operation. The target distance pair refers to the two aiming distances corresponding to two adjacent aiming actions where the aiming distance decreases.

[0050] For each shooting operation, the sudden change in shooting aiming distance for that shooting operation is determined based on the distance difference between each target distance pair corresponding to that shooting operation;

[0051] Based on the sudden changes in aiming distance during each shooting operation, the parameter value of the second parameter of the target to be identified is determined.

[0052] In another aspect, embodiments of this application also provide an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method provided in any optional embodiment of this application.

[0053] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method provided in any optional embodiment of this application.

[0054] This application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the method provided in any optional embodiment of this application.

[0055] The beneficial effects of the technical solution provided in this application are as follows:

[0056] The object recognition method provided in this application, when determining whether an object to be identified is a target type object, can use the parameter value of a first game ability evaluation parameter used to distinguish between target type and non-target type objects to perform preliminary identification of the object to be identified. This preliminary identification can determine whether the object to be identified might be a target type object. Based on this, the parameter value of a second game ability evaluation parameter used to distinguish between the target type and a first type of non-target type objects can be further used to more accurately determine the final identification result of the object to be identified. Using the method provided in this application embodiment can effectively avoid identifying objects of the first type as target type objects, reducing the false identification rate and improving the accuracy of identification. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0058] Figure 1 A flowchart illustrating an object recognition method provided in an embodiment of this application;

[0059] Figure 2 This is a schematic diagram of a virtual game scene in an embodiment of this application;

[0060] Figure 3 This is a schematic diagram illustrating the object recognition performance based on existing solutions.

[0061] Figure 4 This is a comparative diagram showing the object recognition performance of the method based on existing solutions and the embodiments of this application;

[0062] Figure 5 This is a schematic diagram of the structure of an object recognition system applicable to the embodiments of this application;

[0063] Figure 6 A schematic diagram illustrating the implementation principle of an object recognition method provided in this application embodiment;

[0064] Figure 7 A flowchart illustrating an object recognition method provided in an embodiment of this application;

[0065] Figure 8This is a schematic diagram illustrating the changing trends of aiming information for various types of game players provided in the embodiments of this application;

[0066] Figure 9 This is a schematic diagram of the structure of an object recognition device provided in an embodiment of this application;

[0067] Figure 10 This is a schematic diagram of the structure of an electronic device to which this application applies. Detailed Implementation

[0068] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0069] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” can be implemented as “A,” or as “B,” or as “A and B.” When describing multiple (two or more) items, if the relationship between the multiple items is not explicitly defined, the multiple items can refer to one, several or all of the multiple items. For example, the description of "parameter A includes A1, A2, A3" can be implemented as parameter A includes A1 or A2 or A3, or it can be implemented as parameter A includes at least two of the three items A1, A2 and A3.

[0070] This application addresses the problem of poor accuracy in object type recognition in existing game scenarios by proposing an object recognition method that can effectively improve the accuracy of object recognition.

[0071] Optionally, the solutions provided in this application embodiment can be implemented based on artificial intelligence (AI) technology, specifically involving the training and application of neural network models. For example, based on the training method provided in this application embodiment, a machine learning (ML) approach can be used to train an object recognition model for determining object types.

[0072] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. With the research and advancement of AI technology, it has already been researched and applied in many fields, including smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, drones, robots, smart healthcare, smart customer service, vehicle networking, and intelligent transportation. It is believed that with further technological development, this technology will be applied in even more fields and play an increasingly important role.

[0073] Optionally, the data processing involved in the methods provided in this application embodiment can be implemented based on cloud technology. For example, the training method of the neural network model (such as the initial classification model) provided in this application can be implemented based on cloud technology, and various data calculations involved in the training process (such as the calculation of training loss, the adjustment of model parameters, etc.) can be implemented using cloud computing. Optionally, the storage of each training set used in the training process can also be achieved using cloud storage.

[0074] Optionally, the object recognition method provided in this application embodiment can be implemented as an independent application or a functional module / plugin in an application (such as a target application). By running the independent application or a module / plugin with corresponding functions, a specific type (i.e., target type) of player can be accurately identified from the players of the target game application. Optionally, the target type of player can be a cheating player.

[0075] The method provided in this application embodiment can be executed by any electronic device, such as a terminal device or a server. As an optional approach, the method can be executed by the application server of the target game application (referred to as the game server, such as a device that provides cloud gaming services). The game server can determine whether the game player is a player of the target type based on the game data of the game player (i.e. the object to be identified).

[0076] The application server mentioned above can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The user terminal (also referred to as user equipment) mentioned above can be a smartphone, tablet computer, laptop computer, desktop computer, intelligent voice interaction device (e.g., smart speaker), wearable electronic device (e.g., smartwatch), in-vehicle terminal, smart home appliance (e.g., smart TV), AR / VR device, etc., but is not limited thereto. The terminal and server can be directly or indirectly connected via wired or wireless communication, and this application does not impose any restrictions.

[0077] The objects involved in the embodiments of this application may be players of game applications. In the optional embodiments of this application, various data related to the objects (such as the object's game data, game replay files, etc.) require the object's permission or consent when the embodiments of this application are applied to specific products or technologies. Furthermore, the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. In other words, if the embodiments of this application involve data related to the objects, this data needs to be obtained with the object's authorization and consent, and in accordance with the relevant laws, regulations, and standards of the countries and regions.

[0078] To better illustrate and understand the methods provided in the embodiments of this application, some related technical terms used in the embodiments of this application will be explained below.

[0079] FPS (First Person Shooting Game): As the name suggests, it is a shooting game played from the player's first-person perspective, and it is a branch of action games.

[0080] KDA (Kill, Death, Assist): A term in the game, usually expressed as KDR (KD RATIO) to represent the ratio of kills to deaths.

[0081] Ghost step / ghost jump: A technique used in some games that allows you to walk and jump without making any sound by frequently adjusting the directional keys or crouching and jumping repeatedly.

[0082] Replay: A game replay file is usually a binary file. With the player's authorization and consent, the replay file can be obtained through the recording function built into the game application.

[0083] Data stream parsing: By reverse parsing, the corresponding field data is parsed from the game replay file according to the protocol to obtain the player's game data.

[0084] XGBoost (eXtreme Gradient Boosting): An open-source framework for gradient boosting decision trees. The XGBoost model is a widely used machine learning model.

[0085] Precision: Represents the proportion of samples that are predicted as positive but are actually positive; a commonly used model evaluation metric.

[0086] Recall: The proportion of samples that are actually positive that are predicted to be positive. It is also a commonly used model evaluation metric.

[0087] KMP algorithm: An improved string matching algorithm. Its core idea is to use the information after a failed match to minimize the number of times the pattern string matches the main string in order to achieve fast matching.

[0088] High-level players, also known as advanced players, are those with highly skilled gameplay. For example, in the FPS game genre, these players are at the top of the pyramid. Compared to ordinary players, they have accumulated more experience and technical knowledge during the game, so their gameplay and operations are more professional.

[0089] The following description of several optional implementation methods illustrates the technical solutions of the embodiments of this application and the technical effects produced by these solutions. It should be noted that the following implementation methods can be referenced, borrowed from, or combined with each other. Identical terms, similar features, and similar implementation steps in different implementation methods will not be repeated.

[0090] Figure 1 This illustration shows a flowchart of an object recognition method provided in an embodiment of this application. This method can be executed by a server, such as... Figure 1 As shown, the method includes steps S110 to S150.

[0091] Step S110: Obtain the game data of the object to be identified in the target game application.

[0092] In this embodiment, the target game application can theoretically be any type of game application. It can be a game application that requires the player to download and install it on their user terminal, a cloud gaming application, or a game application within a mini-program, etc. The type of game application is not limited in this embodiment; it can be any type of game application, for example, including but not limited to action, competitive, adventure, simulation, role-playing, and casual games. Optionally, the target game application can be an FPS (First Person Shooting Game). For ease of description, some embodiments in this application will use FPS games as examples to describe the target game application.

[0093] Understandably, in practice, the target game application's corresponding object (i.e., the game player) can be represented by a unique object identifier within the target application (such as the object's nickname, name, login account, or other identifier). The object to be identified can be any player in the target game application. Optionally, the object to be identified can include, but is not limited to, players who have been reported by other players for cheating. The game data of the object to be identified can be the game data of the match in which the player was reported for cheating. In practical applications, the object to be identified can be each object in the set of objects to be processed. This set of objects can include multiple player identifiers that need to be identified by object type, with each player identifier representing an object to be identified. Using the object identification method provided in this application's embodiments, each object in the object set can be identified separately, thereby identifying which objects in the object set are objects of the target type, such as which are cheating game players.

[0094] This application does not limit the specific strategy for classifying object types. The object types that may exist in the target game application can be classified according to the actual application scenario and needs. In this embodiment, the object types may include at least target types and non-target types, where non-target types include at least two types. Optionally, the target type object is the cheating player, and the objects other than the cheating player are non-target type players. Non-target type players include high-level players (e.g., some players ranked high in the target game application) and normal players (or ordinary players). Both ordinary players and high-level players are players who abide by the game rules. High-level players have higher game abilities (such as operational skills, ranking, etc.) than ordinary players. Cheating players are players who participate in the game by using cheating methods, such as using game cheating programs to tamper with the original normal settings and rules of the game, seriously damaging the game's balance and fairness, and affecting the game experience of other players.

[0095] For any player in the target game application, their game data can include data from at least one game played within that application. This data can include parameter values ​​for multiple dimensions of game parameters, reflecting the player's specific gameplay. For example, in an FPS game, game parameters might represent a player's combat ability, such as parameters related to their match performance or their shooting skills (e.g., aiming ability). The content of a player's game data may differ across different game applications. Since game data is the source data reflecting a player's gameplay, object type identification can be performed based on the game data of the object to be identified.

[0096] This application does not limit the specific method for obtaining the game data of the object to be identified. As an optional method, a game replay file of at least one game of the object to be identified can be obtained, and the game data of the object to be identified can be obtained by parsing the game replay file. A game replay file can be understood as a file recording the player's gameplay; the format of game replay files may differ in different game applications. Optionally, the game replay file can be parsed according to the game replay file generation protocol to obtain the corresponding field data, and the content of each field data records the player's specific game data.

[0097] In practical applications, for some game parameters in game applications, the parameter values ​​of these parameters may be recorded at certain time intervals during the player's gameplay. For example, the position of the player (the virtual object controlled by the player, i.e., the player's game character) on the game map is recorded at pre-configured time intervals.

[0098] Step S120: Based on the game data, determine the parameter value of at least one first game ability evaluation parameter of the object to be identified.

[0099] The game data of the object to be identified includes the parameter values ​​of multiple game parameters for that object. Each first game ability evaluation parameter can correspond to at least one game parameter. Each first game ability evaluation parameter can be obtained statistically from the parameters of the at least one game parameter corresponding to that evaluation parameter in the game data. The specific evaluation parameters used for the at least one first game ability evaluation parameter can be selected according to actual application requirements. Optionally, the at least one first game ability evaluation parameter can be some specified game parameters in the game data of the object to be identified, and each specified game parameter can be used as a first game ability evaluation parameter. For a specified game parameter, if the game data of a game contains multiple parameter values ​​for that game parameter, the fused value of the multiple parameter values ​​(such as the sum or mean of the multiple parameter values) can be used as a parameter value for that game parameter.

[0100] In optional embodiments of this application, the first game ability evaluation parameter can be selected from some commonly used basic evaluation indicators of the object to be identified, such as some parameters commonly used in existing player type identification methods to determine player type. This application does not limit which specific first game ability evaluation parameters are used. Optionally, the target game application can be an FPS game, and at least one first game ability evaluation parameter can include parameters from multiple dimensions such as the object's performance in FPS game matches, orientation, and aiming. These parameters may include, but are not limited to, kills, deaths, kills through walls, aiming time through walls, and orientation change rate in the game's virtual scene. These characteristics of different types of players have relatively small individual differences, hence they are called low-variance features.

[0101] To facilitate the description and distinction between the first game ability evaluation parameter in the embodiments of this application and the second game ability evaluation parameter in the following text, the first game ability evaluation parameter is referred to as the low variance feature.

[0102] Step S130: Based on the parameter values ​​of each first game ability evaluation parameter, the preliminary recognition result of the object type of the object to be recognized is predicted by the trained first object recognition model.

[0103] The initial identification result is categorized as either a target type or a non-target type. Non-target types include a first type and a second type. In this embodiment, the target type can be cheaters, the first type can be skilled players, and the second type can be ordinary players. The target type and the first type of non-target type are compared; for example, the target type is a cheater in the game, and the first type is a skilled player. Compared to cheaters and ordinary players, cheaters and skilled players are more similar. During the initial identification, the first and second types belong to the same category, i.e., non-target types. In other words, the first game ability evaluation parameter is a parameter that can be used to distinguish between target and non-target types, and can be some basic and commonly used game ability evaluation indicators obtained through simple statistical analysis of game data. Optionally, the first game ability evaluation parameter can include some commonly used indicators from existing cheater identification schemes.

[0104] The first object recognition model, also known as the first object classification model, predicts whether an object is a target type or a non-target type based on the parameter values ​​of various first game ability evaluation parameters. That is, the first object recognition model can be a binary classification model, with the target type in one class and the non-target type in another; the first type and the second type mentioned above belong to the same class. The specific model architecture of the first object recognition model is not limited in this application embodiment and may include, but is not limited to, classification models based on convolutional neural networks, LR (Logistic regression) models, SVM (Support Vector Machine) classification models, or Naive Bayes models. As an optional approach, the first object recognition model can employ the XGBoost model. The trained first object recognition model can predict the probability that the object to be identified belongs to the target type. Based on this probability, the initial recognition result of the object to be identified can be determined, that is, a preliminary judgment can be made as to whether the object to be identified is a target type or a non-target type.

[0105] The first object recognition model can be trained based on a large number of training samples. These training samples include sample data corresponding to target type objects (parameter values ​​of each first game ability evaluation parameter) and sample data of non-target type objects. When training the model, each sample data is labeled, that is, the true object type label of the sample object. For example, the label of the sample data corresponding to the target type object is 1, and the label of the sample data corresponding to the non-target type object is 0. Based on a large number of labeled sample data, the initial classification model can be continuously trained until an object recognition model that meets the application requirements is obtained.

[0106] To further improve the accuracy of the initial recognition results, in an optional embodiment of this application, the first object recognition model may include at least two classification models, wherein the at least two classification models are trained on their respective training datasets, and there are some different training samples between any two of the at least two training datasets corresponding to the at least two classification models.

[0107] Accordingly, in step S130 above, based on the parameter values ​​of each first game ability evaluation parameter, the pre-trained first object recognition model predicts the preliminary recognition result of the object type of the object to be recognized, including:

[0108] Based on the parameter values ​​of each initial game ability assessment parameter, preliminary prediction results of the object type of the object to be identified are obtained through each of at least two classification models;

[0109] Based on at least two preliminary predictions from at least two classification models, determine the preliminary identification result for the object to be identified.

[0110] For the above at least two classification models, since each classification model is trained on different training datasets, there are sample differences between the classification models. During the training process, the introduction of sample differences between multiple models can make the trained multiple classification models have model differences, and each classification model has its own advantages. By using multiple classification models to identify the object to be identified separately, and fusing the preliminary prediction results corresponding to multiple models to achieve the identification of the object to be identified, the accuracy of the preliminary identification results can be effectively improved.

[0111] For each of the at least two classification models mentioned above, the preliminary prediction result corresponding to that model can characterize the object type of the object to be identified. The form of the preliminary prediction result is not limited in this embodiment and can be configured according to actual needs. For example, the preliminary prediction result may include at least one of the probability that the object to be identified belongs to the target type or the probability that the object to be identified belongs to a non-target type; alternatively, the preliminary prediction result may be the identification result of the object to be identified as the target type or the identification result of the object to be identified as not being the target type.

[0112] After obtaining the preliminary prediction results for each of at least two classification models, the preliminary identification result of the object to be identified can be determined based on multiple preliminary prediction results. The specific strategy for determining the preliminary identification result based on multiple preliminary prediction results can also be configured according to actual needs. For example, if at least a set number of prediction results indicate that the object to be identified is of the target type, the initial identification result is determined to be the target type; otherwise, it is a non-target type. Alternatively, the initial identification result can be determined by fusing various preliminary prediction results. For instance, if the preliminary prediction result is the probability that the object to be identified belongs to the target type, the probabilities corresponding to multiple classification models can be fused (e.g., by calculating the average). The initial identification result is determined based on the fused probability and a preset probability. If the fused probability is greater than the preset probability, the initial identification result is determined to be the target type; otherwise, it is a non-target type.

[0113] As an optional approach, the aforementioned preliminary prediction results include the probability that the object type of the object to be identified is the target type; the aforementioned determination of the preliminary identification result corresponding to the object to be identified based on at least two preliminary prediction results corresponding to at least two classification models includes:

[0114] If there is a probability greater than or equal to the first set threshold among at least two probabilities corresponding to at least two classification models, then the preliminary identification result is determined to be the target type;

[0115] If at least two probabilities corresponding to at least two classification models are both less than the first set threshold, then the at least two probabilities are fused to obtain the fused probability. If the fused probability is greater than or equal to the second set threshold, then the preliminary identification result is determined to be the target type.

[0116] If at least two probabilities are less than the first set threshold and the fusion probability is less than the second set threshold, then the preliminary identification result is determined to be a non-target type.

[0117] In this optional approach, if at least one probability among the multiple probabilities corresponding to multiple classification models is greater than or equal to a first preset threshold, the preliminary identification result is determined to be the target type. When multiple probabilities are all less than the first preset threshold, the preliminary identification result is further determined by fusing the probabilities. The specific method for fusing at least two probabilities is not limited in this embodiment. For example, the fusion method can be averaging, in which the first and second preset thresholds can be the same or different. Alternatively, the fusion method can be multiplication, in which case the product of the at least two probabilities is the fusion probability, and in this method, the first preset threshold is greater than the second preset threshold. The specific values ​​of the first and second preset thresholds are not limited in this embodiment and can be configured according to actual application scenarios and requirements. Optionally, they can be determined based on empirical values ​​and / or experimental values.

[0118] By adopting the optional solution provided in this application, it is possible to avoid missing objects of potential target types in the initial identification, thus ensuring the accuracy of the final identification of target types.

[0119] In this embodiment, the model architectures of the at least two classification models can be the same or different. For example, the model architectures of the at least two classification models can both be XGBoost models, but the model parameters of multiple XGBoost models are different. Alternatively, some classification models are XGBoost models, while others are classification models based on convolutional neural networks. To ensure that the at least two classification models have their own distinct capabilities and advantages, the training samples in the training datasets corresponding to each classification model can be completely non-overlapping or at least partially non-overlapping. This embodiment does not limit the methods for obtaining the training datasets corresponding to each classification model, nor the training methods (such as loss functions or objective functions) corresponding to each classification model.

[0120] As an alternative, the above-mentioned at least two classification models, namely the first object recognition model, may include a first classification model and a second classification model. The first object recognition model can be trained in the following way:

[0121] Obtain a first training dataset, which includes multiple labeled training samples. Each training sample includes the parameter values ​​of each first game ability evaluation parameter of a sample object. The label of each training sample indicates whether the real object type of the sample object corresponding to the training sample is the target type or a non-target type.

[0122] The first training dataset is divided into a training set and a validation set;

[0123] The initial classification model is trained based on the training set to obtain a third classification model that meets the first preset condition, and the first classification model is obtained based on the third classification model.

[0124] The third classification model is used to predict the object type of each training sample in the validation set.

[0125] A second training dataset is constructed based on the third prediction results and the training samples whose labels do not match in the validation set.

[0126] The initial classification model is trained based on the second training dataset to obtain a second classification model that meets the second preset conditions.

[0127] In this optional scheme, the aforementioned at least two classification models can include two classification models: a first classification model and a second classification model. The first classification model is obtained by continuously training the initial classification model using the training set in the first training dataset, and the second classification model is obtained by continuously training the initial classification model using the training set in the second training dataset, which is a subset of the first training dataset. The training samples in the first training dataset are labeled samples (i.e., the first game ability evaluation parameters of sample objects with known object types, i.e., low-variance features). Therefore, supervised machine learning can be used to train the initial classification model to obtain the first and second classification models. Specifically, the input to the initial classification model is the low-variance features corresponding to each sample object, and the output is the prediction result of the object type of the sample object. The prediction result can represent the prediction result of the object type of the sample object. The label of the sample object is known, that is, the true object type of the sample object is known. Therefore, the training loss corresponding to the initial classification model can be calculated based on the prediction result and label corresponding to each training sample. If the training loss is less than the set value or the number of training times reaches the set number (that is, the training termination condition is met, that is, the first preset condition or the second preset condition is met), then the classification model that meets the conditions can be used as the trained classification model.

[0128] The specific method for dividing the first training dataset into a training set and a validation set is not limited in the embodiments of this application. Optionally, there can be one training set and one validation set. For example, the first training set can be divided into two parts, one for training and one for validation. A third classification model can be trained using supervised training based on the training set. This third classification model can be used as the first classification model. After obtaining the first classification model, the model can be sampled to predict the object type corresponding to each training sample in the validation set. That is, each training sample in the validation set (i.e., the low variance features of the sample object) is input into the first classification model to obtain the prediction result of the object type of the sample object corresponding to each training sample. Based on the prediction result of the sample object, the object type of the sample object predicted by the model can be known (for example, the prediction result includes the probability that the sample object is the target type, and based on this probability and a set threshold, it can be determined whether the sample object is the target type or not). Since each training sample knows the type of the real object, for each training sample in the validation set, the object type of the sample object corresponding to the training sample can be judged according to the prediction result and label of the sample (i.e., whether the object type corresponding to the third prediction result matches the real object type). If it is incorrect, the training sample can be used as a sample in the second training dataset. In other words, an auxiliary dataset can be constructed by filtering out samples that the model mispredicted in the validation set from the trained first classification model. This auxiliary dataset can then be used as the second training dataset to train the second classification model. The training samples in the auxiliary dataset can be understood as samples that are difficult to identify. Since the samples in the validation set were not used for training the first classification model, the second training dataset and the training set used to train the first classification model are different sample sets. Therefore, the first classification model and the second classification model are two different classification models.

[0129] To ensure the performance of the selected first classification model as much as possible, this application embodiment also provides a method for obtaining the first classification model and the second training dataset. In this method, the first training dataset is divided into a training set and a validation set, including:

[0130] Divide the first training dataset into at least three datasets;

[0131] For each of at least three datasets, use that dataset as the validation set and use the datasets other than that dataset from the at least three datasets as the training set to obtain a sample combination;

[0132] Corresponding to the partitioning method of the dataset, the initial classification model is trained based on the training set to obtain a third classification model that satisfies the first preset condition. The first classification model is then obtained based on the third classification model, including:

[0133] The initial classification model is trained based on the training set of each sample combination to obtain the third classification model that satisfies the first preset condition for each sample combination.

[0134] Obtain the test set, and use the test samples in the test set to test the performance of the third classification model corresponding to each sample combination, and obtain the model test results corresponding to each sample combination;

[0135] Based on the model test results corresponding to each sample combination, the first classification model is selected from the third classification models corresponding to each sample combination.

[0136] The above-mentioned third-classification model predicts the object type corresponding to each training sample in the validation set, including:

[0137] For the validation set of each sample combination, the third prediction result of each training sample in the validation set of that sample combination is predicted using the third classification model corresponding to that sample combination.

[0138] The second training dataset includes: the third prediction results in the validation set of each sample combination and each training sample with mismatched labels.

[0139] In this optional scheme of the application, multiple sample combinations can be constructed based on the first training dataset. Each sample combination includes a training set and a validation set, and the training sets of each sample combination contain differences in the training samples. Similarly, the validation sets of each sample combination also contain different sets of training samples. Based on this, multiple third-classification models satisfying the first preset condition can be trained using the training sets of multiple sample combinations. Then, the third-classification model with better performance can be selected from these multiple third-classification models as the final first-classification model used to identify the object to be identified. Furthermore, since the validation set also includes the validation sets of various sample combinations, after training the third-classification model corresponding to each sample combination, for each third-classification model, the model can be used to predict the training samples in the validation set of the corresponding sample combination to obtain the prediction results for each training sample. Training samples whose prediction results do not match the labels are then selected and added to the second training dataset for training the second-classification model.

[0140] In this embodiment, after training multiple third-class classification models based on training sets with various sample combinations, the method for testing the performance of each third-class classification model is not limited. The performance of each third-class classification model can be tested using one or more commonly used performance evaluation metrics for classification models, based on test samples in the test set. For example, the evaluation metrics may include at least one of the model's accuracy or recall. For each third-class classification model, each test sample in the test set can be input into the model to obtain the prediction result corresponding to each test sample. Based on the prediction results and labels of each test sample, the accuracy and recall of the model can be calculated. Based on the accuracy and recall of each third-class classification model, the best-performing third-class classification model can be determined as the first classification model. The test set includes sample data (low variance features) of multiple known real object types. This embodiment does not limit the method for obtaining test samples in the test set. Optionally, the test samples can be training samples from the validation set of the various sample combinations mentioned above.

[0141] As another alternative, when obtaining the second training dataset based on the validation set, the first classification model can be used to predict the training samples in the validation set for each sample combination. Based on the prediction results and labels of the samples, samples whose prediction results do not match the labels can be used as samples in the second training dataset.

[0142] After obtaining the first and second classification models trained based on different training samples, the object type of the object to be identified can be initially predicted based on these two classification models. Based on the two initial prediction results corresponding to the two models, the initial identification result of the object to be identified can be obtained.

[0143] Step S140: When the preliminary identification result is the target type, based on the above game data, determine the parameter value of at least one second game ability evaluation parameter of the object to be identified, wherein the at least one second game ability evaluation parameter is an evaluation parameter used to measure the difference in game ability between the target type object and the first type object.

[0144] Understandably, in actual implementation, the step of determining the parameter value of the second game ability evaluation parameter of the object to be identified can also be executed after the game data is obtained, and does not necessarily have to be executed after step S130. For example, the determination of the parameter value of the first game ability evaluation parameter and the parameter value of the second game ability evaluation parameter can be executed in parallel.

[0145] Step S150: Determine whether the object to be identified is an object of the target type based on the parameter values ​​of each second game ability evaluation parameter.

[0146] In this embodiment, at least one second game ability evaluation parameter is an evaluation parameter used to measure the difference between a target type object and a first type object; that is, at least one second game ability evaluation parameter is an indicator used to distinguish between a target type object and a first type object. Because of the difference between a target type object and a first type object, compared to the difference between a target type object and a second type object, the initial identification result may misidentify the first type as the target type. To further improve the accuracy of the identification result, when the initial identification result of the object to be identified is the target type, it can be further determined whether the object to be identified is the target type based on the parameter values ​​of each of the second game ability evaluation parameters of the object to be identified.

[0147] Optionally, the aforementioned secondary game ability assessment parameters can be selected based on experience or derived from statistical analysis of sample data. It is understood that at least one of the aforementioned secondary game ability assessment parameters may differ for different game applications. In game applications, although some game parameters in the game data of target-type players (e.g., cheaters) and first-type players (pro players) are similar, cheaters typically rely on cheating software, resulting in less flexibility in adapting to changing situations compared to real players. For example, pro players flexibly utilize information from the game scenario to make game operations, while cheaters rely on more fixed and mechanical operations. These differences are reflected in the game data, resulting in discrepancies in some fine-grained game behavior data between cheaters and pro players. Therefore, based on the game data of the object to be identified, the parameter values ​​of multiple secondary game ability assessment parameters of the object can be obtained, and these parameter values ​​can be used to further determine whether the object to be identified is a target-type object. Optionally, if the parameter values ​​of multiple secondary game ability assessment parameters determine that the object to be identified is not a target-type object, then the object to be identified can be considered a first-type object.

[0148] As an alternative, at least one of the aforementioned second game ability evaluation parameters can be determined in the following way:

[0149] Obtain game data corresponding to multiple sample objects, where the multiple sample objects include multiple first objects and multiple second objects, where the first objects are objects of the target type and the second objects are objects of the first type;

[0150] For each sample object, based on the game data of that sample object, determine the parameter values ​​of multiple initial game ability evaluation parameters corresponding to that sample object;

[0151] For each initial game ability assessment parameter, based on the parameter values ​​of multiple first objects corresponding to the initial game ability assessment parameter, a first reference parameter value of the target type object corresponding to the initial game ability assessment parameter is determined; based on the parameter values ​​of multiple second objects corresponding to the initial game ability assessment parameter, a second reference parameter value of the first type object corresponding to the initial game ability assessment parameter is determined.

[0152] Based on the difference between the first reference parameter value and the second reference parameter value corresponding to each initial game ability assessment parameter, at least one difference assessment parameter corresponding to the target type object and the first type object is determined from multiple initial game ability assessment parameters;

[0153] Based on the identified at least one difference assessment parameter, at least one second game ability assessment parameter is determined.

[0154] In this alternative approach, at least one second game ability evaluation parameter can be automatically analyzed based on game data of sample objects of the target type and sample objects of the first type. This parameter can be used to measure the difference between the target type and the first type of objects. The initial game ability evaluation parameter can be a commonly used parameter for measuring a player's game ability, such as the player's KDA, the number of times the player performs specific actions (such as designated game skills or moves) during the game, the number of times the player aims before each shooting action, and the distance between the player and the target object in the game scene during aiming.

[0155] For each initial game ability assessment parameter, a first reference parameter value for each target type of player can be determined by statistically analyzing the parameter values ​​corresponding to that parameter across multiple target types of players. This reference parameter value can serve as a baseline value for each target type of player's performance. For example, the average of the reference values ​​for multiple target types of players can be used as the first reference parameter value. Similarly, a second reference parameter value can be determined by statistically analyzing the reference parameter values ​​for multiple first-type players. Furthermore, the magnitude of the difference between the reference parameter values ​​for the two types of players can be used to determine whether the assessment parameter can serve as a parameter to distinguish between target type and first-type players—that is, the aforementioned difference assessment parameter. For example, if the difference between the first and second reference parameter values ​​for a certain initial game ability assessment parameter is significant, such as exceeding a set value (different assessment parameters can correspond to different set values), then this assessment parameter can be used as a difference assessment parameter.

[0156] After determining one or more difference assessment parameters, one or more second game ability assessment parameters can be determined based on these parameters. Optionally, each difference assessment parameter can be used as a separate second game ability assessment parameter, or the various difference assessment parameters can be further transformed to obtain at least one second game ability assessment parameter. For example, the difference assessment parameter is the shooting distance corresponding to the aiming action (such as the distance between the crosshair and the target being aimed at), and the second game ability assessment parameter corresponding to this difference assessment parameter can be the variance or standard deviation of multiple shooting distances in the game data.

[0157] For any object in a game application (such as an object to be identified or a sample object), the game data corresponding to that object can be obtained by acquiring the game replay file of that object in the target game application and parsing the game replay file of each sample object separately.

[0158] After obtaining the various second game ability evaluation parameters that can be used to distinguish between target type and first type objects, if the initial identification result of the object to be identified is target type, then the object to be identified can be further judged based on the parameter values ​​of the various second game ability evaluation parameters of the object to be identified, so as to avoid identifying first type objects as target type objects.

[0159] In an optional embodiment of this application, the target game application can be a shooting game application, such as an FPS game, and at least one of the following second game capability evaluation parameters:

[0160] Game ability level assessment parameters;

[0161] The shooting aiming capability variation parameter includes at least one of a first parameter or a second parameter. The parameter value of the first parameter characterizes the fluctuation of the shooting aiming distance during the game, and the parameter value of the second parameter characterizes the sudden change of the shooting aiming distance during the game.

[0162] The number of times a specified action is performed in the game's virtual scene;

[0163] Parameters for evaluating the predictive ability of game situations in virtual game scenarios.

[0164] For any game player, the game ability level assessment parameter reflects the player's overall game level in one or more games. For shooting games, this assessment parameter may include, but is not limited to, KDA. Based on the player's game data, the player's kills, deaths, and assists in a game can be statistically obtained, and the specific values ​​of kills, deaths, and assists can be used as the parameter values ​​of KDA.

[0165] The specified action can be one or more, and can be configured according to the specific gameplay and experience of the target game application. Usually, there are some specific actions in the game. In games, there are usually some skill-based operations that require players to perform one or more consecutive control actions. Generally, they require a high level of operational skills. High-level players in the game may be able to perform these specified actions because they have a lot of game experience and skill theory. On the other hand, cheaters rely on cheating software and will perform some specific actions less often. Therefore, the number of times the target object performs the specified action can be used as a game ability assessment parameter to distinguish the target type object from the first type object.

[0166] Similarly, while cheaters rely on cheating software, advanced players rely on their genuine experience and skill. Advanced players can predict potential situations in the game and take corresponding actions based on the actual gameplay. Therefore, the type of player can be identified based on their predictive ability regarding game situations. The predictive ability evaluation parameter reflects the player's ability to predict game situations and can be characterized by the values ​​of one or more game parameters related to that ability. For example, it can be determined by the number of times one or more specific actions are performed. Optionally... In shooting games, the predictive ability assessment parameter can be represented by the number of times a player performs a special kill operation during the game or the percentage of special kill operations in the total number of kills. Optionally, special kill operations can include, but are not limited to, kill operations completed with the assistance of teammates. For example, in a game, if a player's kill event is completed when the opponent kills their teammate, that is, the player predicts the location of the opponent (the game character corresponding to the opponent) in the virtual game scene based on the game situation of the opponent killing their teammate, and kills the opponent based on the prediction, this kill event can be recorded as a special kill operation.

[0167] The shooting aiming distance change parameter of a game player can reflect the change of aiming information in the game. For target type and first type of players, the shooting aiming ability of these two types of players is usually not much different. For example, the shooting aiming ability of cheaters and high-level players is stronger than that of ordinary players. Although the shooting aiming ability of target type and first type of players is not much different, the two types of players are different because one is achieved through cheating software. Through analysis, it was found that in the virtual game scene, the overall change of distance between the player's virtual shooting tool (such as the crosshair) and the target is different from that of cheaters and high-level players in the process of aiming and killing a target. During the process of aiming and shooting, the distance change between the crosshair and the target fluctuates more during the cheater's aiming and shooting process, while the distance change of the corresponding high-level player is relatively smooth. Therefore, the second game ability evaluation parameter can include the shooting aiming change parameter. The parameter value of this parameter of an object reflects the change of the aiming distance of the object's shooting operation in the game scene.

[0168] Optionally, the aiming distance variation parameter of the target object can be obtained based on the change in aiming distance corresponding to each shooting operation (target being hit or killed) in the game data of the target object. For example, for the first parameter, if the game data of a player's game includes data related to multiple shooting operations, for each shooting operation, the fluctuation value of the aiming fluctuation of that shooting operation can be determined based on the distance corresponding to the aiming action of a set number of times (that is, the aiming action before that shooting operation). The parameter value of the first parameter can be obtained based on the fluctuation value of the aiming fluctuation corresponding to multiple shooting operations. For example, the mean of the fluctuation value of the aiming fluctuation corresponding to multiple shooting operations can be used as the parameter value. Here, the parameter value of the first parameter reflects the overall shooting aiming distance fluctuation in the entire game process, while the parameter value of the second parameter reflects the distance change of aiming actions in which abrupt changes in aiming distance occur. This can be obtained by statistically analyzing the changes in aiming distance corresponding to the player's shooting operations in the game.

[0169] Optionally, in actual implementation, the aforementioned shooting aiming distance variation parameter may include a first parameter and a second parameter. The parameter value for determining at least one second game capability evaluation parameter of the object to be identified based on its game data may include:

[0170] Based on game data, the shooting aiming information in the game virtual scene corresponding to the object to be identified is obtained. The shooting aiming information includes the aiming distance of multiple consecutive aiming actions corresponding to each shooting operation. The aiming distance is the distance between the virtual shooting prop in the game virtual scene and the target being aimed at.

[0171] For each shooting operation, determine the degree of dispersion between the aiming distances of multiple consecutive aiming actions corresponding to that shooting operation, and use the degree of dispersion as the first parameter, and the value of the degree of dispersion as the parameter value of the first parameter.

[0172] For each shooting operation, determine the target distance pair among the aiming distances of the consecutive aiming actions corresponding to that shooting operation. The target distance pair refers to the two aiming distances corresponding to two adjacent aiming actions where the aiming distance decreases (according to the order of description, if the aiming distance of the i-th aiming action is less than the aiming distance of the (i-1)-th aiming action, then the aiming distances of these two aiming actions are considered as a target distance pair).

[0173] For each shooting operation, the sudden change in shooting aiming distance for that shooting operation is determined based on the distance difference between each target distance pair corresponding to that shooting operation;

[0174] Based on the sudden changes in aiming distance during each shooting operation, the parameter value of the second parameter of the target to be identified is determined.

[0175] The aforementioned dispersion can be represented by at least one of standard deviation or variance. For each shooting operation, the parameter value of the first parameter can be obtained by calculating the specific values ​​of the standard deviation or variance of the aiming distances of multiple aiming actions corresponding to that shooting operation. For the second parameter, it can be obtained by statistically analyzing the differences between the aiming distances that meet the conditions among the multiple aiming actions corresponding to each shooting operation. For example, the sum or mean of the absolute values ​​of the differences between the aiming distances that meet the conditions (i.e., the aforementioned target distance pairs) can be used as the characterization value of the abrupt change in aiming distance for each shooting operation. The parameter value of the second parameter of the object to be identified can be obtained by calculating the sum or mean of the characterization values ​​of the abrupt change in aiming distance for multiple shooting operations in a game.

[0176] Since the second game ability assessment parameters can be used to distinguish between the target type of object and the first type of object, after obtaining the parameter values ​​of each second game ability assessment parameter of the object to be identified, the object type of the object to be identified can be further determined based on these parameter values.

[0177] As an optional approach, determining whether the object to be identified is an object of the target type based on the parameter values ​​of each of the second game ability evaluation parameters may include:

[0178] Obtain the parameter thresholds corresponding to each second game ability evaluation parameter;

[0179] If the parameter values ​​of each second game ability assessment parameter and their corresponding parameter thresholds meet the setting conditions of each parameter, then the object to be identified is determined to be a non-target type object; otherwise, the object to be identified is determined to be a target type object.

[0180] In other words, each second game ability assessment parameter can correspond to its own parameter threshold, and each assessment parameter has its own set conditions. For a second game ability assessment parameter to satisfy the set conditions, the parameter value must be greater than the parameter threshold, or the parameter value must be less than the parameter threshold. Whether the parameter value is greater than or less than the parameter threshold satisfies the set conditions can be determined based on empirical values ​​or experimental statistics. The parameter thresholds corresponding to each second game ability assessment parameter can also be determined by the experimenter or based on experimental statistics.

[0181] In the above optional schemes, the above-mentioned setting conditions can be understood as the discrimination conditions for the first type of object. When all the second game ability evaluation parameters and their corresponding parameter thresholds meet the conditions corresponding to their respective parameters, the object to be identified is determined to be a non-target type object; otherwise, it is a target type object. It is understandable that, in practical applications, the above-mentioned setting conditions can also be set as the discrimination conditions for target type objects.

[0182] As an optional approach, the threshold values ​​for each of the aforementioned second game ability evaluation parameters are determined in the following manner:

[0183] Obtain multiple sample objects corresponding to the game replay files of the target game application. The multiple sample objects include multiple first objects and multiple second objects. The first objects are objects of the target type, and the second objects are objects of the first type.

[0184] The game replay file corresponding to each sample object is parsed to obtain the game data corresponding to each sample object;

[0185] For each sample object, the parameter values ​​corresponding to each second game ability evaluation parameter are determined based on the game data of that sample object.

[0186] For each second game ability evaluation parameter, a parameter threshold is determined based on the parameter values ​​of multiple first objects corresponding to the evaluation parameter and the parameter values ​​of multiple second objects corresponding to the evaluation parameter.

[0187] The optional solution provided in this application can automatically determine the parameter thresholds of each second game ability evaluation parameter based on the parameter values ​​of multiple target type sample objects and multiple first type sample objects corresponding to each second game ability evaluation parameter. The parameter threshold of each second game ability evaluation parameter can be used to determine whether the object to be identified is an object of the target type. For example, for a second game ability evaluation parameter, if statistical analysis reveals that the parameter value of the evaluation parameter for the vast majority of skilled players is greater than A, while the parameter value of the evaluation parameter for the vast majority of cheating players is less than A, then A can be used as the parameter threshold for that evaluation parameter.

[0188] As another alternative, determining whether the object to be identified is an object of the target type based on the parameter values ​​of each of the second game ability evaluation parameters may include:

[0189] Based on the parameter values ​​of each second game ability evaluation parameter, the probability that the object type of the object to be identified belongs to the target type is predicted by the trained second object recognition model.

[0190] Based on the probability of belonging to the target type, determine whether the object to be identified is an object of the target type;

[0191] The second object recognition model is trained based on the second training dataset, which includes multiple first samples and multiple second samples. Each first sample is the parameter value of each second game ability evaluation parameter of a target type sample object, and each second sample is the parameter value of each second game ability evaluation parameter of a first type sample object.

[0192] In this scheme, a second object recognition model can be trained based on sample data (i.e., parameter values ​​of each second game ability evaluation parameter) corresponding to sample objects of the target type and sample objects of the first type. This model can be used to identify whether an object to be identified belongs to the target type. The second object recognition model can be a classification model. During the training phase, the input to the classification model is a feature vector composed of the parameter values ​​of each second game ability evaluation parameter of the sample object. The output can include the probability that the sample object is of the target type. Since the true object type of the sample object is known (i.e., it has a labeled tag), the model training can be constrained by the model prediction results corresponding to each sample in the third training dataset and the labels, until a second object classification model that meets the preset conditions is obtained. After obtaining the trained model, the parameter values ​​of each second game ability evaluation parameter of the object to be identified can be input into the model, and the model's output can be used to determine whether the object to be identified belongs to the target type.

[0193] Optionally, in order to ensure better recognition results, the scheme of determining the final recognition result of the object to be identified based on the second object classification model can be implemented when the number of the first and second samples in the third training dataset reaches a certain number, so that the trained second object recognition model has better model performance and the recognition result is more accurate.

[0194] To better illustrate and understand the solutions provided in this application and their practical value, the method provided in this application will be described below with reference to an example of an application scenario. In this application scenario, the target game application is an FPS game as an example. In this game, FPS objects can be divided into three categories: cheating players (target type), skilled players, and ordinary players. Among them, skilled players and ordinary players are non-cheating players, i.e., players who are not the target type.

[0195] The object to be identified can be any player in the game application. Optionally, the object to be identified can be every object in a set of objects to be identified. This set of objects may include, but is not limited to, players who have been reported by other players for cheating. Each player in the game application can be represented by a player identifier in the game application. The player identifier may include, but is not limited to, the player name used by the player in the game application (i.e., the player's ID used in the game, which can be called RaildID), and may also include the player's in-game ID (such as the player number in a game). The player identifier can uniquely identify a player.

[0196] Based on the method provided in this application embodiment, the game server can identify each player in the set of objects to be identified, thereby determining all cheating players in the set. It can also send corresponding prompts to the identified cheating players, such as prompting them to abide by the game rules and not to cheat. Optionally, it can also implement corresponding game permission controls on cheating players according to the game rules, such as banning their accounts. It can also issue corresponding processing notifications to all players.

[0197] In FPS games, players can experience shooting from their own perspective, which greatly enhances the game's agency and realism. For example... Figure 2The diagram shown illustrates an FPS game interface, allowing players to embody a virtual character (the virtual object shown) within a virtual game environment, providing an immersive gaming experience. Every game has its own rules, and players should play fairly while adhering to these rules. However, with technological advancements, cheating methods in games have become increasingly sophisticated and diverse. To achieve better results, some players sometimes resort to cheating, leading to unfairness and negatively impacting other players' experience and the game's reputation. Therefore, identifying cheaters in games is crucial.

[0198] Currently, players can report potential cheaters through a reporting system. However, because it's generally difficult for players to distinguish between skilled players and cheaters in-game, skilled players are frequently reported. Analysis of game data from some skilled players reveals that many reported cheaters are simply misidentified due to their high skill level. Existing technologies also include methods to identify cheaters based on game parameters. For example, a player identification model is trained based on a large number of players' basic game parameters to predict the probability of a player being a cheater, and then a determination is made based on the probability and a penalty threshold. However, due to the complexity and variety of cheating methods in games, the accuracy of existing models is poor, leading to misjudgments of both ordinary and skilled players. To reduce misjudgments, current technologies often involve simply adjusting the penalty threshold, but this results in a large number of missed cheaters.

[0199] Figure 3 This diagram illustrates the effect of cheating player identification in the prior art. In the diagram, the ellipse corresponding to cheating players represents the actual number of cheating players, the ellipse corresponding to skilled players represents the actual number of skilled players, and the ellipse corresponding to penalized players represents the prediction result and penalty threshold based on the player identification model of the prior art. Figure 3 The left-hand diagram corresponds to the initial penalty threshold. Figure 3 The right-hand chart shows the number of cheating players determined by the adjusted penalty threshold. Figure 3 Left side image and Figure 3 As shown in the right figure, the existing technology will identify a considerable proportion of skilled players and ordinary players as cheaters. Even if some false positives can be reduced by adjusting the penalty threshold, it will still cause many cheaters to be missed.

[0200] The object recognition method provided in this application can effectively improve the above-mentioned problems in the prior art. Figure 4The diagram illustrates the identification effectiveness of existing technologies and the methods provided in the embodiments of the present invention. A comparison shows that the solution provided in the embodiments of the present invention can reduce false positives without affecting the coverage of cheating players. Figure 4 The two rectangular areas for misjudgment correction shown in the right figure schematically illustrate the proportion of misjudgments reduced by the solution provided in this application embodiment. It can be seen that it can reduce misjudgments not only by expert players, but also by ordinary players.

[0201] The following describes in detail the method provided in the embodiments of this application and the aforementioned technical effects, taking an FPS game scenario as an example.

[0202] Figure 5 This diagram illustrates the structure of an object recognition system applicable to this scenario embodiment. Figure 5 As shown, the information recommendation system may include a training server 10, a game server 20, and terminal devices (terminal device 1, terminal device 2, and terminal device 3 schematically shown in the figure). The training server 10 can be used to perform the training operations of the neural network involved in this embodiment, such as performing the training step of training a first object recognition model based on a first training dataset. After the trained object recognition model is obtained through the training server 10, the model can be deployed to the game server 20. Players can play the game through their terminal devices. The game server 20 can determine whether a player is cheating based on the player's game data, combined with the object recognition model and corresponding discrimination rules. In actual implementation, with the player's authorization, the game process can be recorded during gameplay, and a corresponding game replay file can be generated according to the corresponding protocol. This replay file can be saved in the database 21. When player identification is needed, the corresponding game replay file can be retrieved from the database 21, and the player's game data can be obtained based on this file. The game replay file records various relevant data of each player during the game process.

[0203] Figure 6 The diagram illustrates the implementation principle of the object recognition method in this scenario embodiment. Figure 7 The implementation flowchart of this method is shown below. Figures 5 to 7 The object recognition process in this scenario embodiment is described below. Figure 6The XGBoost models A and B shown are the first and second classification models, respectively, used for preliminary identification of the object type of the target. For any player, the low-variance features include the parameter values ​​of each of the player's first game ability evaluation parameters, and the high-variance features include the parameter values ​​of each of the player's second game ability evaluation parameters. The low-variance features can be used to initially distinguish between cheating players and normal players (both skilled and ordinary players are considered normal players, i.e., non-cheaters). The high-variance features are used to further determine whether a player is truly cheating. Based on the high-variance features, skilled players who are mistakenly identified as cheaters in the initial identification can be filtered out, preventing skilled players from being identified as cheaters. Figure 7 As shown, the object recognition method in this scenario may include the following steps:

[0204] Step S11: Obtain the first training dataset.

[0205] Step S12: Train the first classification model and the second classification model based on the first training dataset.

[0206] The first training dataset is the training sample set used to train Xgboost models A and B that meet the conditions. The first training dataset includes a large number of training samples, each of which includes a low-variance feature of a sample player (i.e., a sample object), and the label of the sample is known, representing whether the sample player is a cheater or a normal player.

[0207] Optionally, to differentiate between legitimate and cheating players in FPS games, low-variance features can be constructed based on three dimensions: player performance, orientation, and aiming. These low-variance features may include, but are not limited to, kill count, death count, wallhack kills, wallhack aiming time, and orientation change rate in the game's virtual environment. Each of these parameters can serve as a primary game ability assessment parameter, and their specific values ​​are the parameter values ​​for this primary game ability assessment. These parameter values ​​can be combined to obtain the low-variance features. The individual differences between these features of various players in FPS games are usually small, hence the term "low-variance features."

[0208] By acquiring and parsing game replay files of multiple known types of sample objects, game data for each sample object can be obtained. Based on the parsed game data, the parameter values ​​of each parameter in the aforementioned low-variance features can be derived, thus obtaining the training samples corresponding to each sample object. Subsequently, the initial classification model can be trained using a large number of training samples to obtain a trained first classification model and a second classification model.

[0209] As an alternative, five-fold cross-validation can be used during the training of the classification model, such as... Figure 6 As shown, the first training dataset is divided into five equal parts (other division methods and numbers can also be used). Four parts from each fold are selected as the training set to train the XgBoost model. This model is then used to make predictions on the remaining validation set. Based on the true labels of the validation set data, samples that were incorrectly predicted by the model are selected to construct an auxiliary dataset (i.e., the second training dataset). By using each of the five equal parts as the validation set and the other four parts as the training set, five sample combinations can be obtained. The initial XgBoost model is continuously trained using the training set of each sample combination to obtain five trained intermediate models. These five intermediate models can be tested separately, and the model with the best performance is selected as XgBoost model A. In addition, an auxiliary dataset is obtained, which contains the incorrectly predicted samples from the validation set in each fold of training.

[0210] After obtaining the auxiliary dataset, the initial XGBoost model can be continuously trained using the auxiliary dataset to obtain the trained model as XGBoost model B. Since XGBoost model B and XGBoost model A are trained based on different training samples, there are model differences between the two classification models. The above-mentioned scheme for training two classification models provided in this application first uses the full training data to train classification model A, and then uses the samples that model A mispredicted as auxiliary samples to train model B. Because the data used during training is different, model A and model B are different. This scheme introduces model differences through sample differences, thereby improving the accuracy of the initial recognition results by fusing the two classification models.

[0211] After obtaining two trained classification models, these models can be deployed to the game server 20. The game server 20 can sample these two models to perform preliminary identification of the objects to be identified, and can further identify them based on the high variance features of the objects. Figure 7 As shown, the steps for the game server 20 to identify the object to be identified may include the following steps S21 to S26.

[0212] Step 21: Obtain the game replay file of the object to be identified.

[0213] Step 22: Parse the game replay file of the object to be identified to obtain the game data of the object to be identified.

[0214] As an alternative, in practical applications, with player authorization, game replay files for each player can be obtained through the game's built-in recording function. The corresponding field data, such as firing protocols, movement protocols, and aiming protocols, can be extracted from the replay files by reverse engineering according to the protocols. Below is an example of the format of some of these protocols:

[0215] Time | PlayerScreen | My RaildID | Enemy RailID | Visibility | My Coordinates | My Facing Direction;

[0216] Among them, the above time is the generation timestamp of a record, PlayerScreen represents the content displayed in the current player's user interface, My RaildID represents the unique identifier of the current player, Enemy RailID represents the unique identifier of the opponent in the game, Visibility indicates whether the current player's game character can see the opponent's game character in the game virtual scene, and My Coordinates represents the location information of the current player's game character in the game map.

[0217] By parsing the game replay file of the object to be identified, the game data of that object in the game can be obtained.

[0218] Step 23: Based on the game data obtained from the analysis, obtain the low variance features of the object to be identified.

[0219] Step 24: Based on low variance features, predict the preliminary identification results of the object to be identified using the first classification model and the second classification model.

[0220] Based on the parsed game data, low-variance features of the target object can be obtained, such as the number of kills (a parameter value of the first game ability evaluation index) and the number of kills achieved by passing through walls. After obtaining the low-variance features of the target object from the game data, such as... Figure 6 As shown, inputting this low-variance feature into Xgboost model A yields the prediction result output by model A, which includes the probability that the target is a cheater. Inputting this low-variance feature into Xgboost model B yields the prediction result output by model B. Then, the judgment results (predictions) from model A and model B can be fused. Figure 6 The two-sided recall prediction value correction shown in the figure yields the preliminary identification result of the object to be identified (low variance feature prediction result). Optionally, the fusion method can be: if neither model is certain whether a player is cheating, then the probabilities are multiplied; if either model determines that the player is cheating, then the player is judged to be cheating, which can be expressed by the following formula:

[0221]

[0222] in, This indicates the preliminary identification result. A result of 1 indicates that the player to be identified is a cheater, and a result of 0 indicates that the player to be identified is a non-cheating player. The prediction result of model A, The prediction result of model B, The model is designed to determine the threshold for player cheating (i.e., the penalty threshold, which is the first set threshold). The threshold for determining player cheating is the result of multiplying the probabilities of the two models (the second set threshold). < The specific value can be determined based on the actual situation. As can be seen from the formula, if... or Greater than or equal to ,or Greater than or equal to If the initial identification result is the target type (cheating player), then the initial identification result is the non-target type (normal player).

[0223] Step 25: If the preliminary identification result is the target type, then obtain the high variance features of the object to be identified based on the game data.

[0224] Step 26: Based on high variance features and discrimination rules, determine the final identification result of the object to be identified.

[0225] When the initial identification result is a cheater, since the initial judgment may mistakenly identify skilled players as cheaters, in order to reduce false positives, when the initial identification result is confirmed as a cheater, the target can be further confirmed using high-variance features. The high-variance features include the parameter values ​​of at least one second game ability evaluation parameter. High-variance features can be used to distinguish between cheaters and skilled players among normal players. The discrimination rule can be a rule used to determine whether a player is a skilled player; players who meet the rule are considered skilled players. Of course, the discrimination rule can also be a rule used to determine whether a player is a cheater; players who meet the rule are considered cheaters. In this scenario embodiment, the discrimination rule is a skilled player discrimination rule, and the high-variance features can also be called skilled player behavior features. Optionally, in this scenario embodiment, the high-variance features may include, but are not limited to, at least one of the following:

[0226] KDA (Game Ability Level Assessment Parameter); Aiming Fluctuation Parameter (First Parameter) and Aiming Mutation Parameter (Second Parameter); Number of Ghost Step, Ghost Jump, etc. (Number of times a specified action is performed in the game's virtual scene); Special Kills to Total Kills (Predictive Ability Assessment Parameter).

[0227] By parsing the game replay files of the object to be identified, game data can be obtained, which provides fine-grained behavioral data of the object during the game. Based on this game data, the feature values ​​of the aforementioned characteristics of the object to be identified can be obtained, i.e., the parameter values ​​of each of the second game ability evaluation parameters. The following is a description of each of the high-variance features.

[0228] Statistical analysis of game data from skilled and cheating players in FPS games reveals the following key characteristics in the gameplay of skilled FPS players:

[0229] (1) Good awareness, able to pre-aim and fire at high-frequency locations (i.e., anticipate the appearance of enemies in a certain place in the virtual game scene and fire in advance), which is also one of the main reasons for being misjudged; (2) Good combat record, high KDA; (3) Smooth recoil control and aiming during kills; (4) Achieve the effect of walking silently through operations such as ghost step and ghost jump; (5) Strong ability to obtain information in the game. The features extracted based on these characteristics have large differences between individual features, so they can be called high variance features.

[0230] Regarding the aforementioned characteristic (1), since skilled players may be mistakenly identified as cheaters during the initial identification process, and the initial identification results have already demonstrated such characteristics in the behavior of skilled players, it is not necessary to extract additional features for this characteristic.

[0231] For the above feature (2), the player's kills, deaths, assists, etc. can be counted as features through the relevant protocol information of the Replay data stream (i.e., game replay file).

[0232] Regarding the above feature (3), according to data statistics, the average time from a player seeing a target to killing it is generally around 2 seconds. The aiming information in the replay is usually synchronized once at a predetermined time interval (such as 50ms). Therefore, the recoil control and aiming during the killing process can be characterized by the aiming information (the change in the distance between the crosshair of the virtual shooting tool and the target) of the set number of times (such as 50 times) before the player kills the target.

[0233] As an example, Figure 8 This diagram illustrates the changes in aiming information for different types of players in an FPS game. Wallhacks and aimbots correspond to cheaters. The horizontal axis represents the number of aiming attempts, and the vertical axis represents the distance between the crosshair and the target (distance in the game's virtual scene). Data visualization clearly shows the differences in aiming information among the four types of players: normal players have a small range of distance variation, but lack smoothness; wallhack players have a large range of distance variation; and aimbot players experience significant distance fluctuations. Based on these differences, this embodiment of the application designs two features—aiming fluctuation and aiming abrupt change—to characterize these situations.

[0234] Optionally, the aiming fluctuation of each kill (i.e., shooting action) can be measured using the standard deviation. This can be calculated based on aiming information from a set number of kills prior to the current kill (e.g., 50 kills). For example, with a set number of kills of 50, the aiming fluctuation of a single kill action... It can be calculated using the following formula:

[0235]

[0236] in, Indicates the number of kills before the kill. i The distance corresponding to each aiming action. The above formula can be used to calculate the aiming fluctuation value corresponding to each kill action in a game. The average of the aiming fluctuation values ​​corresponding to each kill action can be used as the index value of the player's aiming fluctuation in that game (i.e., the fluctuation of shooting aiming distance).

[0237] For each kill, the aiming abrupt change measures the player's aiming behavior during the game. Optionally, it can be calculated based on the change in aiming distance that results in a significant change, or it can be measured by the negative slope of the aiming distance change. The calculation formula is as follows:

[0238]

[0239] in, This represents the aiming mutation value corresponding to a single kill. As the formula shows, the aiming mutation statistics represent the change in aiming distance over the 50 aiming attempts prior to the kill event. The 50th aiming attempt is the sum of the differences in aiming distance between two consecutive kills. This formula can be used to calculate the aiming mutation value corresponding to each kill in a game. The average of these kill-related aiming mutation values ​​can be used as the index value of the player's aiming mutation in that game (i.e., the change in shooting aiming distance). It can be understood that the 50th aiming attempt refers to the aiming attempt closest to the kill event.

[0240] Regarding the aforementioned characteristic (4), the action sequences of ghost steps and ghost jumps in the game (i.e., the standard action sequence of ghost step operations (also known as the reference action sequence) and the standard action sequence of ghost jump operations) can be obtained through analysis of the action protocol information in the Replay data stream. For the player to be identified, the action sequence of the player's entire game can be extracted based on the player's game data. By matching the standard action sequence with the player's action sequence of the entire game, the number of times the player performs ghost step operations and the number of times they perform ghost jump operations can be determined. Optionally, each action in the player's action sequence can be regarded as a character, and the standard action sequence can be regarded as a word. The KMP algorithm can be used to match and count ghost step and ghost jump behaviors.

[0241] Regarding the aforementioned characteristic (5), this characteristic is based on the following assumption: skilled players can perceive enemy positions through in-game information, while cheaters can directly see enemy positions from a god-like perspective. Therefore, kills performed by players with the assistance of teammates can be considered special kills, and the number of special kills can be used as the parameter value for evaluating player prediction ability. An optional implementation method is to maintain a dictionary to record the latest faction to which each player belongs, and maintain a list where the i-th position in the list represents the player who recently killed the i-th team member. Each time a kill event occurs, it is determined whether the kill is against the player who recently killed their teammate; if so, one point is awarded for a special kill. By recording the total number of kills and the number of special kills in each game in the game data, the total kill score for each player and the ratio of such special kills to the total kills can be calculated.

[0242] Based on the above characteristics, when initially determining that the target is a cheater, the values ​​of various parameters such as the KDA index (e.g., number of kills in the game), aiming fluctuation index, aiming mutation index, number of ghost step operations, number of ghost jump operations, and the ratio of special kills to total kills can be determined based on the target's game data. These values ​​are used as the target's high variance features to determine whether the target is indeed a cheater. Specifically, the target's high variance features can be compared with the discrimination rules of skilled players. If the discrimination rules are met, the target can be determined not to be a cheater.

[0243] To determine the criteria for identifying expert players, statistical analysis can be performed based on sample game data from both cheaters and expert players (for example, data analysis can be conducted using 50 samples of expert players and 50 samples of cheaters, i.e.,...). Figure 6The data distribution analysis shown above, through sample data analysis, can determine the parameter thresholds for identifying high-level players based on the values ​​of the aforementioned parameters for both skilled players and cheaters, thus obtaining the identification rules for skilled players. Optionally, the identification rules, i.e., the filtering rules, for skilled players can be as follows:

[0244]

[0245] in, The result is the prediction result for high-variance features, which is the final identification result. A result of 1 indicates that the object to be identified is a skilled player, and a result of 0 indicates that the object to be identified is a cheating player. To kill a number of people, In order to target fluctuations, To target mutations, b represents the number of ghost steps, t represents the number of ghost jumps, and r represents the percentage of special kills in the total kills. These correspond to the thresholds for the above items, and the specific values ​​can be determined based on the actual situation or experimental values.

[0246] When a player is predicted to be a cheater based on their low variance features, the high variance features of that player can be used to obtain their high variance feature prediction results through the expert player discrimination rule (rule filtering). This achieves the filtering of expert players and avoids misidentifying them as cheaters.

[0247] In summary, the final object recognition result in this scenario embodiment is... ( Figure 6 The detection results (in the data) can be represented as follows:

[0248]

[0249] in, A value of 1 indicates that the player is cheating, and 0 indicates that the player is not cheating. In other words, a player is determined to be cheating only if both the judgment result corresponding to the low variance feature and the cheating result corresponding to the high variance feature are cheating.

[0250] The object recognition method provided in this application introduces model differences through sample differences during classification model training, thereby correcting the model's predicted values. Compared with existing methods, the false positive rate can be reduced by 40% under the premise of the same model coverage. Furthermore, this application also analyzes the behavior patterns of high-level players in the game by focusing on fine-grained in-game behavior dimensions, extracting their behavioral features, and identifying high-level players from the cheating players initially predicted by the model, achieving an accuracy rate exceeding 90%. In practical applications, the method provided in this application can identify and accumulate high-level player data. This data can be used to guide game players through official announcements in game applications. Tests on specific game applications have shown that after adopting the solution provided in this application, the cheating complaint rate of game applications decreased by half. This effectively protects the interests of normal players while cracking down on cheaters, thus safeguarding the game's security reputation.

[0251] Corresponding to the object recognition method provided in the embodiments of this application, the embodiments of this application also provide an object recognition device. Optionally, the object recognition device can be any electronic device, such as a server, etc. Figure 9 As shown, the object recognition device 100 may include a game data acquisition module 110, a first discrimination module 120, and a second discrimination module 130. Among them,

[0252] The game data acquisition module 110 is used to acquire game data of the object to be identified in the target game application;

[0253] The first discrimination module 120 is used to determine the parameter value of at least one first game ability evaluation parameter of the object to be identified based on game data, and predict the preliminary identification result of the object type of the object to be identified by the trained first object recognition model based on the parameter value of each first game ability evaluation parameter. The preliminary identification result is either a target type or a non-target type, and the non-target type includes the first type and the second type.

[0254] The second discrimination module 130 is used to determine the parameter value of at least one second game ability evaluation parameter of the object to be identified based on game data when the preliminary identification result is the target type. The at least one second game ability evaluation parameter is an evaluation parameter used to measure the difference between the target type object and the first type object. Based on the parameter value of each second game ability evaluation parameter, it is determined whether the object to be identified is the target type object.

[0255] Optionally, the aforementioned first object recognition model includes at least two classification models, wherein the at least two classification models are trained on their respective training datasets, and there are at least some different training samples between any two datasets in the at least two training datasets corresponding to the at least two classification models; the aforementioned first discrimination model can be used to: obtain a preliminary prediction result of the object type of the object to be identified by each of the at least two classification models based on the parameter values ​​of each first game ability evaluation parameter; and determine a preliminary recognition result of the object type of the object to be identified based on the at least two preliminary prediction results corresponding to the at least two classification models.

[0256] Optionally, the preliminary prediction results include the probability that the object type of the object to be identified is the target type; the first discrimination module can be used for:

[0257] If there is a probability greater than or equal to the first set threshold among at least two probabilities corresponding to at least two classification models, then the preliminary identification result is determined to be the target type;

[0258] If at least two probabilities corresponding to at least two classification models are both less than the first set threshold, then the at least two probabilities are fused to obtain the fused probability. If the fused probability is greater than or equal to the second set threshold, then the preliminary identification result is determined to be the target type.

[0259] If at least two probabilities are less than the first set threshold and the fusion probability is less than the second set threshold, then the preliminary identification result is determined to be a non-target type.

[0260] Optionally, the first object recognition model includes a first classification model and a second classification model. The first object recognition model can be trained by a model training device in the following way:

[0261] Obtain the first training dataset, which includes multiple labeled training samples. Each training sample includes the parameter values ​​of each first game ability evaluation parameter of a sample object. The label of each training sample represents whether the real object type of the sample object corresponding to the training sample is the target type or a non-target type.

[0262] The first training dataset is divided into a training set and a validation set;

[0263] The initial classification model is trained based on the training set to obtain a third classification model that meets the first preset condition, and the first classification model is obtained based on the third classification model.

[0264] The third classification model is used to predict the object type of each training sample in the validation set.

[0265] A second training dataset is constructed based on the third prediction results and the training samples whose labels do not match in the validation set.

[0266] The initial classification model is trained based on the second training dataset to obtain a second classification model that meets the second preset conditions.

[0267] Optionally, the model training device can be used for:

[0268] The first training dataset is divided into at least three datasets; for each of the at least three datasets, that dataset is used as the validation set, and the datasets other than that dataset in the at least three datasets are used as the training set, thus obtaining a sample combination;

[0269] The initial classification model is trained based on the training set of each sample combination to obtain the third classification model that satisfies the first preset condition for each sample combination.

[0270] Obtain the test set, and use the test samples in the test set to test the performance of the third classification model corresponding to each sample combination, and obtain the model test results corresponding to each sample combination;

[0271] Based on the model test results corresponding to each sample combination, the first classification model is selected from the third classification models corresponding to each sample combination.

[0272] For the validation set of each sample combination, the third prediction result of each training sample in the validation set of that sample combination is predicted using the third classification model corresponding to that sample combination.

[0273] The second training dataset is obtained based on the third prediction results in the validation set of each sample combination and the training samples with mismatched labels.

[0274] Optionally, the second discrimination model can be used to: obtain the parameter threshold corresponding to each second game ability assessment parameter; if the parameter value of each second game ability assessment parameter and its corresponding parameter threshold both satisfy the setting conditions corresponding to each parameter, then determine that the object to be identified is a non-target type object; otherwise, determine that the object to be identified is a target type object.

[0275] Optionally, the threshold values ​​for each of the second game ability evaluation parameters are determined in the following way:

[0276] Obtain multiple sample objects corresponding to the game replay files of the target game application. The multiple sample objects include multiple first objects and multiple second objects. The first objects are objects of the target type, and the second objects are objects of the first type.

[0277] The game replay file corresponding to each sample object is parsed to obtain the game data corresponding to each sample object;

[0278] For each sample object, the parameter values ​​corresponding to each second game ability evaluation parameter are determined based on the game data of that sample object.

[0279] For each second game ability evaluation parameter, a parameter threshold is determined based on the parameter values ​​of multiple first objects corresponding to the evaluation parameter and the parameter values ​​of multiple second objects corresponding to the evaluation parameter.

[0280] Optionally, the second discrimination module can be used to: predict the probability that the object type of the object to be identified belongs to the target type based on the parameter values ​​of each second game ability evaluation parameter and through the trained second object recognition model; determine whether the object to be identified is an object of the target type based on the probability of belonging to the target type; wherein, the second object recognition model is trained based on a third training dataset, which includes multiple first samples and multiple second samples, each first sample being the parameter value of each second game ability evaluation parameter of a target type sample object, and each second sample being the parameter value of each second game ability evaluation parameter of a first type sample object.

[0281] Optionally, the target game application is a shooting game application, and at least one of the above-mentioned second game capability evaluation parameters includes at least one of the following:

[0282] Game ability level assessment parameters;

[0283] The shooting aiming distance variation parameter includes at least one of a first parameter or a second parameter. The parameter value of the first parameter characterizes the fluctuation of the shooting aiming distance during the game, and the parameter value of the second parameter characterizes the sudden change of the shooting aiming distance during the game.

[0284] The number of times a specified action is performed in the game's virtual scene;

[0285] Parameters for evaluating the predictive ability of game situations in virtual game scenarios.

[0286] Optionally, the shooting aiming distance variation parameter includes a first parameter and a second parameter. When the second discrimination module determines the parameter value of at least one second game capability evaluation parameter of the object to be identified based on game data, it can be used for:

[0287] Based on game data, the shooting aiming information in the game virtual scene corresponding to the object to be identified is obtained. The shooting aiming information includes the aiming distance of multiple consecutive aiming actions corresponding to each shooting operation. The aiming distance is the distance between the virtual shooting prop in the game virtual scene and the target being aimed at.

[0288] For each shooting operation, determine the degree of dispersion between the aiming distances of multiple consecutive aiming actions corresponding to that shooting operation, and use the degree of dispersion as the first parameter, and the value of the degree of dispersion as the parameter value of the first parameter.

[0289] For each shooting operation, determine the target distance pair among the aiming distances of the consecutive aiming actions corresponding to that shooting operation. The target distance pair refers to the two aiming distances corresponding to two adjacent aiming actions where the aiming distance decreases.

[0290] For each shooting operation, the sudden change in shooting aiming distance for that shooting operation is determined based on the distance difference between each target distance pair corresponding to that shooting operation;

[0291] Based on the sudden changes in aiming distance during each shooting operation, the parameter value of the second parameter of the target to be identified is determined.

[0292] It is understood that the apparatus of this application embodiment can execute the method provided in this application embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0293] This application provides an electronic device, including a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program stored in the memory, it can implement the method in any optional embodiment of this application.

[0294] Figure 10 A schematic diagram of the structure of an electronic device to which an embodiment of the present invention applies is shown, such as... Figure 10 As shown, the electronic device can be a server or a user terminal, and it can be used to implement the methods provided in any embodiment of the present invention.

[0295] like Figure 10 As shown, the electronic device 2000 may primarily include at least one processor 2001. Figure 10 The diagram shows components such as a memory 2002, a communication module 2003, and an input / output interface 2004. Optionally, these components can be connected and communicate with each other via a bus 2005. It should be noted that... Figure 10 The structure of the electronic device 2000 shown is merely illustrative and does not constitute a limitation on the electronic devices to which the methods provided in the embodiments of this application are applicable.

[0296] The memory 2002 can be used to store operating systems and applications, etc. The applications can include computer programs that implement the methods shown in the embodiments of the present invention when invoked by the processor 2001, and can also include programs for implementing other functions or services. The memory 2002 can be ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices that can store information and computer programs, or it can be EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disk storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.

[0297] Processor 2001 is connected to memory 2002 via bus 2005, and implements corresponding functions by calling application programs stored in memory 2002. Processor 2001 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 2001 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0298] Electronic device 2000 can connect to a network via communication module 2003 (which may include, but is not limited to, components such as a network interface) to communicate with other devices (such as user terminals or servers) through the network and achieve data interaction, such as sending data to or receiving data from other devices. Communication module 2003 may include wired network interfaces and / or wireless network interfaces, meaning the communication module may include at least one of wired or wireless communication modules.

[0299] Electronic device 2000 can connect to required input / output devices, such as keyboards and display devices, via input / output interface 2004. Electronic device 2000 itself may have a display device, and other display devices can also be connected externally via interface 2004. Optionally, storage devices, such as hard drives, can also be connected via interface 2004 to store data from electronic device 2000, retrieve data from storage devices, or store data from storage devices into memory 2002. It is understood that input / output interface 2004 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to input / output interface 2004 can be a component of electronic device 2000 or an external device connected to electronic device 2000 when needed.

[0300] The bus 2005 used to connect the components may include a pathway for transmitting information between the components. The bus 2005 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Depending on its function, the bus 2005 can be divided into address bus, data bus, control bus, etc.

[0301] Optionally, for the solution provided in the embodiments of the present invention, the memory 2002 can be used to store a computer program that executes the solution of the present invention, and the processor 2001 runs the computer program. When the processor 2001 runs the computer program, it implements the operation of the method or apparatus provided in the embodiments of the present invention.

[0302] Based on the same principle as the method provided in the embodiments of this application, the embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the corresponding content of the aforementioned method embodiments.

[0303] This application also provides a computer program product, which includes a computer program that, when executed by a processor, can implement the corresponding content of the aforementioned method embodiments.

[0304] It should be noted that the terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown in the figures or text.

[0305] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0306] The above are only optional implementation methods for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application, without departing from the technical concept of this application, also fall within the protection scope of the embodiments of this application.

Claims

1. An object recognition method, characterized in that, The method includes: Obtain game data of the object to be identified in the target game application; Based on the game data, determine the parameter value of at least one first game ability evaluation parameter of the object to be identified; Based on the parameter values ​​of each of the first game ability evaluation parameters, a preliminary identification result of the object type of the object to be identified is predicted by the trained first object recognition model. The preliminary identification result is either a target type or a non-target type, and the non-target type includes a first type and a second type. The first object recognition model includes at least two classification models, which are trained on their respective training datasets. At least two of the at least two training datasets corresponding to the at least two classification models have at least some different training samples. When the initial identification result is the target type, based on the game data, the parameter value of at least one second game ability evaluation parameter of the object to be identified is determined, wherein the at least one second game ability evaluation parameter is an evaluation parameter used to measure the difference between the target type object and the first type object; Based on the parameter values ​​of each of the second game ability evaluation parameters, determine whether the object to be identified is an object of the target type; Based on the parameter values ​​of each of the first game ability evaluation parameters, and through a trained first object recognition model, a preliminary recognition result of the object type of the object to be recognized is predicted, including: Based on the parameter values ​​of each of the first game ability assessment parameters, a preliminary prediction result of the object type of the object to be identified is obtained through each of the at least two classification models; Based on at least two preliminary prediction results corresponding to the at least two classification models, a preliminary identification result is determined for the object type of the object to be identified.

2. The method according to claim 1, characterized in that, The preliminary prediction result includes the probability that the object type of the object to be identified is the target type; The preliminary identification result for determining the object type of the object to be identified based on at least two preliminary prediction results corresponding to the at least two classification models includes: If there is a probability greater than or equal to a first set threshold among the at least two probabilities corresponding to the at least two classification models, then the preliminary identification result is determined to be the target type; If at least two probabilities corresponding to the at least two classification models are both less than a first set threshold, then the at least two probabilities are fused to obtain a fused probability. If the fused probability is greater than or equal to a second set threshold, then the preliminary identification result is determined to be the target type. If at least two probabilities are both less than a first set threshold and the fusion probability is less than a second set threshold, then the preliminary identification result is determined to be a non-target type.

3. The method according to claim 1, characterized in that, The first object recognition model includes a first classification model and a second classification model, and the first object recognition model is trained in the following way: Obtain a first training dataset, which includes multiple labeled training samples. Each training sample includes the parameter values ​​of each of the first game ability evaluation parameters of a sample object. The label of each training sample indicates whether the real object type of the sample object corresponding to the training sample is the target type or a non-target type. The first training dataset is divided into a training set and a validation set; The initial classification model is trained based on the training set to obtain a third classification model that satisfies the first preset condition, and the first classification model is obtained based on the third classification model. The third classification model is used to predict the object type of each training sample in the validation set. Based on the third prediction results and the training samples with mismatched labels in the validation set, a second training dataset is constructed. The initial classification model is trained based on the second training dataset to obtain a second classification model that meets the second preset conditions.

4. The method according to claim 3, characterized in that, The step of dividing the first training dataset into a training set and a validation set includes: Divide the first training dataset into at least three datasets; For each of the at least three datasets, that dataset is used as the validation set, and the datasets other than that dataset among the at least three datasets are used as the training set to obtain a sample combination; The step of training the initial classification model based on the training set to obtain a third classification model that satisfies the first preset condition, and obtaining the first classification model based on the third classification model, includes: The initial classification model is trained based on the training set of each sample combination to obtain the third classification model that satisfies the first preset condition for each sample combination. Obtain the test set, and use the test samples in the test set to test the performance of the third classification model corresponding to each sample combination, and obtain the model test results corresponding to each sample combination; Based on the model test results corresponding to each sample combination, the first classification model is selected from the third classification models corresponding to each sample combination. The third prediction result, obtained by using the third classification model to predict the object type corresponding to each training sample in the validation set, includes: For the validation set of each sample combination, the third prediction result of each training sample in the validation set of that sample combination is predicted using the third classification model corresponding to that sample combination. The second training dataset includes: the third prediction result in the validation set of each sample combination and training samples with mismatched labels.

5. The method according to claim 1, characterized in that, The step of determining whether the object to be identified is an object of the target type based on the parameter values ​​of each of the second game ability evaluation parameters includes: Obtain the parameter threshold corresponding to each of the second game ability evaluation parameters; If the parameter values ​​of each of the second game ability evaluation parameters and their corresponding parameter thresholds all meet the setting conditions of each evaluation parameter, then the object to be identified is determined to be a non-target type object; otherwise, the object to be identified is determined to be a target type object.

6. The method according to claim 5, characterized in that, The threshold values ​​for each of the second game ability evaluation parameters are determined in the following manner: Obtain multiple sample objects corresponding to the game replay files of the target game application, wherein the multiple sample objects include multiple first objects and multiple second objects, the first objects are objects of the target type, and the second objects are objects of the first type; The game replay file corresponding to each of the sample objects is parsed to obtain the game data corresponding to each of the sample objects; For each of the sample objects, the parameter values ​​corresponding to each of the second game ability evaluation parameters are determined based on the game data of the sample object; For each of the second game ability evaluation parameters, a parameter threshold corresponding to the evaluation parameter is determined based on the parameter values ​​of the plurality of first objects corresponding to the evaluation parameter and the parameter values ​​of the plurality of second objects corresponding to the evaluation parameter.

7. The method according to claim 1, characterized in that, The step of determining whether the object to be identified is an object of the target type based on the parameter values ​​of each of the second game ability evaluation parameters includes: Based on the parameter values ​​of each of the second game ability evaluation parameters, the probability that the object type of the object to be identified belongs to the target type is predicted by the trained second object recognition model. Based on the probability of belonging to the target type, determine whether the object to be identified is an object of the target type; The second object recognition model is trained based on a third training dataset, which includes multiple first samples and multiple second samples. Each first sample is the parameter value of each second game ability evaluation parameter of a target type sample object, and each second sample is the parameter value of each second game ability evaluation parameter of a first type sample object.

8. The method according to any one of claims 1 to 7, characterized in that, The target game application is a shooting game application, and the at least one second game capability evaluation parameter includes at least one of the following: Game ability level assessment parameters; The shooting aiming distance variation parameter includes at least one of a first parameter or a second parameter, wherein the parameter value of the first parameter characterizes the fluctuation of the shooting aiming distance during the game, and the parameter value of the second parameter characterizes the sudden change of the shooting aiming distance during the game. The number of times a specified action is performed in the game's virtual scene; Parameters for evaluating the predictive ability of game situations in virtual game scenarios.

9. The method according to claim 8, characterized in that, The shooting aiming distance variation parameter includes the first parameter and the second parameter. The step of determining the parameter value of at least one second game ability evaluation parameter of the object to be identified based on the game data includes: Based on the game data, shooting aiming information in the game virtual scene corresponding to the object to be identified is obtained. The shooting aiming information includes the aiming distance of multiple consecutive aiming actions corresponding to each shooting operation. The aiming distance is the distance between the virtual shooting prop and the target being aimed at in the game virtual scene. For each shooting operation, the dispersion of the aiming distances between the multiple consecutive aiming actions corresponding to that shooting operation is determined, and the dispersion is used as the first parameter, and the value of the dispersion is used as the parameter value of the first parameter. For each shooting operation, determine the target distance pair among the aiming distances of the multiple consecutive aiming actions corresponding to that shooting operation. The target distance pair refers to the two aiming distances corresponding to two adjacent aiming actions where the aiming distance decreases. For each shooting operation, the sudden change in shooting aiming distance for that shooting operation is determined based on the distance difference between each target distance pair corresponding to that shooting operation; Based on the sudden changes in the aiming distance during each shooting operation, the parameter value of the second parameter of the object to be identified is determined.

10. An object recognition device, characterized in that, The device includes: The game data acquisition module is used to acquire game data of the object to be identified in the target game application; The first discrimination module is used to determine the parameter value of at least one first game ability evaluation parameter of the object to be identified based on the game data, and predict the preliminary identification result of the object type of the object to be identified by using a trained first object recognition model based on the parameter value of each first game ability evaluation parameter, wherein the preliminary identification result is a target type or a non-target type, and the non-target type includes a first type and a second type; the first object recognition model includes at least two classification models, wherein the at least two classification models are trained on their respective corresponding training datasets, and there are at least some different training samples between any two datasets in the at least two training datasets corresponding to the at least two classification models; The second discrimination module is used to determine the parameter value of at least one second game ability evaluation parameter of the object to be identified based on the game data when the preliminary identification result is the target type, and to determine whether the object to be identified is an object of the target type according to the parameter value of each second game ability evaluation parameter, wherein the at least one second game ability evaluation parameter is an evaluation parameter used to measure the difference between the target type object and the first type object; When the first discrimination module is used to predict the preliminary identification result of the object type of the object to be identified based on the parameter values ​​of each of the first game ability evaluation parameters and through the trained first object recognition model, it is specifically used to: obtain the preliminary prediction result of the object type of the object to be identified based on the parameter values ​​of each of the first game ability evaluation parameters and through each of the at least two classification models; and determine the preliminary identification result of the object type of the object to be identified based on the at least two preliminary prediction results corresponding to the at least two classification models.

11. The apparatus according to claim 10, characterized in that, The preliminary prediction result includes the probability that the object type of the object to be identified is the target type; When the first discrimination module determines the preliminary identification result of the object type of the object to be identified based on at least two preliminary prediction results corresponding to the at least two classification models, it is specifically used for: If there is a probability greater than or equal to a first set threshold among the at least two probabilities corresponding to the at least two classification models, then the preliminary identification result is determined to be the target type; If at least two probabilities corresponding to the at least two classification models are both less than a first set threshold, then the at least two probabilities are fused to obtain a fused probability. If the fused probability is greater than or equal to a second set threshold, then the preliminary identification result is determined to be the target type. If at least two probabilities are both less than a first set threshold and the fusion probability is less than a second set threshold, then the preliminary identification result is determined to be a non-target type.

12. The apparatus according to claim 10, characterized in that, The first object recognition model includes a first classification model and a second classification model, and the first object recognition model is trained in the following way: Obtain a first training dataset, which includes multiple labeled training samples. Each training sample includes the parameter values ​​of each of the first game ability evaluation parameters of a sample object. The label of each training sample indicates whether the real object type of the sample object corresponding to the training sample is the target type or a non-target type. The first training dataset is divided into a training set and a validation set; The initial classification model is trained based on the training set to obtain a third classification model that satisfies the first preset condition, and the first classification model is obtained based on the third classification model. The third classification model is used to predict the object type of each training sample in the validation set. Based on the third prediction results and the training samples with mismatched labels in the validation set, a second training dataset is constructed. The initial classification model is trained based on the second training dataset to obtain a second classification model that meets the second preset conditions.

13. The apparatus according to claim 12, characterized in that, The step of dividing the first training dataset into a training set and a validation set includes: Divide the first training dataset into at least three datasets; For each of the at least three datasets, that dataset is used as the validation set, and the datasets other than that dataset among the at least three datasets are used as the training set to obtain a sample combination; The step of training the initial classification model based on the training set to obtain a third classification model that satisfies the first preset condition, and obtaining the first classification model based on the third classification model, includes: The initial classification model is trained based on the training set of each sample combination to obtain the third classification model that satisfies the first preset condition for each sample combination. Obtain the test set, and use the test samples in the test set to test the performance of the third classification model corresponding to each sample combination, and obtain the model test results corresponding to each sample combination; Based on the model test results corresponding to each sample combination, the first classification model is selected from the third classification models corresponding to each sample combination. The third prediction result, obtained by using the third classification model to predict the object type corresponding to each training sample in the validation set, includes: For the validation set of each sample combination, the third prediction result of each training sample in the validation set of that sample combination is predicted using the third classification model corresponding to that sample combination. The second training dataset includes: the third prediction result in the validation set of each sample combination and training samples with mismatched labels.

14. The apparatus according to claim 10, characterized in that, The second discrimination module, when determining whether the object to be identified is an object of the target type based on the parameter values ​​of each of the second game ability evaluation parameters, is specifically used for: Obtain the parameter threshold corresponding to each of the second game ability evaluation parameters; If the parameter values ​​of each of the second game ability evaluation parameters and their corresponding parameter thresholds all meet the setting conditions of each evaluation parameter, then the object to be identified is determined to be a non-target type object; otherwise, the object to be identified is determined to be a target type object.

15. The apparatus according to claim 14, characterized in that, The threshold values ​​for each of the second game ability evaluation parameters are determined in the following manner: Obtain multiple sample objects corresponding to the game replay files of the target game application, wherein the multiple sample objects include multiple first objects and multiple second objects, the first objects are objects of the target type, and the second objects are objects of the first type; The game replay file corresponding to each of the sample objects is parsed to obtain the game data corresponding to each of the sample objects; For each of the sample objects, the parameter values ​​corresponding to each of the second game ability evaluation parameters are determined based on the game data of the sample object; For each of the second game ability evaluation parameters, a parameter threshold corresponding to the evaluation parameter is determined based on the parameter values ​​of the plurality of first objects corresponding to the evaluation parameter and the parameter values ​​of the plurality of second objects corresponding to the evaluation parameter.

16. The apparatus according to claim 10, characterized in that, When the second discrimination module determines whether the object to be identified is an object of the target type based on the parameter values ​​of each of the second game ability evaluation parameters, it is specifically used for: Based on the parameter values ​​of each of the second game ability evaluation parameters, the probability that the object type of the object to be identified belongs to the target type is predicted by the trained second object recognition model. Based on the probability of belonging to the target type, determine whether the object to be identified is an object of the target type; The second object recognition model is trained based on a third training dataset, which includes multiple first samples and multiple second samples. Each first sample is the parameter value of each second game ability evaluation parameter of a target type sample object, and each second sample is the parameter value of each second game ability evaluation parameter of a first type sample object.

17. The apparatus according to any one of claims 10-16, characterized in that, The target game application is a shooting game application, and the at least one second game capability evaluation parameter includes at least one of the following: Game ability level assessment parameters; The shooting aiming distance variation parameter includes at least one of a first parameter or a second parameter, wherein the parameter value of the first parameter characterizes the fluctuation of the shooting aiming distance during the game, and the parameter value of the second parameter characterizes the sudden change of the shooting aiming distance during the game. The number of times a specified action is performed in the game's virtual scene; Parameters for evaluating the predictive ability of game situations in virtual game scenarios.

18. The apparatus according to claim 17, characterized in that, The shooting aiming distance change parameter includes the first parameter and the second parameter. When the second discrimination module determines the parameter value of at least one second game ability evaluation parameter of the object to be identified based on the game data, it is specifically used for: Based on the game data, shooting aiming information in the game virtual scene corresponding to the object to be identified is obtained. The shooting aiming information includes the aiming distance of multiple consecutive aiming actions corresponding to each shooting operation. The aiming distance is the distance between the virtual shooting prop and the target being aimed at in the game virtual scene. For each shooting operation, the dispersion of the aiming distances between the multiple consecutive aiming actions corresponding to that shooting operation is determined, and the dispersion is used as the first parameter, and the value of the dispersion is used as the parameter value of the first parameter. For each shooting operation, determine the target distance pair among the aiming distances of the multiple consecutive aiming actions corresponding to that shooting operation. The target distance pair refers to the two aiming distances corresponding to two adjacent aiming actions where the aiming distance decreases. For each shooting operation, the sudden change in shooting aiming distance for that shooting operation is determined based on the distance difference between each target distance pair corresponding to that shooting operation; Based on the sudden changes in the aiming distance during each shooting operation, the parameter value of the second parameter of the object to be identified is determined.

19. An electronic device, characterized in that, The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program, implements the method of any one of claims 1 to 9.

20. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method of any one of claims 1 to 9.

21. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 9.

Images