Game interface error detection platform based on image recognition technology

Abstract

The invention discloses a game interface error detection platform based on an image recognition technology, and relates to the technical field of game detection, and the platform comprises the following steps: aiming at a game dynamic frame frequency, integrating training samples based on error discrimination elements; generating an error detection module based on an adversarial training principle in combination with the training sample; receiving a frame frequency image of a game interface, assisting an image recognition module, executing the error judgment elements for element extraction and element reconstruction, and determining a judgment element matrix; and the judgment element matrix is transmitted to the error detection module for judgment, error detection data of the frame frequency image is generated, and popup warning is carried out. The technical problems that in the prior art, multiple types of error judgment elements in the game interface cannot be effectively processed, and the error detection efficiency and accuracy are low are solved, and the technical effects that the frame frequency image errors of the game interface are accurately detected, effective pop-up window warning is achieved, and the error detection efficiency and accuracy are improved are achieved.

Description

Technical Field

[0001] The present invention relates to the field of game detection technology, and in particular to a game interface error detection platform based on image recognition technology. Background Art

[0002] With the rapid development of the gaming industry, the complexity and dynamism of game screens are increasing. In traditional game interface error detection technology, the error discrimination in the game screen is often lacking in comprehensiveness. On the one hand, there are many types of game interface elements, including various icons, texts, colors and other directly presented information, as well as character behaviors, scene changes and other elements that are deeply integrated with the game scene. Most of the existing technologies can only perform simple detection on a single type of element, and cannot accurately handle the situation of multiple types of elements that include direct discrimination classes and those constrained by the game scene at the same time. On the other hand, the construction of the discriminator is not accurate enough, and there is a lack of targeted processing mechanisms for elements of different natures (such as static display elements and dynamic interactive elements in the game). Moreover, for elements affected by scene changes, there is no effective way to deal with the discrimination changes caused by scene transfer, which greatly reduces the accuracy of error detection in the complex game scene switching process.

[0003] The existing technology has the technical problem of being unable to effectively handle multiple types of error discrimination factors in the game interface, and having low error detection efficiency and accuracy. Summary of the invention

[0004] The present application provides a game interface error detection platform based on image recognition technology, which is used to solve the technical problems that the existing technology cannot effectively handle multiple types of error discrimination elements in the game interface, and the error detection efficiency and accuracy are low.

[0005] In view of the above problems, the present application provides a game interface error detection platform based on image recognition technology.

[0006] The present application provides a game interface error detection platform based on image recognition technology, the platform comprising: A training sample integration module, the training sample integration module is used to integrate training samples based on error discrimination factors for the dynamic frame rate of the game, the error discrimination factors include a type of direct discrimination category and a type of game scene constraint-containing factor; a discriminator construction module, the discriminator construction module is based on the adversarial training principle, combined with the training samples, to construct a point discriminator for the type one factor, and a sequence discriminator for the type two factors, and generate an error detection module, wherein the sequence discriminator has a built-in discrimination transfer function that obeys the scene transfer law; a discriminant factor matrix determination module, the discriminant factor matrix determination module is used to receive the frame rate image of the game interface, assist the image recognition module, execute the error discrimination factor to perform factor extraction and factor reconstruction, and determine the discriminant factor matrix, wherein the factor reconstruction is guided by the detection requirements; a comprehensive discrimination module, the comprehensive discrimination module is used to transmit the discriminant factor matrix to the error detection module, perform classification discrimination and comprehensive discrimination, generate error detection data of the frame rate image and issue a pop-up window warning.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: The training sample integration module is used to integrate training samples based on error discrimination factors for the dynamic frame rate of the game; the discriminator construction module, based on the adversarial training principle, combines the training samples, constructs a point discriminator for the first type of factors, and a sequence discriminator for the second type of factors, to generate an error detection module; the discriminant factor matrix determination module is used to receive the frame rate image of the game interface, assist the image recognition module, execute the error discrimination factors to extract and reconstruct factors, and determine the discriminant factor matrix; the comprehensive discrimination module transmits the discriminant factor matrix to the error detection module, performs classification and comprehensive discrimination, generates error detection data for the frame rate image, and issues a pop-up window warning. The accurate detection of the frame rate image errors of the game interface and the effective pop-up window warning are achieved, and the technical effect of improving the efficiency and accuracy of error detection is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A schematic diagram of the structure of a game interface error detection platform based on image recognition technology provided in an embodiment of the present application; Figure 2A flow chart of a discriminant factor matrix determination module in a game interface error detection platform based on image recognition technology provided in an embodiment of the present application.

[0010] Explanation of the reference numerals: training sample integration module 10 , discriminator construction module 20 , discriminant factor matrix determination module 30 , comprehensive discrimination module 40 . DETAILED DESCRIPTION

[0011] The present application provides a game interface error detection platform based on image recognition technology, which is used to solve the technical problems that the existing technology cannot effectively handle multiple types of error discrimination elements in the game interface and has low error detection efficiency and accuracy.

[0012] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0013] Examples, such as Figure 1 As shown, the present application provides a game interface error detection platform based on image recognition technology, and the platform includes: The training sample integration module 10 is used to integrate training samples based on error discrimination factors for the dynamic frame rate of the game. The error discrimination factors include a first type of factors of the direct discrimination class and a second type of factors with game scene constraints.

[0014] Specifically, the training sample integration module 10 carefully integrates training samples for the dynamic frame rate of the game in the game interface error detection platform based on image recognition technology, and this process is closely centered on the error discrimination elements. The error discrimination elements cover the first type of elements of the direct discrimination class and the second type of elements with game scene constraints. For the first type of elements, they belong to the basic characteristic elements that can be quickly obtained directly from the game interface image and used to judge errors. For example, in the game interface, certain specific graphic signs, color combinations or specific pixel patterns can be classified as first type of elements. These elements are direct and clear, and can provide important clues in preliminary image analysis. The second type of elements are deeply related to the game scene, and their discrimination needs to be combined with specific game situations and rules. In different game scenes, the expression form and discrimination criteria of the second type of elements are different. For example, in simulation business games, the appearance of the construction progress of building facilities at a specific development stage and the normal working status of resource collection points (such as workers' activities, collection animation effects, etc.) belong to the category of second type of elements. If, in a certain frame rate image, the construction progress displayed by the building does not match the actual resources and time invested, or there are no workers working at the resource collection point but the normal collection status is displayed, these abnormal situations need to be identified based on the second type of elements. The training sample integration module 10 will deeply analyze the dynamic characteristics of the game, accurately locate the way in which various errors may be presented in the frame rate image, and for the first type of elements, by widely collecting image fragments in different states during the game operation, including normal interfaces and artificially created or naturally generated abnormal interfaces, so as to accumulate a rich variety of samples containing various direct discriminant features. For the second type of elements, the module designs and simulates complex operations and event triggers in various game scenarios, and carefully captures the image frames corresponding to each key link, covering comprehensive sample data from normal scene processes to various abnormal situations. By rigorously organizing and collating these samples, a comprehensive, accurate and highly representative training sample set is provided for the subsequent discriminator construction module, laying a solid data foundation for the entire game interface error detection system, ensuring that the system can accurately identify and judge various game interface errors, and ensuring the smoothness and stability of the game experience.

[0015] The discriminator construction module 20 is based on the adversarial training principle, combines the training samples, constructs a point discriminator for the first category of elements, constructs a sequence discriminator for the second category of elements, and generates an error detection module, wherein the sequence discriminator has a built-in discriminant transfer function that obeys the scene transfer law.

[0016] Specifically, the discriminator construction module 20 plays a key role in the game interface error detection platform based on image recognition technology. Its operation is based on the advanced adversarial training principle. Through in-depth analysis and application of training samples, it constructs highly targeted point discriminators and sequence discriminators, and then generates an efficient error detection module. The sequence discriminator has a built-in unique discriminant transfer function that obeys the law of scene transfer, which greatly improves the accuracy and adaptability of error detection. The adversarial training principle provides a solid theoretical framework for the entire construction process. Under this framework, the discriminator and the generator compete with each other and optimize together. For a class of elements, the process of constructing a point discriminator is particularly critical. A class of elements are features that can be directly obtained from the image. The point discriminator aims to accurately identify whether these isolated key feature points are abnormal. For example, in the game interface, the correctness of the shape, color, position and other features of a specific button icon can be judged by a point discriminator. By learning a large number of training samples containing the normal and abnormal states of the button, the point discriminator can establish a discriminant model about the button features, thereby quickly detecting the abnormality of the button in any frame rate image, even if it is only a slight pixel-level change. Constructing a sequence discriminator for the two-class elements requires considering more complex game scene factors. The discrimination of the two-class elements often depends on the information presented by a series of continuous image frames and the logical relationship between them. The sequence discriminator will deeply analyze the change rules of the elements in these continuous frames, such as the continuous action sequence of the character in the battle scene of an action game, the change of the effect of the previous and next frames of the skill release, etc. It masters the change pattern of the elements in normal scenes by learning a large number of samples containing such continuous scenes. The built-in discriminant transfer function obeys the scene transfer law, which means that it can intelligently adjust the discrimination strategy according to the dynamic changes of the game scene. In the game, the scene transition is frequent and complex, and the discrimination criteria of the same element in different scenes may change. For example, in a normal exploration scene of a game, the character's movement speed may be in a lower normal range, but in a chase or escape scene, the speed range will change accordingly. The discriminant transfer function can perceive this scene change in real time and apply it to the discriminant process of the two types of elements. When the scene changes from one type to another, it will quickly adjust the parameters of the discriminant model based on the pre-learned scene transition probability and element change relationship to ensure that errors related to the two types of elements can still be accurately detected in the new scene. In this way, the discriminator construction module 20 successfully constructs a powerful error detection module, providing reliable technical support for game interface error detection.

[0017] The discriminant factor matrix determination module 30 is used to receive the frame rate image of the game interface, assist the image recognition module, execute the error discriminant factor to extract and reconstruct the factor, and determine the discriminant factor matrix, wherein the factor reconstruction is guided by the detection requirements.

[0018] Specifically, the main responsibility of the discriminant element matrix determination module 30 is to receive the frame rate image of the game interface, and closely assist the image recognition module, implement element extraction and element reconstruction operations based on the error discrimination elements, and finally determine the discriminant element matrix. In the whole process, the element reconstruction is always guided by the detection requirements to ensure that the constructed matrix can accurately serve the error detection task. When the frame rate image of the game interface is passed to the module, the first thing to do is to extract the elements. Under the guidance of the error discrimination elements, the module will accurately dig out various types of information related to the error judgment from the image. For the first type of elements of the direct discrimination class, such as various icons, text prompts, color distribution and other obvious features in the game interface, the module can quickly locate and extract its key attributes, such as the outline information of the icon shape, the font and size characteristics of the text, and the pixel distribution of a specific color area. For the second type of elements constrained by the game scene, the module will deeply analyze the complex information such as the character state, action sequence, and scene element layout in the image. Taking the character action sequence as an example, it will extract the character's limb position changes, action amplitude, and time interval between actions in the continuous frame rate image. These elements are crucial to judging whether the character behavior conforms to the game logic. After the extraction of elements is completed, the reconstruction phase begins. Since different error detection requirements have different focuses, the reconstruction of elements will reorganize and optimize the extracted elements according to specific requirements. If the detection focus is on the visual stability of the game interface, then in the reconstruction process, the first type of elements related to the interface layout, color matching, icon display, etc. will be given a higher weight, and they will be arranged and combined in a way that is more conducive to detecting visual anomalies in the interface to form a more targeted feature vector. When the detection needs focus on the correctness of the behavioral logic of the game character, the second type of elements related to the character action, state change, etc. will become the core of the reconstruction, integrating the character action sequence elements in the continuous frame rate image to highlight the logical relationship between the actions, so as to more effectively determine whether the character behavior is wrong. After a series of fine operations such as element extraction and reconstruction, the module finally determines the discriminant element matrix. This matrix will serve as an important input for the subsequent comprehensive discrimination module, providing a key data basis for accurately judging whether there are errors in the game interface and the type of errors, and effectively promoting the smooth progress of the entire game interface error detection process.

[0019] The comprehensive discrimination module 40 is used to transmit the discrimination factor matrix to the error detection module, perform classification discrimination and comprehensive discrimination, generate error detection data of the frame rate image and issue a pop-up window warning.

[0020] Specifically, the comprehensive discrimination module 40 is responsible for transmitting the discriminant element matrix to the error detection module, and based on this, performs comprehensive and in-depth classification discrimination and comprehensive discrimination work, and finally generates error detection data of the frame rate image and triggers a pop-up window warning to ensure that the game interface errors can be discovered and processed in time. After receiving the discriminant element matrix, it is first transmitted to the error detection module. The error detection module classifies and discriminates various elements in the discriminant element matrix based on the point discriminator and sequence discriminator constructed in the early stage. For a type of element of the direct discriminant class, the point discriminator plays a role and quickly determines whether each isolated element is abnormal. For example, it can accurately detect whether the display of a button in the game interface is normal, including whether its shape, color, position, etc. meet the preset standards. If the button is deformed, the color is abnormal, or the position is offset, the point discriminator can quickly identify and mark it. At the same time, the sequence discriminator discriminates the second type of elements constrained by the game scene. It deeply analyzes the changing rules of the elements in the continuous frame rate image to determine whether it conforms to the normal logic under the game scene. For example, in a battle scene, whether the character's skill release action sequence is correct, from the starting action of the skill to the subsequent special effects display and the impact on the enemy, the sequence discriminator will strictly compare it according to the normal mode learned in advance. After the classification and discrimination are completed, the comprehensive discrimination module performs comprehensive discrimination. It integrates the results from different discriminators, considers the relationship between various elements, and their synergy in the entire game interface and scene. For example, the abnormality of a button may not only be its own display problem, but also related to factors such as surrounding text prompts and scene interactions. The comprehensive discrimination module generates more accurate and comprehensive error detection data by comprehensively analyzing these complex relationships. Once an error is detected, the comprehensive discrimination module immediately activates the pop-up warning mechanism, and the pop-up window will clearly display key information such as the type, location, and possible impact range of the error, so that game developers or testers can quickly locate the problem and take corresponding repair measures. At the same time, the pop-up warning can also be displayed in levels according to the severity of the error, ensuring that urgent and serious errors can be handled first, thereby minimizing the impact of the error on the game experience and ensuring the stable operation of the game.

[0021] In a possible implementation, the discriminator building module further includes: Determine a target training scene, wherein the target training scene is any game scene.

[0022] The target training scenario is limited, and adversarial training is performed under the training architecture in combination with training samples based on the two types of elements to generate an initialization architecture, wherein the training architecture is composed of a generator and a discriminator.

[0023] The initialization architecture is disassembled to obtain a trained sequence discriminator.

[0024] Specifically, determining the target training scene is the starting point of the entire process. This choice is extremely critical because the game scenes are rich and diverse, and the visual presentation, element association, and logical rules in different scenes vary greatly. Whether it is the fantasy world scene in a role-playing game, the battle scene in a competitive game, or the puzzle scene in a puzzle game, each has unique interface elements and dynamic change rules. For example, in the fantasy world scene of a role-playing game, elements such as the movement of the character, the interaction with the environment, and the release of magic skills together constitute a complex picture, and these elements have different manifestations in different areas (such as towns, forests, dungeons, etc.). By determining a specific target training scene, subsequent training can focus on the unique error discrimination elements in the scene, especially the second-class elements closely related to the scene, thereby improving the pertinence and accuracy of the discriminator.

[0025] The target training scene is limited, and the focus of subsequent training work is clarified. Taking a large-scale 3D action adventure game as an example, assume that the selected target training scene is a battle scene in the game. At this time, it is necessary to accurately define the various boundary conditions of the battle scene, including the geographical scope of the scene, the type and number of characters involved in the battle, the time conditions when the battle occurs (such as day or night), and environmental factors that may affect the display of the battle interface (such as weather effects, terrain, etc.). Through such a limitation, the two types of elements closely related to the battle scene can be effectively isolated and focused, avoiding the interference of other irrelevant scene factors, and providing a clear and stable context environment for subsequent training. Next, the training samples based on the two types of elements are introduced. These training samples are carefully selected and preprocessed, specifically for the limited battle scenes. For example, the sample contains continuous image frames when different characters release various skills in battle. These image frames record in detail a series of dynamic changes such as the starting action of skill release, the intermediate special effects, and the final impact on the enemy or the environment. At the same time, it also includes the interface display changes of the character in different states (such as normal combat state, injured state, gain state, etc.), as well as various situations of other elements related to combat (such as health bar changes, weapon equipment status updates, combat prompt information display, etc.). These training samples not only cover the standard situations under normal combat processes, but also include various abnormal situations artificially simulated or collected from actual games, such as skill special effects display errors, delayed character status updates, incomplete combat prompt information, etc. Then, these rich training samples are integrated into the training architecture composed of the generator and the discriminator, and the adversarial training process is officially started. In this process, the main responsibility of the generator is to try to generate realistic image samples based on the given combat scene characteristics and the two-category element pattern. It will use the generative model in the deep learning algorithm, the generator part of the generative adversarial network (GAN), learn the data distribution law in the training samples, and create samples that are highly similar to the real combat scene images in terms of visual effects and logical relationships. The generator will generate an image sequence of a character releasing a specific skill, which is as close as possible to the situation in the real game in terms of skill special effects, character actions, and interaction effects with the surrounding environment. The discriminator accurately distinguishes between real training samples and samples generated by the generator. The discriminator will conduct a detailed analysis and evaluation of the input image samples based on pre-defined discrimination criteria. It will consider whether the characteristics of various types of second-class elements in the image conform to the normal logic of the battle scene, such as whether the color, shape, and motion trajectory of skill special effects conform to the game settings, whether the character status changes are reasonable, and whether the health bar updates are consistent with the actual damage calculation. By constantly comparing real samples with generated samples, the discriminator gradually improves its discrimination ability and can more accurately identify subtle differences between real and fake samples. In the interaction process between the generator and the discriminator, they form a dynamic adversarial relationship.The generator constantly tries to deceive the discriminator and generate samples that are more difficult to distinguish; while the discriminator constantly improves its ability to distinguish and is not fooled by the generator. As the training continues, both sides continuously optimize their parameters and model structures in this competition. After a large number of iterative training, the samples generated by the generator become more and more realistic, and the discrimination accuracy of the discriminator becomes higher and higher. The two finally reach a relatively balanced state, at which time the initialization architecture is generated. This initialization architecture is the initial result of the deep training of the generator and the discriminator in a specific combat scenario. It contains the initial recognition and generation capabilities of the relevant features of the two types of elements in this scenario, laying a solid foundation for further training and optimization, thereby providing a more targeted and accurate discrimination tool for game interface error detection.

[0026] The initialization architecture model is the overall model architecture including the generator and the discriminator obtained after the previous target training scenario limitation and adversarial training based on the two-category element training samples. It includes the model's layer structure information (such as the type of each layer, connection relationship, etc.), parameter information (number of parameters, shape, etc.) and the intermediate data related to the discriminator recorded during the training process (such as the change in the loss function value of the discriminator during the training process, etc., which is used to assist in judging the key part of the discriminator). Based on the layer structure in the architecture description information, traverse the layers of the model in reverse from the output layer. For each layer, check its input source and output destination. If the output of a layer mainly flows to the subsequent processing part related to the wrong discrimination (such as the final classification discriminant layer), it is marked as a potential discriminator key layer. At the same time, refer to the intermediate data recorded during the training process. For example, if a certain layer plays a key role in the calculation of the discriminator loss function (such as the gradient calculation found that the parameter update of this layer has a greater impact on the loss function), it is further confirmed as a key layer. Continue to traverse backward until you find a layer boundary that is clearly distinguished from the generator input. This is determined by checking whether the layer's input data source comes from the generator part. Once this boundary layer is determined, all layers between the boundary layer and the previously marked key layer are determined as the core part of the discriminator. Based on the determined discriminator core part, extract the parameters belonging to these layers from the parameter set of the initialization architecture. For each discriminator core layer, accurately copy or extract the corresponding parameter value from the overall parameter set according to its parameter structure (such as weight matrix, bias vector, etc.). During the extraction process, handle the existing shared parameters. If some parameters are found to be shared between the generator and the discriminator, decide whether to copy the shared parameters to the discriminator for independent use or to mark them specially for subsequent processing according to pre-set rules (such as based on the dominant direction of parameter update during training or based on the priority of the current disassembly requirements). Reorganize the extracted parameters according to the order of the discriminator's layer structure to construct a discriminator's independent parameter set. This step ensures that the discriminator can correctly load and use these parameters for calculation after leaving the initialization architecture. According to the layer structure information of the discriminator (obtained in the architecture description information), a new independent model object is created using the extracted and reorganized parameter set to represent the sequence discriminator. For each layer, according to its original layer type (such as convolutional layer, fully connected layer, etc.) and connection relationship, the API of the corresponding deep learning framework (such as TensorFlow or PyTorch) is used to instantiate the layer object and load the reorganized parameters into these layers. Constructing the input layer so that it can receive the second-class feature sequence data of the game interface frame rate image involves the definition of the input data format and the integration of preprocessing operations, for example, setting the shape of the input data to a format suitable for processing sequence data (such as [sequence length, image feature dimension]), and adding necessary normalization or encoding operations. Constructing the output layer so that it can output the discrimination results that meet the error detection requirements.For example, the output layer is set to a binary classification (normal / abnormal) or multi-classification (corresponding to different error types) layer, and its output can be in the form of category labels or probability values. Use a small part of the validation data set (containing known normal and abnormal binary feature sequence data) to verify the constructed sequence discriminator model. Input the validation data into the sequence discriminator and calculate its performance indicators such as discrimination accuracy and recall. If it is found that the performance does not meet expectations, check whether there are problems such as parameter errors and layer connection errors during the model construction process, and make corresponding adjustments and optimizations. Further fine-tune the model, such as adjusting hyperparameters (such as learning rate, regularization parameters, etc.) or making slight modifications to the model structure (such as adding additional hidden layers or adjusting the number of parameters of the layer) to improve the performance of the sequence discriminator. After the above algorithm steps are processed, the trained sequence discriminator model is output, which can independently discriminate errors in the binary feature sequence in the game interface frame rate image, providing key support for subsequent comprehensive discrimination and error detection.

[0027] In a possible implementation, the discriminator building module further includes: Identify training samples of the two types of elements, perform scene clustering processing, and determine multiple scene clusters.

[0028] The plurality of scene clusters are traversed to determine a cluster discrimination probability, wherein the cluster discrimination probability includes a mean value of element discrimination probabilities in each scene cluster.

[0029] Taking the target training scene as a baseline and the cluster discrimination probability as a basis, a probability migration relationship under scene transfer is determined.

[0030] The probability transition relationship is used as the discriminant transition function.

[0031] Specifically, in the game interface error detection system based on image recognition technology, the second-class elements need special treatment due to their scene dependency. First, it is necessary to accurately identify the second-class element training samples that are closely related to specific game scenes. These samples cover rich information in various scenes in the game. For example, in role-playing games, skill release special effects in battle scenes, character injury animations, and character interaction actions and environmental element changes in non-battle scenes are all second-class elements, and their expressions and attribute characteristics in different scenes are different. Next, perform scene clustering processing and train with a specific scene as the standard, such as selecting the main city scene in the game as the benchmark scene. For a large number of collected second-class element training samples, clustering is performed based on the similarity between samples. The similarity measurement standard here is based on the feature differences of the second-class elements in different scenes. For example, the interface display elements (including dialog box position, text color, expression, etc.) when the character and the dialogue in the main city scene have significantly different feature patterns from the elements related to the release of character skills in the battle scene (skill icon, special effect range, damage value display, etc.). Through the clustering algorithm, samples with similar feature patterns are clustered together to determine multiple scene clusters. In this process, the discriminant model trained with the main city scene as the standard needs to compensate for the discriminant results based on the migration probability between scenes when facing other scenes. Because although the same two-category elements exist in different scenes, their attributes may change. For example, the range of the character's moving speed in the main city scene is in a relatively low value range under normal circumstances, but in the chase scene, the range of moving speed will be significantly increased. When detecting whether the character's moving speed is abnormal in the chase scene, the discrimination standard in the main city scene cannot be simply applied, but the discrimination result should be adjusted according to the migration probability of the element of moving speed between the main city scene and the chase scene. In this way, no matter what scene is currently in, a unified analysis can be performed in the selected benchmark scene (main city scene), and then the final discrimination result can be reasonably adjusted and output based on the predetermined migration relationship between the scenes, thereby improving the accuracy and adaptability of error detection and ensuring the normal operation of the game interface in various complex scenes.

[0032] Traversing multiple scene clusters to determine the cluster discrimination probability is a key step. After multiple scene clusters are obtained through scene clustering processing, a detailed analysis is carried out for each scene cluster. For one of the scene clusters, it is necessary to conduct an in-depth investigation of all the training samples contained in it, which are all related to specific game scenes and corresponding two-category elements. For the two-category elements in each sample, the pre-set discrimination criteria and algorithms are used to discriminate and obtain the discrimination probability of each element in each sample. For example, in a scene cluster of an action game, for the element of character skill release, if the skill special effects in the sample are complete and meet expectations, its discrimination probability may be close to 0 (indicating normal); if the skill special effects are missing or abnormal, the discrimination probability may be high (close to 1, indicating abnormality). After obtaining the discrimination probability of each element in each sample, the average of the discrimination probabilities of the same element in all samples in the scene cluster is calculated, which is the mean of the element discrimination probability of the element in the scene cluster. The mean of the discrimination probabilities of all elements in the scene cluster is summed up to obtain the cluster discrimination probability of the scene cluster. This cluster discrimination probability can reflect the discriminative characteristics of the two types of elements in the scene cluster as a whole, and provide an important basis for the subsequent adjustment of the discrimination results based on scene transfer. For example, in a scene cluster containing multiple combat scene samples, if the mean discrimination probability of the character injury animation element is high, it means that the character injury animation in the scene cluster is likely to be abnormal. This information will help to more accurately judge whether the element is normal in different scenes and the corresponding compensation direction and amplitude of the discrimination results when comparing and migrating with other scenes in the future.

[0033] Taking the target training scene as the baseline, it is a crucial task to determine the probability migration relationship under scene transfer with the help of cluster discrimination probability. The target training scene, as the reference point of the entire analysis, has known characteristic distribution and discrimination pattern of the two-category elements. For example, if the town scene in the game is selected as the target training scene, in this scene, the two-category elements such as the character's movement speed and the interface display when interacting with the NPC have their specific normal state range and discrimination criteria. Then, for other scene clusters obtained by clustering, their cluster discrimination probabilities are analyzed. The cluster discrimination probability of each scene cluster reflects the degree of deviation of the two-category elements in the cluster from the normal state. For example, in a battle scene cluster, the cluster discrimination probability of the skill release frequency element is high, which means that there are relatively more abnormal skill release frequencies in this scene cluster. Then, by comparing the differences in cluster discrimination probabilities of the target training scene and other scene clusters in various elements, the probability migration relationship under scene transfer is determined. If we move from the town scene to the battle scene, and compare the cluster discrimination probabilities of the character movement speed factor, we find that the discrimination probability of this factor in the battle scene is significantly higher than that in the town scene. This indicates that the probability of abnormal character movement speed has changed during the scene transfer process. This change trend is the probability migration relationship. For each second-class factor, such a comparative analysis is required to obtain a complete scene transfer probability migration relationship matrix. This matrix records in detail the changing rules of the discrimination probability of each second-class factor when transferring from the target training scene to other scenes, providing a key basis for accurately discriminating the status of the second-class factors in different scenes in the future, so that the model can dynamically adjust the discrimination strategy according to the scene conversion, and improve the accuracy and adaptability of error detection.

[0034] Using the determined probability migration relationship as the discriminant migration function is a key measure to achieve accurate discrimination of the two-category elements in the scene. The probability migration relationship is obtained by deeply comparing the cluster discrimination probabilities of the target training scene and other scene clusters on various elements, and records in detail the changing trends and laws of the discrimination probabilities of each two-category element when transferring from the target training scene to other scenes. When this probability migration relationship is constructed as a discriminant migration function, the sequence discriminator is given the ability to dynamically adapt to different scenes. For example, in a role-playing game, if the target training scene is a normal map scene, when entering a special copy scene, the discrimination criteria of the two-category elements such as monster behavior patterns and environmental special effects in the scene will change. At this time, the discriminant migration function can automatically adjust the discrimination thresholds and weights of elements such as monster behavior patterns and environmental special effects based on the previously determined probability migration relationship and the change of the current scene from the normal map scene to the special copy scene. If the normal range of monster attack frequency in ordinary map scenes corresponds to a low probability of discrimination, but in special copy scenes, the probability of discrimination of this element changes according to the probability migration relationship, the discrimination migration function can adjust the discrimination strategy accordingly, so that the sequence discriminator can still accurately judge whether it is abnormal in the new scene. In this way, no matter what kind of complex and changeable scene switching the game is in, the discrimination migration function can use the probability migration relationship to adjust the discrimination method of the two types of elements in real time, thereby greatly improving the accuracy and adaptability of the entire error detection system to the game interface errors in different scenes, ensuring the stable operation of the game and the good experience of the players.

[0035] In a possible implementation, the discriminator building module further includes: Based on the error determination criterion, a factor threshold is determined.

[0036] The element threshold is used as a constraint to divide positive samples and negative samples, wherein the positive samples are the detected false detection classes.

[0037] Combined with the training samples of the two types of elements, the positive samples and the negative samples are used as the generation targets of the generator, and the element threshold is used as the discrimination criterion. The alternating iterative training of the generator and the discriminator is performed until convergence to obtain the trained discriminator, wherein the convergence condition is that the discriminator satisfies the preset discrimination accuracy.

[0038] Specifically, first, determine the element threshold based on the error judgment standard. The error judgment standard is a series of pre-set rules and indicators used to measure whether various elements in the game interface are in a normal state. These standards involve multiple aspects such as the numerical range, characteristic mode, and change law of the elements. For example, for the health value display element of the game character, its normal numerical range is set between 0 and 100. Then, based on this standard, an element threshold can be determined, such as 80 (when the health value is lower than 80, it may need to be paid attention to). For the cooling time element of the character skill release, its normal cooling time interval has a reasonable range. Exceeding this range may be regarded as abnormal, thereby determining the corresponding element threshold.

[0039] Next, the determined element threshold is used as a constraint to divide positive samples and negative samples. Among them, positive samples are detected error detection classes, that is, those samples that violate the normal range defined by the element threshold in the game interface. For example, in a battle scene, if the character's health suddenly becomes negative or exceeds 100, or the cooldown time of skill release is shorter than the specified minimum value, these samples will be classified as positive samples because they represent possible errors. Negative samples are those that meet the element threshold requirements and are in a normal state, such as samples where the character's health fluctuates within the normal range and the skill cooldown time is normal.

[0040] Then, combined with the training samples of the two types of elements, positive samples and negative samples are used as the generation targets of the generator. The generator attempts to generate realistic samples, which include both negative samples that conform to the normal state and positive samples that simulate error conditions. In this process, the element threshold is used as the discrimination criterion to perform alternating iterative training of the generator and the discriminator. The task of the discriminator is to distinguish between real training samples (including positive samples and negative samples) and samples generated by the generator. In each iteration, the generator adjusts the generation strategy according to the current discrimination results, and strives to generate samples that are more difficult for the discriminator to distinguish; while the discriminator continuously optimizes its discrimination ability based on new samples to improve the recognition accuracy of positive and negative samples. This alternating iterative training continues until the convergence condition is met, that is, the discriminator meets the preset discrimination accuracy. The preset discrimination accuracy is a pre-set target accuracy, such as 95% or higher. When the discriminator reaches or exceeds this preset value when discriminating a large number of verification samples, the training is considered to have converged. At this point, the trained discriminator obtained already has high accuracy and reliability, and can effectively identify error conditions related to the second type of factors in game interface error detection, providing important technical support for ensuring the normal operation of the game and a good experience for players.

[0041] In one possible implementation, Figure 2 The discriminant factor matrix determination module further comprises: A preset error probability is set, wherein the preset error probability limits the consistency of the determination before and after the reconstruction.

[0042] Taking the preset error probability as a constraint and the determination based on the element threshold as a detection requirement, the reconstruction conditions of the error discrimination element are mined.

[0043] Based on the reconstruction condition, the image recognition module is supervised and trained.

[0044] Specifically, in the game interface error detection system, in order to improve the pertinence and efficiency of the discrimination, a series of key operations are carried out around the preset error probability to optimize the training of the image recognition module. First, the preset error probability is set. The setting of this probability value is crucial and provides a key constraint standard for the entire reconstruction and training process. Its main function is to limit the consistency of judgments before and after reconstruction, and to ensure that when processing the erroneous discrimination elements, the judgment logic before and after remains relatively stable and accurate. For example, if the preset error probability is set to a lower value, such as 0.05 (indicating that a 5% error fluctuation is allowed), this means that when performing element reconstruction and subsequent discrimination, the high consistency of the results before and after should be guaranteed as much as possible to avoid misjudgments due to excessive reconstruction or unreasonable adjustments to the discrimination criteria.

[0045] With the preset error probability as a constraint and the judgment based on the element threshold as the detection requirement, the reconstruction conditions of the elements of error judgment are deeply excavated. Taking the dynamic element of the action state in the game task as an example, in the actual judgment, it is found that only judging a certain key position (such as the position of the character's hand action when performing the key operation of the task) can effectively judge whether the task action is normal. Assuming that the task is to carry an object, the key position information such as the contact position between the hand and the object and the angle of the grasping action are important reconstruction conditions. For some elements, only considering its representation in a certain dimension or low dimension can make an accurate judgment. For example, for the lighting effect element in the game scene, only the brightness dimension (low-dimensional representation) can be used to preliminarily judge whether it is abnormal, without the need to fully analyze multiple dimensions such as the color and direction of the light. In this way, under the strict constraint of the preset error probability, the reconstruction conditions that can improve the judgment efficiency without affecting the accuracy are accurately excavated.

[0046] Finally, based on these reconstruction conditions, supervised training is performed on the image recognition module. During the training process, the parameters of the image recognition module are continuously adjusted so that it can better focus on these key reconstruction conditions for judgment. For example, when training the image recognition module to identify the action status of the game task, it focuses on the feature extraction and analysis of the key positions of the hand; when judging the lighting effect, it optimizes the sensitivity to the brightness dimension features. Through this targeted training, the image recognition module can quickly and effectively identify errors in the game interface while ensuring accuracy, thereby improving the performance of the entire error detection system and providing strong guarantees for the stable operation of the game and the high-quality experience of players.

[0047] In a possible implementation, the comprehensive determination module further includes: A detection period is set, wherein the detection period is limited based on a preset determination time zone.

[0048] Detect and determine the updated error detection data of the single frame rate image and store it in a temporary database.

[0049] With the detection period as a constraint, as the updated error detection data of the single frame rate image is stored, the error detection data of the failed frame rate image is deleted.

[0050] Specifically, in order to efficiently and orderly manage error detection data, it is necessary to set a detection cycle, and this detection cycle is limited based on a preset judgment time zone. The preset judgment time zone is a time range determined based on the characteristics of the game, the frequency of operation, and the comprehensive consideration of the timeliness and accuracy of error detection. For example, for a fast-paced action game, a shorter judgment time zone is required, such as a judgment time zone every 0.1 seconds, because the actions and scenes in the game change rapidly, and errors that affect the game experience may occur in a short period of time; for a strategy game, the judgment time zone can be appropriately extended, such as a judgment time zone every 1 second, because its operations and interface changes are relatively less frequent.

[0051] In the complex environment where the game is running, meticulous error detection is required for the updated single-frame images to ensure the smoothness and stability of the game experience. When a new single-frame image is generated, the detection process is immediately started. This process covers in-depth analysis of each element in the image, including the first-class elements of the direct discrimination class and the second-class elements with game scene constraints. For the first-class elements, check whether the icons on the interface are complete and clear, whether the colors are normal, whether the shape and position of the buttons are accurate, and whether the text display is garbled or blurred. For example, if the skill icons in the game have missing pixels or distorted colors, the detection system can quickly capture these anomalies. For the second-class elements, the detection will be based on the rules and logic of the game scene. For example, in the battle scene, whether the character's action sequence conforms to the setting of skill release, whether the action feedback of attack and defense is normal, and whether the position relationship and interaction between different characters conform to the game mechanism. Once the comprehensive detection of the updated single-frame image is completed, all detected error information will be integrated into error detection data. These data carry key information such as the type, location, and severity of the error in the frame rate image. Subsequently, this valuable data will be stored in a temporary database, which is specifically used to store these continuously updated single-frame image error detection data, providing data support for subsequent further analysis, processing, and possible error repair prompts, ensuring that game developers can promptly grasp potential problems during the game operation.

[0052] When new single-frame images are continuously generated and error detection is completed, the corresponding error detection data is continuously stored in the temporary database. As this process continues, the timeliness of each data needs to be examined based on the detection cycle, and the error detection data of invalid frame images that are outdated and have lost their current reference value should be deleted in a timely manner. For example, in a game with fast action scenes, the detection cycle is set to a shorter time interval. As the game screen switches quickly, the error detection data of the frame image data of a previous short action moment may no longer be relevant after the new scene and action unfold. If not deleted in time, these invalid data will gradually accumulate, occupy a large amount of storage space, and may also interfere with the subsequent processing and analysis of valid data. For example, in a game with multi-stage tasks, when the player completes a stage and enters the next stage, the error detection data of some frame images in the previous stage is meaningless to the current stage. At this time, based on the detection cycle, these invalid data are automatically identified and cleaned up to ensure that the temporary database always maintains an efficient storage state, so that resources can be concentrated on processing and analyzing valuable error detection data in the current game state, thereby improving the operating efficiency and accuracy of the entire error detection mechanism.

[0053] In a possible implementation, the comprehensive determination module further includes: As the temporary database is updated, the error detection data within the detection period is subjected to element error determination under the time zone to determine continuous detection data.

[0054] A pop-up window alert is displayed for the continuous detection data.

[0055] Specifically, the update of the temporary database is a continuous and dynamic process. As new error detection data continues to flow into the temporary database, the error judgment of the elements in the time zone is performed on these data within the detection cycle. This judgment process conducts in-depth analysis of the error detection data within each specific time interval (i.e., the time zone defined by the detection cycle). In this time zone, for each element in the error detection data, whether it is a first-class element related to the visual presentation elements of the game interface, such as icons, text, colors, etc., or a second-class element related to the game scene and character behavior, strict error judgment is performed. For example, for a certain action game, within a specific detection cycle, the second-class elements related to the release of the character's skills are checked to see whether the display of its skill special effects is complete and whether the action coherence of the skill release conforms to the game settings; at the same time, the first-class elements such as the skill icons on the interface are also checked for display abnormalities. Through this comprehensive and detailed judgment, the error detection data with correlation in the same time zone are integrated to determine the continuous detection data.

[0056] These continuous detection data represent error messages that are consistent and relevant within a specific time period. When these continuous detection data are determined, a pop-up warning mechanism will be activated. Pop-up warnings are an important part of interacting with players and game developers. They present error messages in a clear and eye-catching way. The pop-up window will list in detail the key content such as the error types contained in the continuous detection data, the frequency of occurrence, and the game links that may be affected, so that game developers can quickly locate the problem. At the same time, it also allows players to be aware of the abnormal situations in the current game, so that they can take corresponding measures, such as giving feedback to developers or waiting for repairs, to ensure the stability and smoothness of the game experience.

[0057] In a possible implementation, the comprehensive determination module further includes: The game state of the frame rate image is determined, and the error determination factors are weighted to determine the factor weights.

[0058] Based on the factor weights, weighted calculations are performed on the discrimination output results based on the first-class factors and the second-class factors to determine a comprehensive discrimination result.

[0059] A preset error tolerance interval is set, and the element error exceeding the limit is judged on the discrimination output result. If there is an element error exceeding the limit, or the comprehensive discrimination result is abnormal, a pop-up warning will be issued.

[0060] Specifically, the game state of the frame rate image is judged. This process requires in-depth analysis of the current situation of the game, such as combat scenes, exploration scenes, plot dialogue scenes, or other specific game links. The importance of error judgment elements varies in different game states. Based on this understanding, the error judgment elements are weighted to determine the element weights. For example, in a combat scene, the accuracy of character skill release and the effectiveness of enemy attacks have higher weights, because they directly affect the game's combat results and player experience; while the weights of some decorative icons on the interface and other elements are relatively low.

[0061] The factor weight plays a key role. Based on the determined factor weight, the weighted calculation of the first-class factor and the second-class factor discrimination output results is carried out. The first-class factor involves the part directly presented in the game interface, such as whether the various icons are clear and complete, whether the text display is wrong, whether the color is normal, etc. The discrimination output results of these factors each have their corresponding weights. At the same time, the second-class factor is closely related to the game scene, such as whether the character's behavior in a specific scene is logical, whether the change of the scene elements follows the set rules, etc., and its discrimination output results also have corresponding weights. When performing the weighted calculation, the influence of each factor is measured, and for each first-class factor discrimination result, it is multiplied by its corresponding weight to obtain the weighted result. Similarly, the discrimination results of the second-class factor are also operated in this way. Then, all the weighted first-class factor results and second-class factor results are summarized, and the importance of each factor in the entire game interface error detection is comprehensively considered. After such a rigorous calculation process, the comprehensive discrimination result is finally determined. This result can comprehensively and accurately reflect whether there is an error in the game interface under the current frame rate image and the severity of the error, providing a reliable basis for subsequent processing.

[0062] Setting a preset error tolerance interval is an important part of ensuring the stability and rationality of the system. This preset error tolerance interval is carefully determined based on the impact on the gaming experience and defines the acceptable error range. After obtaining the judgment output result, the error limit of the element is determined based on this preset error tolerance interval. For each element output result, carefully check whether it exceeds the set tolerance limit. For example, for the element of the game character skill cooldown time, if its normal range is 5-10 seconds, and the preset error tolerance interval is set to fluctuate by 1 second, then when the judgment output result shows that the skill cooldown time is 4 seconds or 11 seconds, it is necessary to further determine whether it exceeds the error range allowed by this element. At the same time, we must also pay attention to the comprehensive judgment result. The comprehensive judgment result is an overall judgment after comprehensive consideration of all elements. If this result shows an abnormality, it means that the game interface as a whole may have problems that affect the gaming experience. Whether it is an error limit of a certain element or an abnormality in the comprehensive judgment result, the pop-up warning function will be immediately activated. As a key way to communicate with players and developers, pop-up alerts will clearly and prominently display error information, including which specific element has a problem, the extent of the error, and the possible impact on the game. This allows players to know the game status in a timely manner, and also facilitates developers to quickly locate and solve problems to ensure the normal operation of the game.

[0063] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0064] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0065] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A game interface error detection platform based on image recognition technology, characterized in that: The platform includes: A training sample integration module, the training sample integration module is used to integrate training samples based on error discrimination factors for the dynamic frame rate of the game, the error discrimination factors include a first type of factors of the direct discrimination class and a second type of factors with game scene constraints; A discriminator construction module, wherein the discriminator construction module is based on the adversarial training principle, combines the training samples, constructs a point discriminator for the first type of elements, constructs a sequence discriminator for the second type of elements, and generates an error detection module, wherein the sequence discriminator has a built-in discriminant transfer function that obeys the scene transfer law; A discriminant factor matrix determination module, the discriminant factor matrix determination module is used to receive a frame rate image of a game interface, assist the image recognition module, execute the error discriminant factor to perform factor extraction and factor reconstruction, and determine a discriminant factor matrix, wherein the factor reconstruction is guided by the detection requirements; A comprehensive discrimination module is used to transmit the discrimination factor matrix to the error detection module, perform classification discrimination and comprehensive discrimination, generate error detection data of the frame rate image and issue a pop-up window warning.

2. The game interface error detection platform based on image recognition technology as claimed in claim 1, characterized in that: The discriminator building module also includes: Determine a target training scene, wherein the target training scene is any game scene; Limiting the target training scenario, combining the training samples based on the two types of elements, performing adversarial training under the training architecture, and generating an initialization architecture, wherein the training architecture consists of a generator and a discriminator; The initialization architecture is disassembled to obtain a trained sequence discriminator.

3. The game interface error detection platform based on image recognition technology as claimed in claim 2, characterized in that: The discriminator building module also includes: Identify training samples of the two types of elements, perform scene clustering processing, and determine multiple scene clusters; Traversing the plurality of scene clusters to determine a cluster discrimination probability, wherein the cluster discrimination probability comprises a mean value of element discrimination probabilities in each scene cluster; Taking the target training scene as a baseline and the cluster discrimination probability as a basis, determining a probability migration relationship under scene transfer; The probability transition relationship is used as the discriminant transition function.

4. The game interface error detection platform based on image recognition technology as claimed in claim 2, characterized in that: The discriminator building module also includes: Determining a factor threshold based on the error discrimination criterion; Using the element threshold as a constraint, dividing positive samples and negative samples, wherein the positive samples are the detected false detection classes; Combined with the training samples of the two types of elements, the positive samples and the negative samples are used as the generation targets of the generator, and the element threshold is used as the discrimination criterion. The alternating iterative training of the generator and the discriminator is performed until convergence to obtain the trained discriminator, wherein the convergence condition is that the discriminator satisfies the preset discrimination accuracy.

5. The game interface error detection platform based on image recognition technology as claimed in claim 4, characterized in that: The discriminant factor matrix determination module also includes: Setting a preset error probability, wherein the preset error probability limits the consistency of determination before and after reconstruction; Taking the preset error probability as a constraint and the determination based on the element threshold as a detection requirement, mining the reconstruction conditions of the error determination element; Based on the reconstruction condition, the image recognition module is supervised and trained.

6. The game interface error detection platform based on image recognition technology as claimed in claim 1, characterized in that: The comprehensive identification module also includes: Setting a detection period, wherein the detection period is limited based on a preset determination time zone; Detect and determine the error detection data of the updated single frame rate image, and store it in a temporary database; With the detection period as a constraint, as the updated error detection data of the single frame rate image is stored, the error detection data of the failed frame rate image is deleted.

7. The game interface error detection platform based on image recognition technology as claimed in claim 6, characterized in that: The comprehensive identification module also includes: As the temporary database is updated, element error determination is performed on the error detection data within the detection period under the time zone to determine continuous detection data; A pop-up window alert is displayed for the continuous detection data.

8. The game interface error detection platform based on image recognition technology as claimed in claim 1, characterized in that: The comprehensive identification module also includes: Determine the game state of the frame rate image, assign weights to the error determination factors, and determine the factor weights; Based on the factor weights, weighted calculation is performed on the discrimination output results based on the first-class factors and the second-class factors to determine a comprehensive discrimination result; A preset error tolerance interval is set, and the element error exceeding the limit is judged on the discrimination output result. If there is an element error exceeding the limit, or the comprehensive discrimination result is abnormal, a pop-up warning will be issued.

Images