Data processing method and device, computer device, and storage medium

Abstract

The application discloses a data processing method and device, computer equipment and a storage medium. The method comprises the following steps: determining a support sample image of a to-be-detected image from a preset sample set, the support sample image being an image subjected to action recognition processing; performing transformation processing on original feature information of the support sample image according to support feature information of the support sample image, to obtain reinforced feature information, the support feature information being feature information contained in an action recognized by the image subjected to the action recognition processing, and the original feature information being feature information obtained by performing feature extraction on the image; and performing action recognition processing on the to-be-detected image according to the reinforced feature information and original feature information of the to-be-detected image, to recognize a target action in the to-be-detected image. By using the method, new actions that cannot be learned in the to-be-recognized image can be better recognized under the condition of limited data.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a data processing method, apparatus, computer equipment, and storage medium. Background Technology

[0002] With the development of the internet, more and more people are choosing to learn online, using online video courses to learn content they are interested in, such as online fitness and online yoga. Because these courses involve physical movements, although people can learn the general movements, they may not be able to grasp the details of the movements well, which may lead to poor results or physical injuries due to improper movements.

[0003] Therefore, there is a need for a way to accurately identify human behavior and actions. The conventional approach is to build a recognition model to identify human behavior and actions. However, due to the limited number of samples of detailed actions, the available sample data is limited, which makes it difficult to identify detailed actions or new actions well during the learning process. This results in unsatisfactory behavior and action recognition performance and reduces the accuracy of action recognition. Summary of the Invention

[0004] This application provides a data processing method, apparatus, computer equipment, and storage medium to solve the technical problem of low recognition accuracy in existing motion recognition technologies.

[0005] In a first aspect, embodiments of this application provide a data processing method, which includes:

[0006] Supporting sample images are determined from a preset sample set; the supporting sample images are images that have undergone action recognition processing.

[0007] Determine the enhancement feature information of the supporting sample image;

[0008] Based on the enhanced feature information and the original feature information of the image to be detected, action recognition processing is performed on the image to be detected to identify the target action in the image to be detected.

[0009] Secondly, embodiments of this application provide a data processing apparatus, which includes:

[0010] The first determining module is used to determine supporting sample images of the image to be detected from a preset sample set; the supporting sample images are images that have undergone action recognition processing.

[0011] The second determining module is used to determine the enhancement feature information of the supporting sample image;

[0012] The recognition module is used to perform action recognition processing on the image to be detected based on the enhanced feature information and the original feature information of the image to be detected, so as to identify the target action in the image to be detected.

[0013] Thirdly, embodiments of this application provide a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the data processing method described in the first aspect above.

[0014] Fourthly, embodiments of this application also provide a storage medium storing a computer program that, when executed by a processor, causes the processor to perform the data processing method described in the first aspect.

[0015] This application provides a data processing method, apparatus, computer device, and storage medium. By determining support sample images for the image to be detected, and by enhancing action features in the support sample images while weakening features other than action features, enhanced feature information is obtained. Then, based on the enhanced feature information and the original feature information obtained by feature extraction from the image to be detected, action recognition processing is performed on the image to be detected. This allows for better identification of new actions in the image to be recognized that cannot be learned under limited data conditions, thus solving the technical problem of low recognition accuracy in existing action recognition technologies. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A schematic flowchart illustrating a data processing method provided in an embodiment of this application;

[0018] Figure 2 A flowchart illustrating the steps for obtaining a supporting sample image according to an embodiment of this application;

[0019] Figure 3 A flowchart illustrating the steps of model training provided in an embodiment of this application;

[0020] Figure 4 A flowchart illustrating the steps of a training exercise provided in one embodiment of this application;

[0021] Figure 5A schematic block diagram of a data processing apparatus provided in an embodiment of this application;

[0022] Figure 6 A schematic block diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0025] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0026] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0027] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating a data processing method provided in an embodiment of this application. The method includes steps S101 to S103.

[0028] S101. Determine the supporting sample images of the image to be detected from the preset sample set.

[0029] Specifically, all sample images in the sample set, including supporting sample images, are images that have undergone action recognition processing. In particular, all sample images in the sample set are images whose actions have been recognized by a trained action recognition model. The feature information contained in the actions identified by the action recognition model in these sample images enables the action recognition model to accurately identify the actions in the sample images.

[0030] Reference Figure 2 , Figure 2This is a flowchart illustrating the steps for obtaining a supporting sample image according to an embodiment of the present application, wherein the steps include steps S201 to S203.

[0031] Step S201: Obtain the first encoding matrix corresponding to each sample image in the sample set and the second encoding matrix corresponding to the image to be detected.

[0032] In this embodiment, since even images of the same category may exhibit significant differences in certain parts during practical applications, after feature extraction for each sample image, the extracted encoding matrix undergoes further processing to better focus on the differences. For example, an attention module can be used to process the encoding matrix to obtain the corresponding feature matrix. Specifically, the step of obtaining the first encoding matrix corresponding to each sample image in the sample set and the second encoding matrix corresponding to the image to be detected involves: obtaining the first initial encoding matrix corresponding to each sample image in the sample set and the second initial encoding matrix corresponding to the image to be detected; and inputting the first initial encoding matrix and the second initial matrix into the attention module to obtain the first encoding matrix corresponding to each sample image in the sample set and the second encoding matrix corresponding to the image to be detected.

[0033] Step S202: Calculate the similarity between the first encoding matrix and the second encoding matrix.

[0034] Step S203: Based on the similarity, determine the supporting sample images of the image to be detected from the sample set.

[0035] Specifically, features are extracted for each sample image in the sample set to obtain the feature information corresponding to each sample image. Then, the attention module is used to filter the feature information corresponding to each sample image to obtain the encoding matrix corresponding to each sample image. Then, the similarity value between the encoding matrix corresponding to the image to be detected and the encoding matrix of each sample image is calculated. Based on the obtained similarity value, the supporting sample image corresponding to the image to be detected is determined in the sample set.

[0036] When determining the supporting sample images corresponding to the image to be detected, further filtering is performed by calculating the similarity value between feature matrices. For example, the image to be detected X... tar The characteristic matrix is ​​x attention,tar The feature matrix of the sample images in the sample set is x attention,j Where tar ≠ j, then calculate x. attention,tar With x attention,j The similarity value is used to determine the supporting sample images corresponding to the image to be detected. The similarity value can be calculated as follows:

[0037] sim(x attention,tar ,x attention,j )=σ(x attention,tar T x attention,j )

[0038] Where σ is the activation function.

[0039] It should be noted that when obtaining supporting sample images based on the obtained similarity values, they are compared with a corresponding preset threshold. For example, if the preset threshold is set to 0.85, then under this condition, when the image to be detected X... tar If the similarity between the image to be detected and the sample image in the sample set is greater than 0.85, it is used as a supporting sample image. Otherwise, if the similarity between the image to be detected and the sample image in the sample set is less than 0.85, the most similar sample image is selected as the supporting sample image.

[0040] Step S102: Determine the enhancement feature information of the supporting sample image.

[0041] In this embodiment, when identifying feature information in a sample image, it is necessary to consider not only the information contained in the sample image but also the feature information of the actions already identified in the supporting sample image. Therefore, to better obtain the feature information contained in the identified actions in the supporting sample set, the feature information of the sample image is made more intuitive by enhancing the target feature information and weakening the information of other non-target features. The target feature information mainly refers to the feature information of the identified actions in the sample image, while the non-target feature information mainly refers to the feature information in the sample image other than the identified actions. Specifically, the step of determining the enhanced feature information of the supporting sample image includes: obtaining the feature matrix of the supporting sample image; performing a transformation on the feature matrix to obtain the supporting feature information of the supporting sample image; and performing a linear transformation on the supporting feature information and the original feature information of the supporting sample image to obtain the enhanced feature information.

[0042] The supporting feature information refers to the feature information contained in the action identified by the image after action recognition processing, and the original feature information refers to the feature information obtained by feature extraction of the image.

[0043] It should be noted that this embodiment superimposes the feature information of two images, allowing identical target features to be superimposed while preventing the superposition of dissimilar non-target features. This enhances the target features and weakens the non-target features. In other words, when the feature information of two images is superimposed, the target features are enhanced, while the non-target features are weakened because they are not superimposed.

[0044] To weaken non-target feature information in the support sample image, this embodiment defines it as noise and then weakens the non-target feature information through noise removal. Specifically, to eliminate noise, this embodiment superimposes the support feature information and the original feature information of the support sample image. Specifically, the feature information corresponding to the actions contained in the support sample image is used as the superimposed feature to enhance and eliminate noise in the original feature information of the support sample image, thereby obtaining feature information that includes enhanced target feature information and weakened non-target information, i.e., the enhanced feature information.

[0045] For example, the transformation process of the feature matrix can be performed as follows:

[0046]

[0047] K = ∑ i∈Ω L(y i )y i

[0048] Among them, y i Let W represent the feature matrix of the i-th support sample image, Ω represent the index of the support sample image in the corresponding support sample set, and W represent the feature matrix of the i-th support sample image. e These are the hyperparameters that need to be learned, and K represents the supporting feature information corresponding to the supporting sample images.

[0049] After obtaining the supporting feature information corresponding to the supporting sample images, a linear transformation can be performed on the original feature information of the supporting sample images based on the obtained supporting feature information to obtain the enhanced feature information. The method for performing the linear transformation on the original feature information of the supporting sample images based on the obtained supporting feature information can be as follows:

[0050]

[0051] in, To enhance feature information, Y represents the original feature information of the supporting sample image, LeakyReLu is a non-linear function, and α is the parameter to be learned.

[0052] Step S103: Based on the enhanced feature information and the original feature information of the image to be detected, perform action recognition processing on the image to be detected to identify the target action in the image to be detected.

[0053] In this embodiment, to enhance feature information to aid in the identification of target actions in the image to be detected, the enhanced feature information is combined with the original feature information of the image to be detected. Thus, by performing action recognition processing on the image to be detected using the combined target feature information, the actions of the person in the image to be detected can be better identified. Specifically, the step of performing action recognition processing on the image to be detected based on the enhanced feature information and the original feature information of the image to be detected to identify the target action in the image to be detected includes: linearly adding the enhanced feature information and the original feature information of the image to be detected to obtain target feature information; and performing action recognition processing on the image to be detected based on the target feature information to identify the target action in the image to be detected.

[0054] As an optional embodiment, when combining the enhanced feature information with the original feature information of the image to be detected, it is necessary not only to consider the similarity between the image to be detected and the supporting sample image, but also to consider the information of the supporting sample image itself. The specific combination method is as follows:

[0055]

[0056]

[0057] Where, p i It is the information in the support sample image calculated based on the same position i in the image to be detected. The coefficients are calculated using linear transformation based on the degree of similarity and information from supporting sample images.

[0058] Specifically, for support sample images, since there may be multiple sample images with similarity values ​​greater than the preset threshold, it is necessary to average all the information contained in multiple support sample images to obtain the total information. The specific calculation method is as follows:

[0059]

[0060] Where K is the number of supporting sample images.

[0061] It should be noted that, in order to better utilize the feature information contained in the actions already identified in the supporting sample images, this embodiment also needs to use an action recognition model that has undergone action recognition processing on the supporting sample images to perform action recognition processing on the image to be detected. Specifically, the step of performing action recognition processing on the image to be detected based on the target feature information to identify the target action in the image to be detected includes: obtaining an action recognition model that has undergone action recognition processing on the supporting sample images; and performing action recognition processing on the image to be detected based on the action recognition model and the target feature information to identify the target action in the image to be detected.

[0062] As an optional embodiment, before performing action recognition processing on the image to be recognized using the action recognition model and target feature information, it is necessary to first obtain the action recognition model, which is obtained through pre-training. For details, please refer to... Figure 3 , Figure 3 This is a flowchart illustrating the steps of model training according to an embodiment of the present application, wherein the steps include steps S301 to S303.

[0063] Step S301: Obtain several video data.

[0064] Each video data point corresponds to a set of action procedures.

[0065] Step S302: Take each frame of each video data as a sample image to obtain an action sample set.

[0066] It should be noted that since each video data corresponds to a set of action procedures, each frame of each video data contains the corresponding action, so each frame of the video data can be directly used as a sample image.

[0067] Step S303: Train the action recognition model to be trained according to the action sample set, and enable the action recognition model to be trained to recognize the action in each of the sample images, so as to obtain the trained action recognition model.

[0068] For example, when constructing an action sample set, it can be constructed based on different categories of video data. For the use case of this action recognition model, when acquiring video data, the corresponding sample images are obtained by acquiring video frames in the video. For example, when constructing an action sample set, it is possible to collect m types of videos, where each type of video embodies a set of action processes. By acquiring each frame image contained in each video to construct an image set, after completing the processing of all types of videos, the image set of m types is obtained, which is the action sample set.

[0069] Meanwhile, when the action sample set is obtained, the action in each sample image in the action sample set is labeled. In this way, the action recognition model to be trained can be trained based on the sample images and the corresponding labels.

[0070] In some embodiments, when training the action recognition model to be trained based on the action sample set, the action sample set needs to be divided into a training set and a test set, thereby completing the training of the action recognition model to be trained based on the training set and the test set. Specifically, when dividing the action sample set to obtain the training set and the test set, the sample images corresponding to each category of image set are divided. Specifically, a certain number of sample images are randomly selected from the image set of one category as the training set, and other sample images are selected from the image set of that category as the test set. For example, if the image set of one category has 100 sample images, 70 sample images can be selected as the training set, and then N sample images can be selected from the remaining 30 sample images as the test set, where N is less than or equal to 30. When determining the training set and the test set, it is only necessary to ensure that there is no overlap between the training set and the test set. The number of sample images contained in the training set and the test set is not limited, but usually, the number of sample images in the training set is greater than the number of sample images in the test set.

[0071] During training, it is necessary to determine whether the model has completed training. Upon completion, the model is tested to determine if it meets the usage requirements. Specifically, during training, a loss is obtained for each training iteration. Therefore, by analyzing the loss, it is possible to determine when training based on the training set is complete. Then, upon determining that training is complete, the trained action recognition model is tested using the test set. If the test passes, the trained action recognition model is stored.

[0072] For example, when performing detection on a trained action recognition model based on a test set, the model's prediction accuracy is used to determine whether the model has passed the detection.

[0073] In addition, it is necessary to determine when the model training is completed before conducting detection. Determining when the training is completed includes: when it is determined that a training session has ended, obtaining the loss values ​​corresponding to several training sessions adjacent to the current training session; and determining whether the training is complete based on the loss values ​​corresponding to the several training sessions.

[0074] Specifically, during the training process, a judgment is made after each training session to determine whether training is complete. When making the judgment, the loss values ​​corresponding to several training sessions adjacent to the current training session are obtained, and then the completion of training is determined based on the obtained loss values.

[0075] For example, when determining whether training is complete based on several obtained loss values, based on the characteristics of model training, the more stable the final model loss during training, the more stable the model is. Therefore, when determining whether training is complete based on several loss values, the volatility of model training is determined based on the loss values, and then the volatility is used to determine whether training is complete. This includes: calculating the volatility value based on the loss values ​​corresponding to several training sessions, and comparing the volatility value with a preset threshold; if the volatility value is less than or equal to the preset threshold, training is determined to be complete; if the volatility value is greater than the preset threshold, training is determined to be incomplete.

[0076] There are many methods for calculating volatility, such as range, variance, and standard deviation. Any method can be used here. Taking variance calculation as an example, the variance corresponding to several loss values ​​is calculated, and then the obtained variance is compared with a set threshold. The comparison determines whether training is complete. The smaller the variance, the more stable the data. Therefore, training is considered complete when the obtained variance is less than or equal to the set threshold; otherwise, training is considered incomplete.

[0077] In one embodiment, after training the action recognition model to be trained based on the training set to obtain the trained action recognition model, it is also necessary to test it to ensure the accuracy of the training. Specifically, after obtaining the trained action recognition model, the method further includes: testing the trained action recognition model based on the test set; storing the trained action recognition model when the test is determined to be passed; and adjusting the trained action recognition model when the test is determined to be failed.

[0078] Specifically, when testing the trained action recognition model using a test set, if the test fails, the model's structure will be adjusted accordingly, and the model will be retrained. Adjustments to the model's structure include, but are not limited to, adding modules or changing the number of layers. For example, adding a new convolutional module or adjusting the structure of existing convolutional modules, such as changing from 3x3 to 4x4. After adjusting the model's structure, it will be trained again based on the training set until a usable action recognition model is obtained.

[0079] Reference Figure 4 , Figure 4 This is a flowchart illustrating the steps of a training exercise provided in one embodiment of this application.

[0080] During training, a set of action samples is first acquired, consisting of images from each frame of video data from several different categories. The sample set is then randomly divided based on the category to obtain a training set and a test set. Next, an image is randomly selected from the training set as the image to be detected and used as the training sample for the first training iteration. Simultaneously, the similarity between the image to be detected and the images in the training set is calculated, such as by calculating the similarity value between the images. Based on this similarity, a support set corresponding to the image to be detected is constructed. The support set contains at least one support sample image. Then, features are extracted from the support sample images in the obtained support set to obtain the corresponding support features. Finally, based on the obtained support features, target feature enhancement and noise reduction are performed on the images in the support set to generate new support set images containing enhanced features.

[0081] After obtaining a new support set of images containing enhanced features, the feature matrix of the new support set images is acquired, along with the feature matrix of the image to be detected. These two matrices are linearly added to combine the feature information. Finally, the combined feature information is used to predict and recognize the action in the image to be detected. When the action in the image to be detected is recognized, one training iteration of the model is completed. At this point, the model parameters are adaptively updated based on the loss between the recognition result and the corresponding labels of the training samples. Furthermore, the training loss value is obtained after each training iteration to determine when model training is complete.

[0082] After many such Figure 4 The training steps shown are followed, and the trained action recognition model can be obtained after the model parameters converge.

[0083] Thus, after obtaining the trained action recognition model, the action recognition process of the image to be detected can be completed by using the trained action recognition model and the target feature information obtained by linearly adding the enhanced feature information of the supporting sample image and the original feature information of the image to be detected. This enables better accurate recognition of new actions appearing in the image to be detected under the condition of fewer samples, ensuring that the details of each action are identified and better recognizing the human body actions appearing in the image to be detected.

[0084] In summary, the data processing method provided in this application determines the supporting sample images of the image to be detected, and strengthens the action features in the supporting sample images while weakening the features other than action features in the supporting sample images to obtain enhanced feature information. Then, based on the enhanced feature information and the original feature information obtained by feature extraction of the image to be detected, action recognition processing is performed on the image to be detected. This allows for better identification of new actions in the image to be recognized that cannot be learned under limited data conditions, thus solving the technical problem of low recognition accuracy in the prior art.

[0085] This application also provides a data processing apparatus for executing any embodiment of the aforementioned data processing method. Specifically, please refer to... Figure 5 , Figure 5 This is a schematic block diagram of a data processing apparatus provided in an embodiment of this application. The data processing apparatus 500 provided in this embodiment includes:

[0086] The first determining module 501 is used to determine supporting sample images of the image to be detected from a preset sample set; the supporting sample images are images that have undergone action recognition processing.

[0087] The second determining module 502 is used to determine the enhancement feature information of the supporting sample image.

[0088] The recognition module 503 is used to perform action recognition processing on the image to be detected based on the enhanced feature information and the original feature information of the image to be detected, so as to identify the target action in the image to be detected.

[0089] In specific implementation, the above modules and / or units can be implemented as independent entities, or they can be arbitrarily combined and implemented as the same or several entities. For the specific implementation of the above modules and / or units, please refer to the previous method embodiments. For the specific beneficial effects that can be achieved, please also refer to the beneficial effects in the previous method embodiments, which will not be repeated here.

[0090] Please see Figure 6 , Figure 6 A schematic block diagram of a computer device provided in an embodiment of this application.

[0091] See Figure 6 The computer device 600 includes a processor 602, a memory, and a network interface 605 connected via a system bus 601. The memory may include a non-volatile storage medium 603 and internal memory 604.

[0092] The non-volatile storage medium 603 may store an operating system 6031 and a computer program 6032. When the computer program 6032 is executed, it causes the processor 602 to perform a data processing method.

[0093] The processor 602 provides computing and control capabilities to support the operation of the entire computer device 600.

[0094] The internal memory 604 provides an environment for the execution of the computer program 6032 in the non-volatile storage medium 603. When the computer program 6032 is executed by the processor 602, the processor 602 can perform data processing methods.

[0095] This network interface 605 is used for network communication, such as providing data transmission. Those skilled in the art will understand that... Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 600 to which the present application is applied. The specific computer device 600 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0096] The processor 602 is used to run the computer program 6032 stored in the memory to implement any process step in the data processing method provided in this embodiment.

[0097] Those skilled in the art will understand that Figure 6 The embodiments of the computer device shown do not constitute a limitation on the specific configuration of the computer device. In other embodiments, the computer device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, in some embodiments, the computer device may include only memory and a processor. In such embodiments, the structure and function of the memory and processor are different from those shown. Figure 6 The embodiments shown are consistent and will not be repeated here.

[0098] It should be understood that in the embodiments of this application, the processor 602 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0099] In another embodiment of this application, a storage medium is provided. This storage medium may be a non-volatile, computer-readable storage medium. The storage medium stores a computer program, which, when executed by a processor, implements the aforementioned data processing method.

[0100] Those skilled in the art will readily understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Units with the same function may be grouped into one unit. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, or it may be an electrical, mechanical, or other form of connection.

[0102] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0103] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0104] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks.

[0105] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A data processing method, characterized in that, The method includes: Supporting sample images are determined from a preset sample set; the supporting sample images are images that have undergone action recognition processing. Determine the enhancement feature information of the supporting sample image; Based on the enhanced feature information and the original feature information of the image to be detected, action recognition processing is performed on the image to be detected to identify the target action in the image to be detected; The determination of the enhancement feature information of the supporting sample image includes: Obtain the feature matrix of the supporting sample image; The feature matrix is ​​transformed to obtain the support feature information of the support sample image; Linear transformation is performed on the supporting feature information and the original feature information of the supporting sample image to obtain enhanced feature information; the supporting feature information is the feature information contained in the action identified by the action recognition processing of the image, and the original feature information is the feature information obtained by feature extraction of the image; The step of performing action recognition processing on the image to be detected based on the enhanced feature information and the original feature information of the image to be detected, in order to identify the target action in the image to be detected, includes: The enhanced feature information and the original feature information of the image to be detected are linearly added together to obtain the target feature information; The target feature information is used to perform action recognition processing on the image to be detected in order to identify the target action in the image to be detected.

2. The method according to claim 1, characterized in that, The step of determining supporting sample images for the image to be detected from a preset sample set includes: Obtain the first encoding matrix corresponding to each sample image in the sample set and the second encoding matrix corresponding to the image to be detected; Calculate the similarity between the first encoding matrix and the second encoding matrix; Based on the similarity, supporting sample images for the image to be detected are determined from the sample set.

3. The method according to claim 2, characterized in that, The step of obtaining the first encoding matrix corresponding to each sample image in the sample set and the second encoding matrix corresponding to the image to be detected includes: Obtain the first initial encoding matrix corresponding to each sample image in the sample set and the second initial encoding matrix corresponding to the image to be detected; The first initial encoding matrix and the second initial matrix are respectively input into the attention module to obtain the first encoding matrix corresponding to each sample image in the sample set and the second encoding matrix corresponding to the image to be detected.

4. The method according to claim 1, characterized in that, The step of performing action recognition processing on the image to be detected based on the target feature information to identify the target action in the image to be detected includes: Obtain the action recognition model that performs action recognition processing on the supported sample images; Based on the action recognition model and the target feature information, action recognition processing is performed on the image to be detected to identify the target action in the image to be detected.

5. The method according to claim 4, characterized in that, Prior to the step of obtaining the action recognition model for completing action recognition processing on the supporting sample image, the method further includes: Acquire several video data sets; each video data set corresponds to a set of action procedures; Each frame of each video data is used as a sample image to obtain an action sample set; The action recognition model to be trained is trained based on the action sample set, and the action recognition model to be trained is made able to recognize the action in each of the sample images, thus obtaining the trained action recognition model.

6. A data processing apparatus, characterized in that, The device includes: The first determining module is used to determine supporting sample images of the image to be detected from a preset sample set; the supporting sample images are images that have undergone action recognition processing. The second determining module is used to determine the enhancement feature information of the supporting sample image; The recognition module is used to perform action recognition processing on the image to be detected based on the enhanced feature information and the original feature information of the image to be detected, so as to identify the target action in the image to be detected; The step of determining the enhancement feature information of the supporting sample image includes: Obtain the feature matrix of the supporting sample image; The feature matrix is ​​transformed to obtain the support feature information of the support sample image; Linear transformation is performed on the supporting feature information and the original feature information of the supporting sample image to obtain enhanced feature information; the supporting feature information is the feature information contained in the action identified by the action recognition processing of the image, and the original feature information is the feature information obtained by feature extraction of the image; The step of performing action recognition processing on the image to be detected based on the enhanced feature information and the original feature information of the image to be detected, in order to identify the target action in the image to be detected, includes: The enhanced feature information and the original feature information of the image to be detected are linearly added together to obtain the target feature information; The target feature information is used to perform action recognition processing on the image to be detected in order to identify the target action in the image to be detected.

7. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the data processing method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, causes the processor to perform the data processing method as described in any one of claims 1 to 5.

Images