Data processing method, device, electronic device and storage medium

Abstract

The embodiments of the present application provide a data processing method, device, electronic device and storage medium, the method comprising: providing a data interaction page to obtain input data to be recognized based on the data interaction page; determining a recognition result based on the input data and a trained character recognition model, the character recognition model is trained based on multiple groups of labeled training data, the multiple groups of training data are grouped based on the credibility of the input data input to the recognition model, and the labels of each group of training data are labeled based on a batch labeling method; issuing the recognition result; and improving the recognition accuracy of the character recognition model.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method, a data processing device, an electronic device and a storage medium. Background Art

[0002] Optical Character Recognition (OCR) refers to the process of using a character recognition model to examine characters in an image, determine their shapes by detecting dark and light patterns, and then use character recognition methods to translate the shapes into computer text.

[0003] Currently, character recognition models are usually trained using a small amount of manually annotated training data, resulting in low recognition accuracy of the character recognition models. Summary of the invention

[0004] An embodiment of the present application provides a data processing method to improve the recognition accuracy of a character recognition model.

[0005] Correspondingly, an embodiment of the present application also provides a data processing device, an electronic device and a storage medium to ensure the implementation and application of the above system.

[0006] In order to solve the above problems, an embodiment of the present application discloses a data processing method, which includes: providing a data interaction page to obtain input data to be recognized based on the data interaction page; determining a recognition result based on the input data and a trained character recognition model, wherein the character recognition model is trained based on multiple groups of labeled training data, and the multiple groups of training data are grouped based on the credibility of the input data input into the recognition model, and the annotation of each group of training data is obtained by batch annotation; and issuing the recognition result.

[0007] In order to solve the above problems, an embodiment of the present application discloses a data processing method, which includes: providing a data interaction page, the data interaction page including a data upload control; based on the triggering of the data upload control, obtaining the input data to be recognized and uploading it to a server, the server is used to determine the recognition result based on the character recognition model and the input data, the character recognition model is trained based on multiple groups of labeled training data, the multiple groups of training data are grouped based on the credibility of the input data input into the recognition model, and the annotation of each group of training data is obtained by annotating according to a batch annotation method; receiving the recognition result.

[0008] In order to solve the above problems, an embodiment of the present application discloses a data processing method, which includes: providing a data interaction page to obtain ancient character image data to be identified based on the data interaction page; determining a recognition result based on the ancient character image data and a trained character recognition model, wherein the character recognition model is trained based on multiple groups of annotated training data, and the multiple groups of training data are grouped based on the credibility of the ancient character image data input into the recognition model, and the annotation of each group of ancient character image data is obtained by batch annotation; and issuing the recognition result.

[0009] In order to solve the above problems, an embodiment of the present application discloses a data processing method, which includes: inputting unlabeled input data into a recognition model to determine the corresponding recognition result; grouping the input data according to the credibility of the recognition result; and configuring labeled data in batches for the input data of each group to form training data.

[0010] In order to solve the above problems, an embodiment of the present application discloses a data processing method, which includes: inputting unlabeled input data into a recognition model to determine the corresponding recognition result; grouping the input data according to the credibility of the recognition result; configuring labeled data in batches for the input data of each group to form training data; training a character recognition model based on the training data, so as to recognize the input data to be recognized based on the trained character recognition model.

[0011] In order to solve the above problems, an embodiment of the present application discloses a data processing method, which includes: providing a service page to obtain uploaded input data based on the service page; inputting the unlabeled input data into a recognition model to determine the corresponding recognition result; grouping the input data according to the credibility of the recognition result; and configuring labeled data for each group of input data in batches to form training data.

[0012] In order to solve the above problems, an embodiment of the present application discloses a data processing method, which includes: providing a service page, the service page including a data upload interface; based on the triggering of the data upload interface, obtaining input data and uploading it to a server, the server is used to input the unlabeled input data into a recognition model to determine the corresponding recognition result; and grouping the input data according to the credibility of the recognition result; sending the input data of each group; receiving the input data of each group sent by the server, and displaying it on the service page; receiving labeling information for the input data of each group to form training data.

[0013] In order to solve the above problems, an embodiment of the present application discloses a data processing device, including: a data interaction page providing module, used to provide a data interaction page to obtain input data to be recognized based on the data interaction page; an input data recognition processing module, used to determine the recognition result based on the input data and a trained character recognition model, the character recognition model is trained based on multiple groups of labeled training data, the multiple groups of training data are grouped according to the credibility of the input data input to the recognition model, and the annotation of each group of training data is obtained by batch annotation; a recognition result sending processing module, used to send the recognition result.

[0014] In order to solve the above problems, an embodiment of the present application discloses an electronic device, including: a processor; and a memory, on which executable code is stored. When the executable code is executed, the processor executes one or more methods described in the above embodiments.

[0015] In order to solve the above problems, the embodiments of the present application disclose one or more machine-readable media on which executable codes are stored. When the executable codes are executed, the processor executes one or more methods described in the above embodiments.

[0016] Compared with the prior art, the embodiments of the present application have the following advantages:

[0017] In an embodiment of the present application, unlabeled input data can be input into a recognition model to obtain corresponding recognition results, and the input data can be grouped according to the credibility of the recognition results. For the grouped input data, the labeling method corresponding to the grouping is used to configure labeling data for the input data of each group in batches, thereby forming training data, and the training data is used to train the character recognition model. Afterwards, the input data to be recognized is obtained based on the interactive page, and the trained character recognition model is used to recognize it, determine the recognition result, and issue the recognition result. In an embodiment of the present application, the input data can be grouped according to the credibility of the recognition result of the input data, and the input data in each group can be labeled in batches. Compared with the method of manually labeling each input data separately, the manual workload can be reduced, and more training data can be used to train the character recognition model, thereby improving the recognition accuracy of the character recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1A It is a flowchart of a data processing method according to an embodiment of the present application;

[0019] Figure 1B is a flowchart of a method for data grouping according to an embodiment of the present application;

[0020] Figure 2 is a flowchart of a data processing method according to another embodiment of the present application;

[0021] Figure 3 is a flowchart of a data processing method according to another embodiment of the present application;

[0022] Figure 4A is a flowchart of a data processing method according to another embodiment of the present application;

[0023] Figure 4B is a flowchart of a data processing method according to another embodiment of the present application;

[0024] Figure 5 is a flowchart of a data processing method according to another embodiment of the present application;

[0025] Figure 6 is a flowchart of a data processing method according to another embodiment of the present application;

[0026] Figure 7 is a flowchart of a data processing method according to another embodiment of the present application;

[0027] Figure 8 is a flowchart of a data processing method according to another embodiment of the present application;

[0028] Fig. 9 is a flowchart of a data processing method according to another embodiment of the present application;

[0029] Fig.10 is a structural schematic diagram of a data processing device according to an embodiment of the present application;

[0030] Fig.11 is a structural schematic diagram of a data processing device according to another embodiment of the present application;

[0031] Fig.12 is a structural schematic diagram of a data processing device according to another embodiment of the present application;

[0032] Fig.13 is a structural schematic diagram of a data processing device according to another embodiment of the present application;

[0033] Fig.14 is a structural schematic diagram of a data processing device according to another embodiment of the present application;

[0034] Fig.15 is a structural schematic diagram of a data processing device according to another embodiment of the present application;

[0035] Fig.16 It is a schematic diagram of the structure of an exemplary device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0036] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0037] The embodiments of the present application can be applied in the field of optical character recognition, wherein optical character recognition (OCR) refers to the process of using a character recognition model to check characters in an image, determine their shapes by detecting dark and light patterns, and then translate the shapes into computer text using a character recognition method. The character recognition model can also be called a recognition model, which can be understood as a neural network model (NN). The neural network model is a complex network system formed by a large number of simple processing units (called neurons) that are widely connected to each other. It is a highly complex nonlinear dynamic learning system.

[0038] In the embodiment of the present application, input data can be input into the recognition model to obtain recognition results, and the recognition results can be used to divide the input data into different groups, and different groups can be labeled in different ways. For example, the recognition results can be divided into three groups according to the credibility of the recognition results, and the credibility can be from high to low, namely the first group, the second group and the third group. For the first group, the recognition result of the recognition model can be used as its label; the input data in the second group can be clustered according to the characteristics of the data, and the input data of each category can be labeled in batches manually; the data in the third group can be unlabeled. In the embodiment of the present application, on the one hand, some input data that does not need to be manually labeled can be screened out; on the other hand, the input data that needs to be manually labeled can be clustered, so that the labeling personnel can configure corresponding labels for the input data in a category in batches, thereby reducing the workload of the staff and reducing the waste of human resources.

[0039] In the embodiments of the present application, Figure 1AAs shown, the recognition model may include a first recognition model and a second recognition model, and the first recognition model and the second recognition model may be trained using part of the marked training data, the first recognition model may also be referred to as a single-word recognition model, a single-word recognition network, the first recognition model is used to determine the single-word semantics, specifically, the first recognition model is used to extract the features of the input data, form a feature vector, and according to the feature vector, determine the single-word recognition result and the recognition confidence, as the first recognition result, wherein the single-word recognition result represents the semantics of the character, and the recognition confidence represents the credibility (or accuracy) of the single-word recognition result, the higher the recognition confidence, the higher the credibility of the single-word recognition result, otherwise, the credibility of the single-word recognition result is lower. The second recognition model may also be referred to as a feature extraction model, a feature extraction network. The second recognition model is used to determine the single-word classification to which the single word belongs, specifically, the second recognition model is used to extract the features of the input data, form a feature vector, and according to the feature vector, determine the corresponding feature class and feature similarity, as the second recognition result. Among them, the feature class can also be called a feature anchor point. The feature class is determined by multiple characters of the same category. The feature class is used to represent the features of characters of the same category. Specifically, for the same word, there may be a small number of distinguishing features in the features of different sub-images. Therefore, for the same word, there may be multiple similar feature vectors. Multiple similar feature vectors can be used to determine a feature anchor point that can express the characteristics of a single word. For example, the average value of multiple feature vectors can be used as a feature anchor point. Feature similarity refers to the degree of similarity between the feature vector of the input data to be identified and the feature anchor point. Feature similarity is used to characterize the accuracy (or credibility) of the identified feature class. The higher the feature similarity, the higher the credibility of the identified feature class. The lower the feature similarity, the lower the credibility of the identified feature class.

[0040] After the training of the first recognition model and the second recognition model is completed, the unlabeled input data can be input into the first recognition model and the second recognition model to obtain the first recognition result and the second recognition result, wherein the first recognition result includes the single word recognition result and the corresponding recognition confidence, and the second recognition result includes the feature class corresponding to the input data and the corresponding feature similarity, wherein the unlabeled input data can be understood as a sub-image containing a single word (or character), and the input data can be obtained by segmenting an image containing multiple words. After determining the first recognition result and the second recognition result, the credibility of the recognition result can be determined based on the first recognition result and the second recognition result, and then the input data is grouped to obtain the first group, the second group and the third group. Among them, the credibility of the recognition result represents the quality (or accuracy) of the recognition result, and the credibility of the recognition result is determined based on whether the single word recognition result is consistent with the feature class recognition result, the confidence level of the single word recognition result, and the level of feature similarity. After the input data is grouped, different annotation methods can be adopted for the input data in different groups. For example, for the input data of the first group, the single word recognition result can be obtained as an annotation to form annotated training data; for the input data of the second group, after clustering, the manual annotation information of the input data of each category can be obtained to form annotated training data; for the input data of the third group, it can be unannotated and used as the input of the next iteration of the first recognition model and the second recognition model. After determining the training data, the character recognition model can be trained based on the training data so as to determine the corresponding recognition result based on the trained character recognition model. In the embodiment of the present application, on the one hand, the first recognition result and the second recognition result can be used to filter out some input data that do not need to be manually annotated, reducing the manual workload and reducing the waste of manual resources; on the other hand, the input data that needs to be manually annotated can be clustered, so that the annotation personnel can configure the corresponding annotations for the input data in a category in batches, reducing the workload of the staff and further reducing the waste of manual resources.

[0041] Specifically, if Figure 1BAs shown, the embodiment of the present application can match the single-word recognition results in the first recognition result and the second recognition result with the feature-class recognition results to obtain a matching result, wherein the matching result indicates whether the single-word recognition result is consistent with the feature-class recognition result, and the matching result includes a first matching result and a second matching result, wherein the first matching result indicates that the single-word recognition result is consistent with the feature-class recognition result, and the second matching result indicates that the single-word recognition result is inconsistent with the feature-class recognition result; the feature similarity is compared with the similarity threshold to obtain a first comparison result; the recognition confidence is compared with the confidence threshold to obtain a second comparison result, and the credibility of the recognition result is determined based on the matching result, the first comparison result and the second comparison result, and the input data is grouped, wherein the first grouping may include an existing category simple sample group, the second grouping may include an existing category similar sample group, an existing category difficult sample group, and an unclassified sample group; the third grouping may include a low-quality sample group. Specifically, the input data in the representation group of the existing category simple sample group corresponds to the existing category in the recognition model, and the recognition accuracy is high and the recognition is simple. Therefore, when the feature similarity is greater than the similarity threshold and the recognition confidence is greater than the confidence threshold, the grouping described in the input data is determined to be the existing category simple sample group. The input data in the representation group of the existing category similar sample group is similar to the existing category in the recognition model, but the similarity is low. Therefore, when the feature similarity is greater than the similarity threshold and the recognition confidence is less than the confidence threshold, the grouping described in the input data is determined to be the existing category similar sample group. The input data in the representation group of the existing category difficult sample group is similar to the existing category in the recognition model, and the similarity is high, and the recognition difficulty is great. Therefore, when the feature similarity is less than the similarity threshold, the recognition confidence is greater than the confidence threshold, and the single-word recognition result is consistent with the feature class, the grouping described in the input data is determined to be the existing category difficult sample group. The input data in the unclassified sample group is characterized by a low similarity with the existing categories in the recognition model, and the corresponding recognition result cannot be determined. The recognition model needs to be trained with samples of this type. Therefore, when the feature similarity is less than the similarity threshold, the recognition confidence is less than the confidence threshold, and the single-word recognition result is inconsistent with the feature class, the grouping described in the input data is determined to be the unclassified sample group. The low-quality sample group represents that the input data in the group has low data quality. Therefore, when the feature similarity is less than the similarity threshold, the recognition confidence is less than the confidence threshold, and the single-word recognition result is consistent with the feature class, the grouping described in the input data is determined to be the low-quality sample group.

[0042] In an embodiment of the present application, part of the labeled training data can be used to train the first recognition model and the second recognition model, and part of the unlabeled training data can use the trained recognition model to cluster the unlabeled training data, thereby screening out part of the input data after clustering for manual labeling to obtain training data for training the first recognition model and the second recognition model.

[0043] The embodiment of the present application is an improvement on the annotation process of training data in the field of character recognition. Therefore, the embodiment of the present application can be applied to the training data annotation scenarios of various character recognition models, and the embodiment of the present application can also be applied to various scenarios for character recognition. For example, the character recognition model can be applied in the scenarios of image text recognition, image text semantic recognition, and ancient text recognition. The embodiment of the present application can group the input data and adopt the corresponding annotation method to configure the annotation for the input data to form training data, so as to use the training data to train the character recognition model applied in different scenarios. Among them, the character recognition model can also be divided according to the application field, so as to take the input data corresponding to the field to be grouped and annotated to obtain training data, so as to use the training data to train the character recognition model applied in different scenarios. For example, the character recognition model can be applied in the medical field, and the input data in the medical field can be annotated to obtain the training data for training the model; medical-related data can also be identified to determine medical information. For another example, the character recognition model can also be applied in the education field, and the input data in the education field can be annotated to obtain the training data for training the model; education-related data can also be identified to determine education information. The character recognition model can also be used in the e-commerce field. It can label product-related data to determine the training data for the training model; it can also identify product-related data to determine the product information of the product.

[0044] The embodiment of the present application provides a data processing method, which can be applied to a server, can group input data, and batch-label input data in different groups using different labeling methods, thereby reducing the workload of manual labeling and reducing the waste of manual resources. Specifically, Figure 2 As shown, the method includes:

[0045] Step 202, input the unlabeled input data into the recognition model to determine the corresponding recognition result. Wherein, the unlabeled input data can be understood as a sub-image containing a single word (or character), and the input data can be obtained by segmenting an image containing multiple sub-images. Specifically, as an optional embodiment, the method also includes: obtaining image data to be processed, and identifying the image data to obtain multiple sub-images containing a single word as input data. The embodiment of the present application can identify the image data and determine multiple object frames, the object frames are used to locate the single word in the image data, and then the sub-images are segmented from the image data according to the object frames as input data containing a single word. Wherein, in order to increase the diversity of the input data, the embodiment of the present application can also adjust the sub-image to expand the number of input data. Specifically, as an optional embodiment, the method also includes: adjusting the sub-image data to obtain an expanded sub-image as input data, and the data adjustment includes at least one of image rotation, image translation, image cropping, affine transformation, and noise enhancement processing. Among them, image translation, image rotation, image cropping, affine transformation, noise enhancement processing, etc. can be understood as adjusting the entire sub-image or a part of the sub-image.

[0046] In an embodiment of the present application, the recognition result of the recognition model may include a first recognition result and a second recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity. In an embodiment of the present application, a recognition model may be used to obtain a first recognition result and a second recognition result. Specifically, as an optional embodiment, the input data without annotations is input into the recognition model to determine the corresponding recognition result, including: inputting the input data without annotations into the recognition model, using the first output layer of the recognition model to output the first recognition result, and using the second output layer of the recognition model to output the second recognition result. In this embodiment, the recognition model may use a feature extraction layer to extract features of the input data, form a feature vector, and pass the feature vector to the output layer. For the output layer, two different output layers may be pre-configured for the recognition model, so that the recognition model may output the first recognition result and the second recognition result respectively through two different output layers.

[0047] In addition, the embodiment of the present application may also use two recognition models to respectively determine the first recognition result and the second recognition result. Specifically, as an optional embodiment, the input data without annotations is input into the recognition model to determine the corresponding recognition result, including: inputting the input data without annotations into the first recognition model to determine the first recognition result; inputting the input data without annotations into the second recognition model to determine the second recognition result. The first recognition model is used to extract the features of the input data, form a feature vector, and determine the single word recognition result and recognition confidence based on the feature vector. The second recognition model is used to extract the features of the input data, form a feature vector, and determine the corresponding feature class and feature similarity based on the feature vector.

[0048] The following is a specific example to describe the first recognition model and the second recognition model in detail. In an optional example, for the first recognition model, the first recognition model can adopt the network architecture of the residual network (Residual Network, ResNet). The characteristics of the residual network are that it is easy to optimize and can improve the accuracy by increasing the depth. The residual blocks inside the residual network use jump connections to alleviate the gradient vanishing problem caused by increasing the depth in the deep neural network. The embodiment of the present application can also adopt other types of network architectures. For example, the first recognition model can also adopt the network architecture of the original network (InceptionNet) and the network architecture of the terminal streamlined network (MobileNet). In addition, cross entropy can be used as a loss function to reversely adjust the first recognition model. Cross entropy (Cross Entropy) is mainly used to measure the difference information between two probability distributions. The cross entropy L CE It can be expressed by the following formula 1.

[0049]

[0050] Among them, K is the total number of word categories, y k is the value of the kth bit of the sample, p k It is the probability value predicted by the first recognition model (single-word recognition model) as the kth class.

[0051] As an optional example, for the second recognition model, the second recognition model can use the same network architecture as the first recognition model (such as a residual network), or other types of network architectures, which will not be described here. The training process of the second recognition model can use a triplet loss function (Triplet Loss) to reversely adjust the second recognition model, where the triplet can be understood as the current sample (Anchor), a negative sample (Negative), and a positive sample (Positive). Anchor is a sample randomly selected from the training data set, Positive is a sample that belongs to the same class as the Anchor, and Negative is a sample that is not of the same class as the Anchor. Triplet loss function L Tri It can be expressed by the following formula 2.

[0052]

[0053] Wherein, N is the total number of samples of training data, i represents the i-th sample, and f() is the feature extraction network. The training sample is composed of a triplet (a, p, n), where a represents the current sample (Anchor), p represents the positive sample of the same category (Positive), and n represents the negative sample of a different category (Negative). Based on the above triplet, a positive sample group (positive pair) and a negative sample group (negative pair) can be constructed. The purpose of Triplet loss is to separate the positive sample group and the negative sample group at a certain distance, so that the positive sample group is closer to the current sample and the negative sample group is far away from the current sample, so as to improve the resolution of the features extracted by the feature extraction network. Here, the Euclidean distance is used to measure the distance between two features (feature similarity). In an optional embodiment, during the training process of the second recognition model, the feature class (or feature anchor) of the second recognition model can be dynamically updated, such as the feature anchor F can be dynamically updated by the following formula 3 i .

[0054]

[0055] After determining the first recognition result and the second recognition result, the input data can be grouped according to the credibility of the recognition result in step 204. The credibility of the input data can be determined according to the first recognition result and the second recognition result, so that the input data can be grouped according to the credibility, and different groups can be annotated according to corresponding annotation methods, wherein, as an optional embodiment, the input data can be grouped according to the credibility of the recognition result, including: matching the feature class with the single word recognition result to determine the matching result; determining the credibility of the recognition result according to the matching result, the feature similarity and the recognition confidence, and grouping the input data. The single word recognition result in the first recognition result can be matched with the feature class in the second recognition result to determine whether the single word recognition result is consistent with the feature class, thereby obtaining the matching result, and the input data can be grouped according to the matching result, the feature similarity and the recognition confidence, so as to process the input data.

[0056] For feature similarity and recognition confidence, they can be compared with corresponding thresholds, and the input data can be grouped according to the comparison results and matching results. Specifically, as an optional embodiment, the reliability of the recognition result is determined based on the matching results, the feature similarity and the recognition confidence, and the input data is grouped, including: comparing the feature similarity with the similarity threshold to obtain a first comparison result; comparing the recognition confidence with the confidence threshold to obtain a second comparison result; determining the reliability of the recognition result based on the first comparison result, the second comparison result and the matching result, and grouping the input data. The first comparison result represents the degree of similarity between the feature class recognition result obtained by the second recognition model and the input data. When the feature similarity exceeds the similarity threshold, it represents that the similarity between the feature class recognition result obtained by the recognition model and the input data is high. Otherwise, it represents that the similarity between the feature class recognition result obtained by the recognition model and the input data is low. The second comparison result represents the degree of similarity between the single word recognition result obtained by the second recognition result and the input data. When the recognition confidence exceeds the confidence threshold, it represents that the similarity between the single word recognition result obtained by the second recognition result and the input data is high. Otherwise, it represents that the similarity between the single word recognition result obtained by the second recognition result and the input data is low.

[0057] The embodiment of the present application can divide the input data into three groups of data based on the matching results, the first comparison results and the second comparison results, which are respectively the first group, the second group and the third group in descending order of credibility. Among them, the second group can be further divided into a group of similar samples with existing categories, a group of difficult samples with existing categories, and a group of samples without categories.

[0058] After the input data is grouped, in step 206, annotation data can be configured in batches for the input data of each group to form training data. The embodiment of the present application can cluster the input data in each group, and batch annotate the input data after clustering. As an optional embodiment, the configuration of annotation data for the input data of each group to form training data includes: clustering the grouped input data according to the characteristics of the data, and batch determining the annotation data of each category of input data to form training data. In the embodiment of the present application, different annotation methods can be used for different groups. For example, for the first group, the recognition result of the recognition model can be used as its annotation; for the input data in the second group, clustering can be performed according to the characteristics of the data, and the input data of each category can be manually annotated in batches; the data in the third group can be unannotated. Specifically, as an optional embodiment, the grouped input data is clustered according to the characteristics of the data, and the labeling data of each category of input data is determined in batches to form training data, including at least one of the following steps: when the group to which the input data belongs is the first group, the single-word recognition results of each input data are obtained as the labeling data; when the group to which the input data belongs is the second group, the input data is clustered according to the single-word recognition results of the input data, and the labeling information of each category is obtained as the labeling data; when the group to which the input data belongs is the third group, the input data is used as the input for the next iteration of the recognition model. Wherein, as an optional embodiment, the first group includes a simple sample group of existing categories, the second group includes a similar sample group of existing categories, a difficult sample group of existing categories, and a sample group without categories; the third group includes a low-quality sample group.

[0059] The input data in the first grouping corresponds to the existing categories of the first recognition model, and the recognition confidence is high, and the single word recognition result of the first recognition model can be used as the labeling data. The input data in the second grouping includes a similar sample group of existing categories, a difficult sample group of existing categories, and a sample group without a category, wherein, for the input data in the similar sample group of existing categories and the difficult sample group of existing categories, the input data with the same single word recognition result can be clustered according to the single word recognition result of each input data, and the input data after clustering processing can be obtained, and then the labeling information of the manual labeling can be obtained as the labeling data of each category. For the input data in the sample group without a category, the feature can be extracted to obtain the feature vector, and the input data can be grouped according to the similarity between the feature vectors of each input data. In an optional example, the k-means clustering algorithm can be adopted to cluster the input data, wherein the k-means clustering algorithm is an iterative clustering analysis algorithm. For the input data in the third grouping, the quality of the input data is poor, therefore, the unlabeled method can be adopted to treat it as unlabeled data and perform the next iteration process of the recognition model. In some optional examples, the input data in the third group may also be discarded, which may be specifically set according to requirements.

[0060] It should be noted that the embodiment of the present application can continuously generate training data, and use the training data to train the first recognition model and the second recognition model, so as to continuously improve the capabilities of the first recognition model and the second recognition model. In addition, the embodiment of the present application can pre-set a processing threshold for unlabeled data as a termination condition for unlabeled data, so as to continuously add annotations to the unlabeled input data through the first recognition model and the second recognition model until the remaining amount of unlabeled input data is lower than the processing threshold, stop execution, and complete the annotation of the unlabeled data.

[0061] In order to facilitate the staff to annotate the unannotated input data in batches, the embodiment of the present application can also provide an interactive page to complete the input data annotation process. Specifically, as an optional embodiment, the method also includes: providing an interactive page to obtain the annotation information of the input data based on the interactive page. The interactive page may include a display area and an annotation control for each category of input data. The interactive page can be provided to the staff who annotate the input data. The display area in the interactive page can display the clustered input data. The staff can use the annotation control to annotate the input data of a category, thereby batch determining the annotation data corresponding to the input data of the category to form training data.

[0062] In addition, the embodiments of the present application can also display the labeled data and obtain corresponding correction information to adjust the labeled data, improve the accuracy of the labeled data, and further improve the recognition accuracy of the recognition model. Specifically, as an optional embodiment, the method also includes: outputting input data and corresponding labeled data, and obtaining corresponding correction information to adjust the labeled data according to the correction information. The interactive page also includes a display area for displaying input data and corresponding labeled data. The interactive page also includes a correction control. The user can trigger the correction control to form correction information and upload the correction information to correct the input data or labeled data according to the correction information. Among them, the corrected data can also be used as training data for the first recognition model and the second recognition model to further improve the recognition accuracy of the second recognition model and the second recognition model.

[0063] After the input data is clustered according to the characteristics of the data, there may be similar characters in the input data of a category that do not belong to the category. Therefore, the staff can also adjust the input data through the interactive page. Specifically, as an optional embodiment, the method also includes: receiving adjustment information for the input data to adjust the input data to the target category. The interactive page also includes a data category adjustment control. The target category can be a category generated after clustering processing or a category created by the staff. The staff can determine the input data that does not belong to the category from the input data of a category, and adjust the input data that does not belong to the category to the corresponding target category through the data category adjustment control, so as to batch label the input data belonging to the same category to form training data.

[0064] In an embodiment of the present application, unlabeled input data can be input into a recognition model to obtain corresponding recognition results, and the input data can be grouped according to the credibility of the recognition results. For the grouped input data, clustering processing can be performed according to the characteristics of the data, and the labeling method corresponding to the grouping can be used to batch configure labeling data for each category of input data, thereby forming training data. In an embodiment of the present application, the recognition results can be used to group the input data, and the input data in each group can be clustered and then the corresponding labeling method can be used to batch label the input data of each category. Compared with the method of manually labeling each input data separately, it can reduce the workload of manual labor and thus reduce the waste of manual resources.

[0065] Based on the above embodiments, the present application also provides a data processing method, which can be applied on the server side, such as Figure 3 As shown, the method includes:

[0066] Step 302: Obtain image data to be processed, and identify the image data to obtain a plurality of sub-images containing a single word as input data.

[0067] Step 304: perform data adjustment on the sub-image to obtain an expanded sub-image as input data, wherein the data adjustment includes at least one of image rotation, image translation, image cropping, affine transformation, and noise enhancement processing.

[0068] Step 306: Input the unlabeled input data into the first recognition model to determine a first recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence.

[0069] Step 308: Input the unlabeled input data into the second recognition model to determine a second recognition result. The second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity.

[0070] Step 310: Match the feature class with the single word recognition result to determine the matching result.

[0071] Step 312: Compare the feature similarity with a similarity threshold to obtain a first comparison result.

[0072] Step 314: Compare the recognition confidence with the confidence threshold to obtain a second comparison result.

[0073] Step 316: Determine the credibility of the recognition result according to the first comparison result, the second comparison result and the matching result, and group the input data.

[0074] Step 318: When the group to which the input data belongs is the first group, obtain the single-word recognition result of each input data as the marked data.

[0075] Step 320: When the group to which the input data belongs is the second group, clustering processing is performed on the input data according to the single-word recognition result of the input data, and labeling information of each category is obtained as labeling data.

[0076] Step 322: When the group to which the input data belongs is the third group, use the input data as the input for the next iteration of the recognition model. As an optional embodiment, the first group includes a simple sample group with existing categories, the second group includes a similar sample group with existing categories, a difficult sample group with existing categories, and a sample group with no categories; the third group includes a low-quality sample group. As an optional embodiment, the method further includes: providing an interactive page, the interactive page including a display area and a labeling control for input data of each category; and determining the labeling information of the input data based on triggering the labeling control. As an optional embodiment, the interactive page also includes a data category adjustment control, and the method further includes: adjusting the input data to the target category based on triggering the data category adjustment control.

[0077] In an embodiment of the present application, unlabeled input data can be input into the first recognition model and the second recognition model to obtain the first recognition result and the second recognition result, and the input data can be grouped according to the first recognition result and the second recognition result. After that, for the input data in the first group, the single-word recognition result can be obtained as a label to form labeled training data; for the input data in the second group, after clustering, the manual labeling information of the input data of each category can be obtained to form labeled training data; for the input data in the third group, it can be unlabeled and used as the input of the next iteration of the first recognition model and the second recognition model. In an embodiment of the present application, on the one hand, the first recognition result and the second recognition result can be used to filter out some input data that do not need to be manually labeled, reducing the manual workload and reducing the waste of manual resources; on the other hand, the input data that needs to be manually labeled can be clustered, so that the labeling personnel can batch configure corresponding labels for the input data in a category, reducing the workload of the staff and further reducing the waste of manual resources.

[0078] On the basis of the above embodiments, the embodiments of the present application also provide a data processing method, which can be applied to the server. The server can be understood as a device that interacts with the terminal to analyze the unlabeled input data using the recognition model. The method of this embodiment can receive the input data uploaded by the terminal, and input it into the recognition model, determine the recognition result, and group the input data according to the credibility of the recognition result, and then cluster the input data in each group, and return the clustered input data to the terminal so that the terminal personnel can label the input data of each category. Specifically, Figure 4A As shown, the method includes:

[0079] Step 402: Provide a service page to obtain uploaded input data according to the service page.

[0080] Step 404: input the unlabeled input data into the recognition model to determine the corresponding recognition result.

[0081] Step 406: group the input data according to the credibility of the recognition result.

[0082] Step 408: configuring the labeled data for each grouped input data in batches to form training data. Specifically, as an optional embodiment, configuring the labeled data for each grouped input data in batches to form training data includes: clustering the grouped input data according to the characteristics of the data, and sending the input data of each category to the terminal for display in the service page; obtaining the labeled data based on the service page, so as to label the input data in batches by category to form training data.

[0083] In an embodiment of the present application, the staff can also adjust the input data of each category to adjust the input data to the correct category. Specifically, as an optional embodiment, the method also includes: receiving adjustment information for the input data of each category; adjusting the input data of each category according to the adjustment information, and issuing annotation reference information to annotate the adjusted input data of each category.

[0084] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0085] In the present application embodiment, Figure 4BAs shown, the server can provide a service page to the terminal, and the service page includes a data upload interface. The terminal personnel can upload input data to the server by triggering the data upload interface. After the server receives the unlabeled input data, it can input the unlabeled input data into the recognition model to obtain the corresponding recognition result, and group the input data according to the credibility of the recognition result. For the grouped input data, clustering processing can be performed according to the characteristics of the data to obtain input data of various categories for annotation. In an optional example, the input data of various categories can be sent to the annotation personnel, and the annotation personnel can annotate the input data in batches according to the category to form training data. After determining the training data, the server can return the training data to the user of the terminal. After the user of the terminal receives the training data, the character recognition model can be trained based on the training data so as to recognize the characters through the character recognition model. In an optional example, the interactive page can also include an adjustment control. The user can trigger the adjustment control to make correction information for the labeled data and upload it to the server. The server adjusts the training data based on the correction information, thereby improving the accuracy of the training data, and then improving the accuracy of the character recognition model. In another optional example, after the input data of each category is determined, the input data of each category can be returned to the terminal so that the user of the terminal can batch-label the input data of each category to form training data, so as to train the character recognition model based on the training data, wherein, in the scheme of sending the input data of each category to the user for labeling, the present embodiment can also send the labeled training data to other labelers for confirmation, and other labelers can label the user's labeling to further improve the accuracy of the labeling. In the scheme of this embodiment, for the user, the user can upload the unlabeled input data to the server, and the server will group and cluster the input data to form input data of each category, so that the user of the terminal can more conveniently obtain the labeled data.

[0086] On the basis of the above embodiments, the embodiments of the present application further provide a data processing method, which can be applied to a terminal. The terminal can be understood as interacting with a server to upload input data to the server and receive unlabeled data after clustering processing, so as to label the unlabeled data. Specifically, Figure 5 As shown, the method includes:

[0087] Step 502: Provide a service page, wherein the service page includes a data upload interface.

[0088] Step 504: based on the triggering of the data upload interface, the input data is obtained and uploaded to the server, the server is used to input the unlabeled input data into the recognition model to determine the corresponding recognition result; and group the input data according to the credibility of the recognition result; and send the input data of each group.

[0089] Step 506: Receive input data of each group sent by the server and display it on the service page.

[0090] Step 508: Receive labeling information for each group of input data to form training data.

[0091] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0092] In the embodiment of the present application, the server can provide a service page to the terminal, and the service page includes a data upload interface. The terminal personnel can upload input data to the server by triggering the data upload interface. After receiving the unlabeled input data, the server can input the unlabeled input data into the recognition model to obtain the corresponding recognition result, and group the input data according to the credibility of the recognition result. For the grouped input data, clustering processing can be performed according to the characteristics of the data, and the input data of each category can be sent to the terminal. The terminal can receive the annotation information of the input data of each category to form training data. In the scheme of this embodiment, before the unlabeled input data is annotated, it can be uploaded to the server, and the server will group and cluster the input data to form input data of each category, so that the terminal personnel can annotate the input data in batches, which can reduce the manual workload and reduce the waste of manual resources.

[0093] On the basis of the above embodiments, the embodiments of the present application further provide a data processing method, which can be applied on the server side, and can input the input data to be annotated into the recognition model, determine the recognition result, and group the input data according to the credibility of the recognition result, and annotate the input data of each group in batches, thereby forming training data, and using the training data to train the character recognition model. Specifically, Figure 6 As shown, the method includes:

[0094] Step 602: input the unlabeled input data into the recognition model to determine the corresponding recognition result.

[0095] Step 604: group the input data according to the credibility of the recognition result.

[0096] Step 606: configure label data for each group of input data in batches to form training data.

[0097] Step 608: Train the character recognition model based on the training data so as to recognize the input data to be recognized based on the trained character recognition model.

[0098] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0099] In an embodiment of the present application, unlabeled input data can be input into a recognition model to obtain corresponding recognition results, and the input data can be grouped according to the credibility of the recognition results, and label data can be configured in batches for the input data of each group to form training data. Afterwards, the training data can be used to train a character recognition model so that the input data to be recognized can be recognized based on the trained character recognition model. In an embodiment of the present application, the input data can be labeled in batches by labeling the input data of each group, which can be more convenient to label. Therefore, a larger amount of training data can be used to train the character recognition model, thereby improving the recognition accuracy of the character recognition model.

[0100] On the basis of the above embodiments, the embodiments of the present application also provide a data processing method, which can be applied to a server, and the server can interact with a terminal. The server can train a character recognition model through batch labeled training data, and use the trained character recognition model to recognize the input data to be recognized, determine the corresponding recognition result, and send it to the terminal. Specifically, Figure 7 As shown, the method includes:

[0101] Step 702, provide a data interaction page to obtain input data to be recognized based on the data interaction page. Specifically, as an optional embodiment, the method further includes: obtaining image data to be processed, and recognizing the image data to obtain multiple sub-images containing a single word as input data. The embodiment of the present application can also expand the input data to increase the data volume of the input data and improve the recognition accuracy of the character recognition model. Specifically, as an optional embodiment, the method further includes: adjusting the sub-image data to obtain an expanded sub-image as input data, and the data adjustment includes at least one of image rotation, image translation, image cropping, affine transformation, and noise increase processing.

[0102] Step 704: determine the recognition result based on the input data and the trained character recognition model, the character recognition model is trained based on multiple groups of labeled training data, the multiple groups of training data are grouped based on the credibility of the input data input to the recognition model, and the annotation of each group of training data is obtained by batch annotation.

[0103] Step 706: Send the recognition result.

[0104] Specifically, as an optional embodiment, the method also includes the step of determining training data: inputting unlabeled input data into the recognition model to determine the corresponding recognition result; grouping the input data according to the credibility of the recognition result; and configuring labeled data in batches for the input data of each group to form training data.

[0105] The embodiment of the present application can cluster the input data of each group according to the characteristics, and batch-label the input data of each category. Specifically, as an optional embodiment, the batch configuration of label data for the input data of each group to form training data includes: clustering the grouped input data according to the characteristics of the data, and batch-determining the label data of the input data of each category to form training data. The recognition model may include a first recognition model and a second recognition model. The embodiment can group the input data according to the recognition results of the first recognition model and the second recognition model. Specifically, as an optional embodiment, the inputting of unlabeled input data into the recognition model to determine the corresponding recognition result includes: inputting the unlabeled input data into the first recognition model to determine the first recognition result; inputting the unlabeled input data into the second recognition model to determine the second recognition result. Specifically, as an optional embodiment, the recognition result includes a first recognition result and a second recognition result, the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity; wherein, grouping the input data according to the credibility of the recognition result includes: matching the feature class with the single-word recognition result to determine the matching result; determining the credibility of the recognition result based on the matching result, the feature similarity and the recognition confidence, and grouping the input data.

[0106] Specifically, as an optional embodiment, the credibility of the recognition result is determined based on the matching result, the feature similarity and the recognition confidence, and the input data is grouped, including: comparing the feature similarity with a similarity threshold to obtain a first comparison result; comparing the recognition confidence with a confidence threshold to obtain a second comparison result; determining the credibility of the recognition result based on the first comparison result, the second comparison result and the matching result, and grouping the input data.

[0107] For different groups, corresponding processing methods can be adopted to determine the annotations. Specifically, as an optional embodiment, the grouped input data is clustered according to the characteristics of the data, and the annotation data of each category of input data is determined in batches to form training data, including at least one of the following steps: when the group to which the input data belongs is the first group, the single-word recognition results of each input data are obtained as the annotation data; when the group to which the input data belongs is the second group, the input data is clustered according to the single-word recognition results of the input data, and the annotation information of each category is obtained as the annotation data; when the group to which the input data belongs is the third group, the input data is used as the input of the next iteration of the recognition model. Specifically, as an optional embodiment, the first group includes a simple sample group of existing categories, the second group includes a similar sample group of existing categories, a difficult sample group of existing categories, and a sample group without categories; the third group includes a low-quality sample group.

[0108] The embodiment of the present application may provide an interactive page to obtain the annotation of the input data. Specifically, as an optional embodiment, the method further includes: providing an interactive page to obtain the annotation information of the input data based on the interactive page. The input data of one category may contain input data belonging to other categories. The user can adjust the input data in the interactive page to adjust the input data to the correct category. Specifically, as an optional embodiment, the method further includes: receiving adjustment information for the input data to adjust the input data to the target category. The embodiment of the present application can also display the input data and the corresponding annotation to the user so that the user can make corrections. Specifically, as an optional embodiment, the method further includes: outputting the input data and the corresponding annotation data, and obtaining the corresponding correction information to adjust the annotated data according to the correction information.

[0109] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0110] In an embodiment of the present application, unlabeled input data can be input into a recognition model to obtain corresponding recognition results, and the input data can be grouped according to the credibility of the recognition results. For the grouped input data, the labeling method corresponding to the grouping is used to configure labeling data for the input data of each group in batches, thereby forming training data, and the training data is used to train the character recognition model. Afterwards, the input data to be recognized is obtained based on the interactive page, and the trained character recognition model is used to recognize it, determine the recognition result, and issue the recognition result. In an embodiment of the present application, the recognition result of the input data can be used to group the input data, and the input data in each group can be labeled in batches. Compared with the method of manually labeling each input data separately, it can reduce the manual workload and thus reduce the waste of manual resources.

[0111] Based on the above embodiments, the embodiments of the present application further provide a data processing method, which can be applied to a terminal. The terminal can interact with a server to upload input data to be identified to the server. The server identifies the input data to be identified, determines the identification result, and feeds back the identification result to the terminal. Specifically, Figure 8 As shown, the method includes:

[0112] Step 802: Provide a data interaction page, wherein the data interaction page includes a data upload control.

[0113] Step 804: based on triggering the data upload control, the input data to be recognized is obtained and uploaded to the server, and the server is used to determine the recognition result based on the character recognition model and the input data. The character recognition model is trained based on multiple groups of labeled training data, and the multiple groups of training data are grouped based on the credibility of the input data input into the recognition model. The annotation of each group of training data is obtained by batch annotation.

[0114] Step 806: Receive the recognition result.

[0115] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0116] In the embodiment of the present application, the server can provide a data interaction page to the terminal, and the user of the terminal can upload the input data to be identified to the server by triggering the data upload control. After receiving the input data to be identified, the server can determine the recognition result based on the trained recognition model and feed back the recognition result to the terminal.

[0117] On the basis of the above embodiments, the embodiments of the present application further provide a data processing method, which can be applied to a server, and the server can interact with a terminal. The embodiments of the present application can be applied to the scene of recognizing ancient characters, and the ancient character image data can be recognized through the pre-trained character recognition model in the server to determine the recognition result. Specifically, Fig. 9 As shown, the method includes:

[0118] Step 902: Provide a data interaction page to obtain ancient character image data to be identified based on the data interaction page.

[0119] Step 904: determine the recognition result based on the ancient character image data and the trained character recognition model. The character recognition model is trained based on multiple groups of labeled training data. The multiple groups of training data are grouped based on the credibility of the ancient character image data input into the recognition model. The annotation of each group of ancient character image data is obtained by batch annotation.

[0120] Step 906: Send the recognition result.

[0121] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0122] Among them, ancient characters can include ancient Chinese characters, and ancient Chinese characters can be understood as Chinese characters used in China before the appearance of simplified characters. Simplified characters are officially announced simplified characters, usually referring to the characters in the "General List of Simplified Characters" announced in 1956. Ancient Chinese characters can be understood as Chinese characters used in China before simplified characters, and can include original ancient Chinese characters, traditional characters, oracle bone inscriptions, bronze inscriptions, Warring States characters, Qin characters, and ancient official scripts in the process of official script change from Qin to the early Han Dynasty. Modern Chinese characters can be understood as simplified characters, and whether traditional characters belong to modern Chinese characters or ancient Chinese characters can be determined based on demand, for example, based on the font of traditional characters, the time to which they belong, etc. In the embodiment of the present application, after training the character recognition model, the ancient character image data to be recognized can be obtained based on the interactive page, and the trained character recognition model can be used for recognition, the recognition result can be determined, and the recognition result can be issued. In the process of training the character recognition model, the embodiment of the present application can group the ancient character image data according to the credibility of the ancient character image data used for training, and batch annotate the ancient character images in each group, which can facilitate the annotation of the ancient character images.

[0123] In addition to being applied in the above-mentioned scenarios, the embodiments of the present application can also be applied in other scenarios. For example, the embodiments of the present application can also be applied in e-commerce scenarios to recognize characters in e-commerce product promotion pages, extract corresponding characters, and determine product information of the product, such as product name, product type, product attributes, etc. Specifically, the method includes:

[0124] Get product-related data for the product to be identified.

[0125] The recognition result is determined based on the product-related data and the trained character recognition model. The character recognition model is trained based on multiple groups of labeled training data. The multiple groups of training data are grouped based on the credibility of the product-related data input into the recognition model. The annotations of each group of training data are obtained by batch annotation.

[0126] The commodity information of the commodity is determined according to the recognition result.

[0127] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0128] In the embodiment of the present application, the commodity-related data of the commodity to be identified may include commodity title, commodity promotion data (such as a promotional video with characters), commodity description and other data. In the embodiment of the present application, the commodity-related data to be identified can be obtained, and the identification can be performed using a pre-trained recognition model to obtain a recognition result, and the commodity information of the commodity can be determined based on the recognition result. For example, the name of the commodity, commodity specifications and other information of the commodity can be determined based on the recognition result so as to perform corresponding processing.

[0129] In addition, in an optional embodiment, the embodiment of the present application can also be applied in an e-commerce scenario, and the relevant data of the goods to be marked can be grouped, and corresponding marking methods can be adopted for different groups to perform batch marking, which can reduce the manual workload. Specifically, the method includes:

[0130] The unlabeled commodity-related data is input into the recognition model to determine the corresponding recognition results.

[0131] The commodity-related data are grouped according to the credibility of the recognition result.

[0132] The batch configuration labeling data for the commodity-related data of each group forms the training data.

[0133] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0134] In the embodiment of the present application, product-related data can be obtained, and the recognition model can be used to determine the recognition result of the product-related data. Then, the product-related data can be grouped according to the credibility of the recognition result, and the product-related data of each group can be labeled in batches to form training data. After the training data is determined, the character recognition model can be trained based on the training data so that the product-related information can be recognized by the character recognition model.

[0135] In addition to being applied in the above-mentioned scenarios, the embodiments of the present application can also be applied in other scenarios. For example, the embodiments of the present application can also be applied in medical scenarios, and can recognize characters in documents in medical scenarios, extract corresponding characters, and determine corresponding medical information, such as determining the patient's personal information, the patient's disease information, etc. Specifically, the method includes:

[0136] Access medical related data.

[0137] The recognition result is determined based on the medical-related data and the trained character recognition model. The character recognition model is trained based on multiple groups of labeled training data. The multiple groups of training data are grouped based on the credibility of the medical-related data input into the recognition model. The annotations of each group of training data are obtained by batch annotation.

[0138] Medical information is determined based on the recognition result.

[0139] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0140] The embodiments of the present application can also be applied to the scene of identifying medical-related data, which may include medical documents, medical teaching materials, etc. The embodiments of the present application can obtain medical-related data, and use a pre-trained character recognition model to determine the recognition result, and then determine the corresponding medical information based on the recognition result. For example, the disease type of the user's disease and the user's medical expenses can be determined based on the recognition result, so as to perform corresponding processing.

[0141] In addition, in an optional embodiment, the embodiment of the present application can also be applied in a medical scenario, and the medical-related data to be labeled can be grouped, and different groups can be labeled in a corresponding manner to perform batch labeling, which can reduce the manual workload. Specifically, the method includes:

[0142] Input the unlabeled medical-related data into the recognition model to determine the corresponding recognition results.

[0143] The medical-related data are grouped according to the credibility of the recognition results.

[0144] Configure labeled data for each group of medical-related data in batches to form training data.

[0145] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0146] In the embodiment of the present application, medical-related data can be obtained, and the recognition model can be used to determine the recognition result of the medical-related data. Then, the medical-related data can be grouped according to the credibility of the recognition result, and the medical-related data of each group can be labeled in batches to form training data. After the training data is determined, the character recognition model can be trained based on the training data so that the medical-related information can be recognized by the character recognition model.

[0147] In addition to being applied in the above-mentioned scenarios, the embodiments of the present application can also be applied in other scenarios. For example, the embodiments of the present application can also be applied in educational scenarios, and can identify target images in educational scenarios, determine corresponding recognition results, and perform corresponding processing according to the recognition results. For example, the recognition results can be determined based on images of student homework, and the recognition results can be displayed to the teacher. For another example, the recognition results can be determined based on images of handwritten student academic records, and the recognition results can be entered into the academic record management system. Specifically, the method includes:

[0148] Get the target image data.

[0149] The recognition result is determined based on the target image data and the trained character recognition model. The character recognition model is trained based on multiple groups of labeled training data. The multiple groups of training data are grouped based on the credibility of the image data input into the recognition model. The annotations of each group of training data are obtained by batch annotation.

[0150] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0151] The embodiments of the present application can be applied in educational scenarios, and the target image data can include images of student homework, images of answer sheets, images of school textbooks, etc. The embodiments of the present application can obtain image data related to education, and determine the recognition result through a trained character recognition model, and then perform corresponding processing based on the recognition result. For example, when the image data is an image of a student's answer sheet, the embodiments of the present application can obtain the recognition result, and determine whether the answer is correct based on the recognition result, and then determine the student's corresponding score.

[0152] In addition, in an optional embodiment, the embodiment of the present application can also be applied in an educational scenario, and the education-related data to be labeled can be grouped, and corresponding labeling methods can be adopted for different groups to perform batch labeling, which can reduce the manual workload. Specifically, the method includes:

[0153] The unlabeled education-related data is input into the recognition model to determine the corresponding recognition results.

[0154] The education-related data are grouped according to the confidence of the identification results.

[0155] Configure labeled data in batches for each group of education-related data to form training data.

[0156] The implementation of this embodiment is similar to the implementation of the above method embodiment. The specific implementation process can refer to the specific implementation process of the above method embodiment, which will not be repeated here.

[0157] The embodiment of the present application can obtain education-related data and input the education-related data into a recognition model to determine a recognition result. Thereafter, the education-related data can be grouped based on the credibility of the recognition result, and the education-related data in each group can be batch-labeled to obtain training data for training a character recognition model. In the embodiment of the present application, the training data can be determined by batch-labeling the education-related data in each group, which can make data labeling simpler and more convenient.

[0158] It should be noted that, for the method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present application are not limited by the described order of actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present application.

[0159] Based on the above embodiment, this embodiment further provides a data processing device, referring to Fig.10, specifically it can include the following modules:

[0160] The recognition result acquisition module 1002 is used to input the unlabeled input data into the recognition model to determine the corresponding recognition result.

[0161] The grouping result acquisition module 1004 is used to group the input data according to the credibility of the recognition result.

[0162] The training data acquisition module 1006 is used to configure annotation data for each group of input data in batches to form training data.

[0163] In summary, in the embodiment of the present application, the unlabeled input data can be input into the recognition model to obtain the corresponding recognition result, and the input data can be grouped according to the credibility of the recognition result. For the grouped input data, clustering processing can be performed according to the characteristics of the data, and the labeling method corresponding to the grouping can be used to batch configure labeling data for each category of input data, thereby forming training data. In the embodiment of the present application, the recognition result can be used to group the input data, and the input data in each group can be clustered and then the corresponding labeling method can be used to batch label the input data of each category. Compared with the method of manually labeling each input data separately, it can reduce the workload of manual labor and thus reduce the waste of manual resources.

[0164] On the basis of the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0165] The image data acquisition and processing module is used to acquire the image data to be processed, and to identify the image data to obtain a plurality of sub-images containing a single word as input data.

[0166] The expanded image acquisition processing module is used to perform data adjustment on the sub-image to obtain an expanded sub-image as input data, wherein the data adjustment includes at least one of image rotation, image translation, image cropping, affine transformation, and noise enhancement processing.

[0167] The first data input processing module is used to input the unlabeled input data into the first recognition model to determine the first recognition result. The first recognition result includes a single word recognition result and a corresponding recognition confidence.

[0168] The second data input processing module is used to input the unlabeled input data into the second recognition model to determine a second recognition result. The second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity.

[0169] The matching result acquisition processing module is used to match the feature class with the single word recognition result to determine the matching result.

[0170] The first comparison result acquisition processing module is used to compare the feature similarity with a similarity threshold to obtain a first comparison result.

[0171] The second comparison result acquisition processing module is used to compare the recognition confidence with the confidence threshold to obtain a second comparison result.

[0172] The grouping result acquisition processing module is used to determine the credibility of the recognition result according to the first comparison result, the second comparison result and the matching result, and to group the input data.

[0173] The first group data processing module is used to obtain the single word recognition result of each input data as the marking data when the group to which the input data belongs is the first group.

[0174] The second grouping data processing module is used to cluster the input data according to the single-word recognition result of the input data when the group to which the input data belongs is the second group, and obtain the labeling information of each category as the labeling data.

[0175] The third grouping data processing module is used to use the input data as the input for the next iteration of the recognition model when the group to which the input data belongs is the third group. As an optional embodiment, the first group includes a simple sample group with existing categories, the second group includes a similar sample group with existing categories, a difficult sample group with existing categories, and a sample group without categories; the third group includes a low-quality sample group. As an optional embodiment, the device further includes: a page providing processing module, which is used to provide an interactive page, and the interactive page includes a display area and a labeling control for input data of each category; and a labeling information acquisition processing module, which is used to determine the labeling information of the input data based on the triggering of the labeling control. As an optional embodiment, the interactive page also includes a data category adjustment control, and the device further includes: a data category adjustment processing module, which is used to adjust the input data to the target category based on the triggering of the data category adjustment control.

[0176] In summary, in the embodiment of the present application, the unlabeled input data can be input into the first recognition model and the second recognition model to obtain the first recognition result and the second recognition result, and the input data can be grouped according to the first recognition result and the second recognition result. After that, for the input data in the first group, the single word recognition result can be obtained as a label to form labeled training data; for the input data of the second group, after clustering, the manual labeling information of the input data of each category can be obtained to form labeled training data; for the input data of the third group, it can be unlabeled and used as the input of the next iteration of the first recognition model and the second recognition model. In the embodiment of the present application, on the one hand, the first recognition result and the second recognition result can be used to filter out some input data that do not need to be manually labeled, reducing the manual workload and reducing the waste of manual resources; on the other hand, the input data that needs to be manually labeled can be clustered, so that the labeling personnel can configure the corresponding labels for the input data in a category in batches, reducing the workload of the staff and further reducing the waste of manual resources.

[0177] Based on the above embodiment, this embodiment further provides a data processing device, referring to Fig.11 , specifically it can include the following modules:

[0178] The service page providing module 1102 is used to provide a service page to obtain uploaded input data according to the service page.

[0179] The recognition result determination module 1104 is used to input the unlabeled input data into the recognition model to determine the corresponding recognition result.

[0180] The grouping result determination module 1106 is used to group the input data according to the credibility of the recognition result.

[0181] The classification result sending module 1108 is used to configure the annotation data for the input data of each group in batches to form training data. Specifically, as an optional embodiment, the classification result sending module 1108 specifically includes: clustering the grouped input data according to the characteristics of the data, and sending the input data of each category to the terminal for display in the service page; obtaining the annotation data according to the service page, so as to mark the input data in batches by category to form training data. In the embodiment of the present application, the staff can also adjust the input data of each category to adjust the input data to the correct category. Specifically, as an optional embodiment, the device also includes: a data category adjustment module, which is used to receive adjustment information for the input data of each category; adjust the input data of each category according to the adjustment information, and send annotation reference information to mark the adjusted input data of each category.

[0182] In summary, in the embodiment of the present application, the server can provide a service page to the terminal, and the service page includes a data upload interface. The terminal personnel can upload input data to the server by triggering the data upload interface. After the server receives the unlabeled input data, it can input the unlabeled input data into the recognition model to obtain the corresponding recognition result, and group the input data according to the credibility of the recognition result. For the grouped input data, clustering can be performed according to the characteristics of the data, and the input data of each category is sent to the terminal. The terminal personnel can mark the input data in batches according to the category to form training data. In the embodiment of the present application, the recognition result can also be used to mark some unlabeled data, which can be set according to the needs. In the scheme of this embodiment, before marking the unlabeled input data, it can be uploaded to the server, and the server groups and clusters the input data to form input data of each category, so that the terminal personnel can mark the input data in batches, which can reduce the manual workload and reduce the waste of manual resources.

[0183] Based on the above embodiment, this embodiment further provides a data processing device, referring to Fig.12 , specifically it can include the following modules:

[0184] The service page display module 1202 is used to provide a service page, and the service page includes a data upload interface.

[0185] The input data upload module 1204 is used to obtain input data based on the triggering of the data upload interface and upload it to the server. The server is used to input the unlabeled input data into the recognition model to determine the corresponding recognition result; and group the input data according to the credibility of the recognition result; and send the input data of each group.

[0186] The classification result display module 1206 is used to receive the input data of each group sent by the server and display it in the service page.

[0187] The training data determination module 1208 is used to receive the labeling information of each group of input data to form training data.

[0188] In summary, in the embodiment of the present application, the server can provide a service page to the terminal, and the service page includes a data upload interface. The terminal personnel can upload input data to the server by triggering the data upload interface. After receiving the unlabeled input data, the server can input the unlabeled input data into the recognition model to obtain the corresponding recognition result, and group the input data according to the credibility of the recognition result. For the grouped input data, clustering processing can be performed according to the characteristics of the data, and the input data of each category can be sent to the terminal. The terminal can receive the annotation information of the input data of each category to form training data. In the scheme of this embodiment, before the unlabeled input data is annotated, it can be uploaded to the server, and the server will group and cluster the input data to form input data of each category, so that the terminal personnel can annotate the input data in batches, which can reduce the manual workload and reduce the waste of manual resources.

[0189] Based on the above embodiment, this embodiment further provides a data processing device, referring to Fig.13 , specifically it can include the following modules:

[0190] The unlabeled data recognition module 1302 is used to input the unlabeled input data into the recognition model to determine the corresponding recognition result.

[0191] The unlabeled data grouping module 1304 is used to group the input data according to the credibility of the recognition result.

[0192] The unlabeled data configuration module 1306 is used to configure labeled data for each group of input data in batches to form training data.

[0193] The recognition model training module 1308 is used to train the character recognition model based on the training data so as to recognize the input data to be recognized based on the trained character recognition model.

[0194] In an embodiment of the present application, unlabeled input data can be input into a recognition model to obtain corresponding recognition results, and the input data can be grouped according to the credibility of the recognition results, and label data can be configured in batches for the input data of each group to form training data. Afterwards, the training data can be used to train a character recognition model so that the input data to be recognized can be recognized based on the trained character recognition model. In an embodiment of the present application, the input data can be labeled in batches by labeling the input data of each group, which can be more convenient to label. Therefore, a larger amount of training data can be used to train the character recognition model, thereby improving the recognition accuracy of the character recognition model.

[0195] Based on the above embodiment, this embodiment further provides a data processing device, referring to Fig.14 , specifically it can include the following modules:

[0196] The data interaction page providing module 1402 is used to provide a data interaction page to obtain input data to be identified based on the data interaction page.

[0197] The input data recognition processing module 1404 is used to determine the recognition result based on the input data and the trained character recognition model. The character recognition model is trained based on multiple groups of labeled training data, and the multiple groups of training data are grouped based on the credibility of the input data input into the recognition model. The annotation of each group of training data is obtained by batch annotation.

[0198] The recognition result sending processing module 1406 is used to send the recognition result.

[0199] In an embodiment of the present application, unlabeled input data can be input into a recognition model to obtain corresponding recognition results, and the input data can be grouped according to the credibility of the recognition results. For the grouped input data, the labeling method corresponding to the grouping is used to configure labeling data for the input data of each group in batches, thereby forming training data, and the training data is used to train the character recognition model. Afterwards, the input data to be recognized is obtained based on the interactive page, and the trained character recognition model is used to recognize it, determine the recognition result, and issue the recognition result. In an embodiment of the present application, the recognition result of the input data can be used to group the input data, and the input data in each group can be labeled in batches. Compared with the method of manually labeling each input data separately, it can reduce the manual workload and thus reduce the waste of manual resources.

[0200] Based on the above embodiment, this embodiment further provides a data processing device, referring to Fig.15 , specifically it can include the following modules:

[0201] The data interaction page display module 1502 is used to provide a data interaction page, and the data interaction page includes a data upload control.

[0202] The input data acquisition and upload module 1504 is used to obtain the input data to be recognized and upload it to the server according to the triggering of the data upload control. The server is used to determine the recognition result according to the character recognition model and the input data. The character recognition model is trained based on multiple groups of labeled training data. The multiple groups of training data are grouped according to the credibility of the input data input into the recognition model. The annotation of each group of training data is obtained by batch annotation.

[0203] The recognition result receiving and processing module 1506 is used to receive the recognition result.

[0204] In the embodiment of the present application, the server can provide a data interaction page to the terminal, and the user of the terminal can upload the input data to be identified to the server by triggering the data upload control. After receiving the input data to be identified, the server can determine the recognition result based on the trained recognition model and feed back the recognition result to the terminal.

[0205] On the basis of the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0206] An ancient character image acquisition module, used for providing a data interaction page to acquire ancient character image data to be identified based on the data interaction page;

[0207] An ancient character image recognition module, used to determine a recognition result based on the ancient character image data and a trained character recognition model, wherein the character recognition model is trained based on multiple groups of annotated training data, the multiple groups of training data are grouped based on the credibility of the ancient character image data input to the recognition model, and the annotations of each group of ancient character image data are annotated based on a batch annotation method;

[0208] The ancient character recognition result sending module is used to send the recognition result.

[0209] In the embodiment of the present application, after training the character recognition model, the ancient character image data to be recognized can be obtained based on the interactive page, and the trained character recognition model can be used for recognition, the recognition result can be determined, and the recognition result can be issued. In the process of training the character recognition model, the ancient character image data can be grouped according to the credibility of the ancient character image data used for training, and the ancient character images in each group can be labeled in batches, which can facilitate the labeling of the ancient character images.

[0210] On the basis of the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0211] The commodity-related data determination module is used to obtain commodity-related data of the commodity to be identified.

[0212] The commodity-related data recognition module is used to determine the recognition result based on the commodity-related data and a trained character recognition model, wherein the character recognition model is trained based on multiple groups of labeled training data, and the multiple groups of training data are grouped based on the credibility of the commodity-related data input into the recognition model, and the annotations of each group of training data are obtained by batch annotation.

[0213] The commodity-related data processing module is used to determine the commodity information of the commodity according to the recognition result.

[0214] In the embodiment of the present application, the commodity-related data of the commodity to be identified may include commodity title, commodity promotion data (such as a promotional video with characters), commodity description and other data. In the embodiment of the present application, the commodity-related data to be identified can be obtained, and the identification can be performed using a pre-trained recognition model to obtain a recognition result, and the commodity information of the commodity can be determined based on the recognition result. For example, the name of the commodity, commodity specifications and other information of the commodity can be determined based on the recognition result so as to perform corresponding processing.

[0215] On the basis of the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0216] The commodity-related data acquisition module is used to input the unlabeled commodity-related data into the recognition model to determine the corresponding recognition results.

[0217] The commodity-related data grouping module is used to group the commodity-related data according to the credibility of the recognition result.

[0218] The commodity-related data configuration module is used to configure the annotation data for the commodity-related data of each group in batches to form training data.

[0219] In the embodiment of the present application, product-related data can be obtained, and the recognition model can be used to determine the recognition result of the product-related data. Then, the product-related data can be grouped according to the credibility of the recognition result, and the product-related data of each group can be labeled in batches to form training data. After the training data is determined, the character recognition model can be trained based on the training data so that the product-related information can be recognized by the character recognition model.

[0220] On the basis of the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0221] The medical-related data determination module is used to obtain medical-related data.

[0222] The medical-related data recognition module is used to determine the recognition result based on the medical-related data and a trained character recognition model. The character recognition model is trained based on multiple groups of labeled training data. The multiple groups of training data are grouped based on the credibility of the medical-related data input into the recognition model. The annotations of each group of training data are obtained by batch annotation.

[0223] The medical-related data processing module is used to determine the medical information based on the recognition result.

[0224] The embodiments of the present application can also be applied to the scene of identifying medical-related data, which may include medical documents, medical teaching materials, etc. The embodiments of the present application can obtain medical-related data, and use a pre-trained character recognition model to determine the recognition result, and then determine the corresponding medical information based on the recognition result. For example, the disease type of the user's disease and the user's medical expenses can be determined based on the recognition result, so as to perform corresponding processing.

[0225] On the basis of the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0226] The medical-related data acquisition module is used to input unlabeled medical-related data into the recognition model to determine the corresponding recognition results.

[0227] The medical-related data grouping module is used to group the medical-related data according to the credibility of the recognition result.

[0228] The medical-related data configuration module is used to configure the annotation data for each group of medical-related data in batches to form training data.

[0229] In the embodiment of the present application, medical-related data can be obtained, and the recognition model can be used to determine the recognition result of the medical-related data. Then, the medical-related data can be grouped according to the credibility of the recognition result, and the medical-related data of each group can be labeled in batches to form training data. After the training data is determined, the character recognition model can be trained based on the training data so that the medical-related information can be recognized by the character recognition model.

[0230] On the basis of the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0231] The education-related data determination module is used to obtain target image data.

[0232] The education-related data recognition module is used to determine the recognition result based on the target image data and the trained character recognition model. The character recognition model is trained based on multiple groups of labeled training data, and the multiple groups of training data are grouped based on the credibility of the image data input into the recognition model. The annotation of each group of training data is obtained by batch annotation.

[0233] The embodiments of the present application can be applied in educational scenarios, and the target image data can include images of student homework, images of answer sheets, images of school textbooks, etc. The embodiments of the present application can obtain image data related to education, and determine the recognition result through a trained character recognition model, and then perform corresponding processing based on the recognition result. For example, when the image data is an image of a student's answer sheet, the embodiments of the present application can obtain the recognition result, and determine whether the answer is correct based on the recognition result, and then determine the student's corresponding score.

[0234] On the basis of the above embodiment, this embodiment further provides a data processing device, which may specifically include the following modules:

[0235] The education-related data acquisition module is used to input unlabeled education-related data into the recognition model to determine the corresponding recognition results.

[0236] The education-related data grouping module is used to group the education-related data according to the credibility of the recognition results.

[0237] The education-related data configuration module is used to configure the labeled data for each group of education-related data in batches to form training data.

[0238] The embodiment of the present application can obtain education-related data and input the education-related data into a recognition model to determine a recognition result. Thereafter, the education-related data can be grouped based on the credibility of the recognition result, and the education-related data in each group can be batch-labeled to obtain training data for training a character recognition model. In the embodiment of the present application, the training data can be determined by batch-labeling the education-related data in each group, which can make data labeling simpler and more convenient.

[0239] The embodiment of the present application also provides a non-volatile readable storage medium, which stores one or more modules (programs). When the one or more modules are applied to a device, the device can execute instructions (instructions) of each method step in the embodiment of the present application.

[0240] The present application embodiment provides one or more machine-readable media, on which instructions are stored, and when executed by one or more processors, an electronic device executes one or more methods as described in the above embodiments. In the present application embodiment, the electronic device includes a server, a terminal device, and the like.

[0241] The embodiments of the present disclosure may be implemented as a device configured as desired using any appropriate hardware, firmware, software, or any combination thereof, and the device may include electronic devices such as a server (cluster), a terminal, etc. Fig.16An exemplary apparatus 1600 that can be used to implement various embodiments described in this application is schematically illustrated.

[0242] For one embodiment, Fig.16 An exemplary apparatus 1600 is shown having one or more processors 1602, a control module (chip set) 1604 coupled to at least one of the (one or more) processors 1602, a memory 1606 coupled to the control module 1604, a non-volatile memory (NVM) / storage device 1608 coupled to the control module 1604, one or more input / output devices 1610 coupled to the control module 1604, and a network interface 1612 coupled to the control module 1604.

[0243] Processor 1602 may include one or more single-core or multi-core processors, and processor 1602 may include any combination of general-purpose processors or special-purpose processors (such as graphics processors, application processors, baseband processors, etc.). In some embodiments, device 1600 can be used as a server, terminal, or other device described in the embodiments of the present application.

[0244] In some embodiments, the device 1600 may include one or more computer-readable media (e.g., memory 1606 or NVM / storage device 1608) having instructions 1614 and one or more processors 1602 configured to execute the instructions 1614 in combination with the one or more computer-readable media to implement the module to perform the actions described in the present disclosure.

[0245] For one embodiment, the control module 1604 may include any suitable interface controller to provide any suitable interface to at least one of the processor(s) 1602 and / or any suitable device or component in communication with the control module 1604 .

[0246] The control module 1604 may include a memory controller module to provide an interface to the memory 1606. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0247] The memory 1606 may be used, for example, to load and store data and / or instructions 1614 for the device 1600. For one embodiment, the memory 1606 may include any suitable volatile memory, such as a suitable DRAM. In some embodiments, the memory 1606 may include double data rate type four synchronous dynamic random access memory (DDR4 SDRAM).

[0248] For one embodiment, control module 1604 may include one or more input / output controllers to provide an interface to NVM / storage device 1608 and input / output device(s) 1610 .

[0249] For example, NVM / storage 1608 may be used to store data and / or instructions 1614. NVM / storage 1608 may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable non-volatile storage device(s) (e.g., one or more hard disk drives (HDDs), one or more compact disk (CD) drives, and / or one or more digital versatile disk (DVD) drives).

[0250] NVM / storage device 1608 may include storage resources that are part of the device on which apparatus 1600 is installed, or it may be accessible to the device without being part of the device. For example, NVM / storage device 1608 may be accessed via (one or more) input / output devices 1610 over a network.

[0251] (One or more) input / output devices 1610 may provide an interface for the apparatus 1600 to communicate with any other appropriate device, and the input / output device 1610 may include a communication component, an audio component, a sensor component, etc. The network interface 1612 may provide an interface for the apparatus 1600 to communicate through one or more networks, and the apparatus 1600 may wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, for example, accessing a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof for wireless communication.

[0252] For one embodiment, at least one of the processor(s) 1602 may be packaged together with the logic of one or more controllers (e.g., a memory controller module) of the control module 1604. For one embodiment, at least one of the processor(s) 1602 may be packaged together with the logic of one or more controllers of the control module 1604 to form a system-in-package (SiP). For one embodiment, at least one of the processor(s) 1602 may be integrated on the same die with the logic of one or more controllers of the control module 1604. For one embodiment, at least one of the processor(s) 1602 may be integrated on the same die with the logic of one or more controllers of the control module 1604 to form a system-on-chip (SoC).

[0253] In various embodiments, the device 1600 may be, but is not limited to, a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). In various embodiments, the device 1600 may have more or fewer components and / or a different architecture. For example, in some embodiments, the device 1600 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including a touch screen display), a non-volatile memory port, multiple antennas, a graphics chip, an application-specific integrated circuit (ASIC), and a speaker.

[0254] Among them, the main control chip can be used as a processor or control module in the detection device, sensor data, location information, etc. are stored in a memory or NVM / storage device, the sensor group can be used as an input / output device, and the communication interface may include a network interface.

[0255] An embodiment of the present application further provides an electronic device, comprising: a processor; and a memory, on which executable code is stored, and when the executable code is executed, the processor executes one or more methods described in the embodiments of the present application.

[0256] The embodiments of the present application also provide one or more machine-readable media on which executable codes are stored. When the executable codes are executed, the processor executes one or more methods described in the embodiments of the present application.

[0257] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0258] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0259] The present application embodiment is described with reference to the flowchart and / or block diagram of the method, terminal device (system) and computer program product according to the embodiment of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, and the combination of the process and / or box in the flowchart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device produce a device for realizing the function specified in one process or multiple processes in the flowchart and / or one box or multiple boxes in the block diagram.

[0260] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0261] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce computer-implemented processing, so that the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0262] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0263] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0264] The data processing method, the data processing device, the electronic device and the storage medium provided by the present application are introduced in detail above. The principles and implementation methods of the present application are explained in this article by using specific examples. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A data processing method, characterized in that: The method includes: Inputting unlabeled input data into a recognition model to determine a corresponding recognition result, wherein the recognition result includes a first recognition result and a second recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity, wherein the input data is an image containing characters; Match the feature class with the single word recognition result to determine the matching result; Determining the credibility of the recognition result according to the matching result, the feature similarity and the recognition confidence, and grouping the input data; The grouped input data is clustered according to the characteristics of the data, and the labeled data of each category of input data is determined in batches to form training data; Providing a data interaction page to obtain input data to be identified based on the data interaction page; Determine a recognition result based on the input data and a trained character recognition model, wherein the character recognition model is trained based on multiple groups of labeled training data, the multiple groups of training data are grouped based on the credibility of the input data input to the recognition model, and the labels of each group of training data are labeled based on a batch labeling method; The recognition result is issued.

2. The method according to claim 1, characterized in that The determining the credibility of the recognition result based on the matching result, the feature similarity and the recognition confidence, and grouping the input data includes: Comparing the feature similarity with a similarity threshold to obtain a first comparison result; Comparing the recognition confidence with the confidence threshold to obtain a second comparison result; The credibility of the recognition result is determined according to the first comparison result, the second comparison result and the matching result, and the input data is grouped.

3. The method according to claim 1, characterized in that The step of inputting the unlabeled input data into the recognition model to determine the corresponding recognition result includes: Inputting the unlabeled input data into a first recognition model to determine a first recognition result; The unlabeled input data is input into a second recognition model to determine a second recognition result.

4. The method according to claim 1, characterized in that Also includes: The image data to be processed is acquired, and the image data is recognized to obtain a plurality of sub-images containing a single word as input data.

5. The method according to claim 4, characterized in that Also includes: The sub-image is subjected to data adjustment to obtain an expanded sub-image as input data, wherein the data adjustment includes at least one of image rotation, image translation, image cropping, affine transformation, and noise enhancement processing.

6. The method according to claim 1, characterized in that The clustering process is performed on the grouped input data according to the characteristics of the data, and the labeled data of each category of the input data is determined in batches to form training data, which includes at least one of the following steps: When the group to which the input data belongs is the first group, obtaining a single-word recognition result of each input data as the marked data; When the group to which the input data belongs is the second group, clustering the input data according to the single-word recognition result of the input data, and obtaining annotation information of each category as the annotation data; When the group to which the input data belongs is the third group, the input data is used as input for the next iteration of the recognition model.

7. The method according to claim 6, characterized in that The first group includes a simple sample group with existing categories, the second group includes a similar sample group with existing categories, a difficult sample group with existing categories, and a sample group without categories; and the third group includes a low-quality sample group.

8. The method according to claim 1, characterized in that Also includes: An interactive page is provided to obtain annotation information of input data based on the interactive page.

9. The method according to claim 8, characterized in that The method further comprises: Receive adjustment information for input data to adjust the input data to a target category.

10. The method according to claim 8, characterized in that Also includes: The input data and the corresponding annotated data are output, and the corresponding correction information is obtained to adjust the annotated data according to the correction information.

11. A data processing method, characterized in that: The method comprises: Inputting unlabeled input data into a recognition model to determine a corresponding recognition result, wherein the recognition result includes a first recognition result and a second recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity, wherein the input data is an image containing characters; Match the feature class with the single word recognition result to determine the matching result; Determining the credibility of the recognition result according to the matching result, the feature similarity and the recognition confidence, and grouping the input data; The grouped input data is clustered according to the characteristics of the data, and the labeled data of each category of input data is determined in batches to form training data; Providing a data interaction page, wherein the data interaction page includes a data upload control; According to the triggering of the data upload control, the input data to be recognized is obtained and uploaded to the server, and the server is used to determine the recognition result according to the character recognition model and the input data, the character recognition model is trained according to multiple groups of labeled training data, the multiple groups of training data are grouped according to the credibility of the input data input to the recognition model, and the annotation of each group of training data is annotated according to the batch annotation method; Receive the recognition result.

12. A data processing method, characterized in that: The method comprises: Inputting unlabeled input data into a recognition model to determine a corresponding recognition result, wherein the recognition result includes a first recognition result and a second recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity, wherein the input data is an image containing characters, and the characters include ancient characters; Match the feature class with the single word recognition result to determine the matching result; Determining the credibility of the recognition result according to the matching result, the feature similarity and the recognition confidence, and grouping the input data; The grouped input data is clustered according to the characteristics of the data, and the labeled data of each category of input data is determined in batches to form training data; Providing a data interaction page to obtain ancient character image data to be recognized based on the data interaction page; Determine a recognition result based on the ancient character image data and a trained character recognition model, wherein the character recognition model is trained based on multiple groups of annotated training data, the multiple groups of training data are grouped based on the credibility of the ancient character image data input to the recognition model, and the annotations of each group of ancient character image data are annotated based on a batch annotation method; The recognition result is issued.

13. A data processing method, characterized in that: The method comprises: Inputting unlabeled input data into a recognition model to determine a corresponding recognition result, wherein the recognition result includes a first recognition result and a second recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity, wherein the input data is an image containing characters; Match the feature class with the single word recognition result to determine the matching result; Determining the credibility of the recognition result according to the matching result, the feature similarity and the recognition confidence, and grouping the input data; The grouped input data is clustered according to the characteristics of the data, and the labeled data of each category of input data is determined in batches to form training data.

14. A data processing method, characterized in that: The method comprises: Inputting unlabeled input data into a recognition model to determine a corresponding recognition result, wherein the recognition result includes a first recognition result and a second recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity, wherein the input data is an image containing characters; Match the feature class with the single word recognition result to determine the matching result; Determining the credibility of the recognition result according to the matching result, the feature similarity and the recognition confidence, and grouping the input data; The grouped input data is clustered according to the characteristics of the data, and the labeled data of each category of input data is determined in batches to form training data; The character recognition model is trained based on the training data so that the input data to be recognized can be recognized based on the trained character recognition model.

15. A data processing method, characterized in that: The method comprises: Providing a service page to obtain uploaded input data according to the service page, wherein the input data is an image containing characters; Inputting unlabeled input data into a recognition model to determine a corresponding recognition result, wherein the recognition result includes a first recognition result and a second recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity; Match the feature class with the single word recognition result to determine the matching result; Determining the credibility of the recognition result according to the matching result, the feature similarity and the recognition confidence, and grouping the input data; The grouped input data is clustered according to the characteristics of the data, and the labeled data of each category of input data is determined in batches to form training data.

16. The method according to claim 15, characterized in that The batch configuration of label data for each group of input data to form training data includes: The grouped input data is clustered according to the characteristics of the data, and the input data of each category is sent to the terminal for display on the service page; Obtain labeled data based on the service page so that the input data can be labeled in batches by category to form training data.

17. The method according to claim 16, characterized in that Also includes: receiving adjustment information for each category of input data; The input data of each category is adjusted according to the adjustment information, and the marking reference information is issued so as to mark the adjusted input data of each category.

18. A data processing method, characterized in that: The method comprises: Providing a service page, the service page including a data upload interface; According to the triggering of the data upload interface, input data is obtained and uploaded to a server, the server is used to input the unlabeled input data into a recognition model, determine a corresponding recognition result, the recognition result includes a first recognition result and a second recognition result, the first recognition result includes a single-word recognition result and a corresponding recognition confidence, the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity; match the feature class with the single-word recognition result to determine a matching result; according to the matching result, the feature similarity and the recognition confidence, determine the credibility of the recognition result, and group the input data, the input data is an image containing characters; Receive input data of each group sent by the server and display it on the service page; The labeling information of the input data of each group is received to form training data, wherein the labeling information of the input data of each group is determined in batches by clustering the grouped input data according to the characteristics of the data.

19. A data processing device, characterized in that: include: A recognition result acquisition module, used for inputting unlabeled input data into a recognition model to determine a corresponding recognition result, wherein the recognition result includes a first recognition result and a second recognition result, wherein the first recognition result includes a single-word recognition result and a corresponding recognition confidence, and the second recognition result includes a feature class corresponding to the input data and a corresponding feature similarity, wherein the input data is an image containing characters; A grouping result acquisition module is used to match the feature class with the single word recognition result to determine the matching result; determine the credibility of the recognition result based on the matching result, the feature similarity and the recognition confidence, and group the input data; The training data acquisition module is used to perform clustering processing on the grouped input data according to the characteristics of the data, and batch determine the labeling data of each category of input data to form training data; A data interaction page providing module, used for providing a data interaction page, so as to obtain input data to be identified based on the data interaction page; An input data recognition processing module is used to determine a recognition result based on the input data and a trained character recognition model, wherein the character recognition model is trained based on multiple groups of labeled training data, the multiple groups of training data are grouped based on the credibility of the input data input to the recognition model, and the labels of each group of training data are labeled based on a batch labeling method; The recognition result sending processing module is used to send the recognition result.

20. An electronic device, characterized in that: include: processor; and A memory having executable codes stored thereon, which, when executed, cause the processor to perform the method according to one or more of claims 1-18.

21. One or more machine-readable media having executable codes stored thereon, which, when executed, cause a processor to perform the method of one or more of claims 1-18.

Images