Field information extraction method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a field information extraction method and device, equipment and a storage medium, and relates to the field of artificial intelligence, and the method comprises the steps: obtaining a target picture, carrying out the information extraction of characters in the target picture, and obtaining a field extraction result; determining field extraction features corresponding to the target picture according to the field extraction result; traversing a plurality of feature classifiers, and determining a target feature classifier corresponding to the field extraction features; based on the target feature classifier, performing classification processing on the field extraction result to obtain a confidence score of each piece of field information in the field extraction result; performing bucket dividing processing on field information in the field extraction result according to the confidence score to obtain a target array bucket; and if the accuracy rate of the target array bucket is higher than a preset accuracy rate threshold value, taking the field information in the target array bucket as target field information. The method and the device aim at solving the problem that a field extraction result still needs to be verified manually.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a field information extraction method, a field information extraction device, a computer device, and a computer-readable storage medium. Background Art

[0002] At present, in medical scenarios, hospitals have accumulated a large amount of image data. In related technologies, structured information in image data, such as name, date, diagnosis information, etc., can be automatically extracted through information extraction models. At present, optical character recognition (OCR) + natural language processing (NLP) solutions are generally used for information extraction.

[0003] However, this solution cannot guarantee 100% accuracy. It may cause typos when performing OCR and entity boundary extraction errors when performing NLP, which may lead to errors in information extraction results. Moreover, taking multiple field information as an example, if the accuracy of these field information extraction is only 0.8 and the recall rate is 0.88, that is, the model extracts 10 results, of which it can be determined that 2 field information is extracted incorrectly, but it cannot determine which 2 field information is wrong. This will result in that when using the field extraction results, although 10 field information is provided, it still needs to be manually reviewed one by one to find out which field information is accurate, which is very cumbersome, inefficient, and low in intelligence. Summary of the invention

[0004] The present application provides a field information extraction method, a field information extraction device, a computer device and a computer-readable storage medium, aiming to solve the problem that field extraction results still need to be manually verified, resulting in low efficiency.

[0005] To achieve the above purpose, the present application provides a field information extraction method, the method comprising:

[0006] Obtain a target image, extract information from the text in the target image, and obtain a field extraction result;

[0007] Determine the field extraction feature corresponding to the target image according to the field extraction result;

[0008] Traversing multiple feature classifiers to determine a target feature classifier corresponding to the field extraction feature;

[0009] Based on the target feature classifier, the field extraction result is classified to obtain a confidence score of each field information in the field extraction result;

[0010] Bucketing the field information in the field extraction result according to the confidence score to obtain a target array bucket;

[0011] If the accuracy of the target array bucket is higher than a preset accuracy threshold, the field information in the target array bucket is used as the target field information.

[0012] To achieve the above object, the present application provides a field information extraction device, the field information extraction device comprising:

[0013] A field extraction module is used to obtain a target image, extract information from the text in the target image, and obtain a field extraction result;

[0014] A feature determination module, used to determine the field extraction feature corresponding to the target image according to the field extraction result;

[0015] A classifier determination module, used to traverse multiple feature classifiers and determine a target feature classifier corresponding to the field extraction feature;

[0016] A confidence determination module, used to classify the field extraction results based on the target feature classifier to obtain the confidence score of each field information in the field extraction results;

[0017] An information bucketing module, used to bucket the field information in the field extraction result according to the confidence score to obtain a target array bucket;

[0018] The information screening module is used to use the field information in the target array bucket as the target field information if the accuracy of the target array bucket is higher than a preset accuracy threshold.

[0019] To achieve the above-mentioned purpose, the present application provides a computer device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the field information extraction method as described above when executing the computer program.

[0020] To achieve the above objectives, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the field information extraction method as described above.

[0021] The field information extraction method, field information extraction device, computer equipment and computer-readable storage medium disclosed in the embodiment of the present application obtain a target image, extract information from the text in the target image, and obtain a field extraction result; determine the field extraction feature corresponding to the target image according to the field extraction result; traverse multiple feature classifiers to determine the target feature classifier corresponding to the field extraction feature; classify the field extraction feature based on the target feature classifier to obtain the confidence score of each field information in the field extraction result; bucket the field information in the field extraction result according to the confidence score to obtain a target array bucket; if the accuracy of the target array bucket is higher than the preset accuracy threshold, the field information in the target array bucket is used as the target field information. In this way, the confidence determination method can be introduced to improve the accuracy of the field extraction result, without the need for manual intervention to verify the field extraction result, thereby improving the efficiency and intelligence of field extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 It is a scenario diagram of a method for extracting field information provided by an embodiment of the present application;

[0024] Figure 2 It is a flowchart of a method for extracting field information provided by an embodiment of the present application;

[0025] Figure 3 is a schematic block diagram of a field information extraction device provided in one embodiment of the present application;

[0026] Figure 4 It is a schematic block diagram of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor do they have to be executed in the order described. For example, some operations / steps can also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions. In addition, although the functional modules are divided in the device schematic, in some cases, the module division can be different from that in the device schematic.

[0029] The term "and / or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0030] The present application proposes a field information extraction method, a field information extraction device, a computer device and a computer-readable storage medium, which can introduce a confidence judgment method to improve the accuracy of the field extraction results without the need for manual intervention to verify the field extraction results, thereby improving the efficiency and intelligence of field extraction.

[0031] The method can be applied to a server, and of course can also be applied to a terminal device, wherein the terminal device may include a fixed terminal such as a mobile phone, a tablet computer, a personal digital assistant (PDA), etc. The server may be, for example, a separate server or a server cluster. However, for ease of understanding, the following embodiments will be described in detail with the field information extraction method applied to the server.

[0032] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0033] like Figure 1 As shown, the field information extraction method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. The application environment includes a terminal device 110 and a server 120, wherein the terminal device 110 can communicate with the server 120 through a network. Specifically, the server 120 obtains a target image, extracts information from the text in the target image, and obtains a field extraction result; determines the field extraction feature corresponding to the target image according to the field extraction result; traverses multiple feature classifiers to determine the target feature classifier corresponding to the field extraction feature; based on the target feature classifier, classifies the field extraction result to obtain the confidence score of each field information in the field extraction result; buckets the field information in the field extraction result according to the confidence score to obtain a target array bucket; if the accuracy of the target array bucket is higher than the preset accuracy threshold, the field information in the target array bucket is used as the target field information, and the target field information is sent to the terminal device 110. Among them, the server 120 can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. The terminal device 110 can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected via wired or wireless communication, and this application does not limit this.

[0034] See also Figure 2 , Figure 2 This is a schematic flow chart of a field information extraction method provided in an embodiment of the present application. The field information extraction method can introduce a confidence determination method to improve the accuracy of the field extraction result, without the need for manual intervention to verify the field extraction result, thereby improving the efficiency and intelligence of field extraction.

[0035] like Figure 2 As shown, the field information extraction method includes steps S101 to S106.

[0036] S101, obtaining a target image, extracting information from text in the target image, and obtaining a field extraction result.

[0037] The target image may be an image that needs information extraction, that is, an image containing text information, such as an image corresponding to a diagnosis sheet, etc. The field extraction result is the field information extracted from the target image.

[0038] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0039] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0040] Specifically, optical character recognition (OCR) technology and natural language processing (NLP) technology can be used to extract information from text in the target image, thereby obtaining a field extraction result corresponding to the target image.

[0041] Exemplarily, the picture corresponding to the diagnosis form can be input into the information extraction model, so that the information extraction model extracts information from the insurance policy picture and obtains the field extraction result corresponding to the diagnosis form.

[0042] It should be noted that any method may be used to extract information from the text in the target image to obtain a field extraction result, and no specific limitation is made here.

[0043] S102: Determine the field extraction features corresponding to the target image according to the field extraction result.

[0044] The field extraction feature may be feature information in the information extraction process, such as OCR features, grouping features, normalization features, and entity boundary features.

[0045] Since the information extraction process involves many stages of information processing, errors may occur in each stage of information processing, resulting in errors in the final field extraction results. For example, taking the diagnosis field as an example, for the diagnosis field, it is generally necessary to first group the reports, then extract the discharge Western medicine diagnosis information in the discharge record, and finally obtain the field extraction result through normalization and deduplication. In the above information processing process, problems such as grouping, paragraphing, normalization, entity boundaries, and OCR typos will cause errors in the final field extraction results. Therefore, it is necessary to analyze the relevant features based on the error situation. The wider the error situation analysis, the more comprehensive the design features need to be.

[0046] In some embodiments, the field extraction result is parsed to obtain field semantic information; the business scenario corresponding to the target image is determined according to the field semantic information; and the field extraction features corresponding to the target image are determined according to the business scenario. Thus, the field extraction features corresponding to the target image can be accurately determined according to the business scenario.

[0047] The field semantic information may be the semantic information of each field information in the field extraction result. Business scenarios may include medical scenarios, insurance scenarios, and other business scenarios. Different business scenarios generally correspond to different field extraction features.

[0048] Specifically, semantic prediction can be performed on the field extraction results to obtain the semantic information of each field information in the field extraction results; the semantic information can then be used to perform scene matching to determine the business scenario corresponding to the target image; finally, the field extraction features corresponding to the target image are determined based on the business scenario.

[0049] Exemplarily, semantic prediction can be performed on the field extraction results, so that the semantic information of each field information in the field extraction results may include information such as A hospital, surgery, fracture, surgery, etc., and scene matching can be performed based on the above semantic information to determine that the business scenario corresponding to the target image is a medical scenario. Finally, the field extraction features corresponding to the target image are determined based on the medical scenario.

[0050] In some embodiments, the field extraction method corresponding to the target image is determined according to the business scenario; the risk information corresponding to the field extraction method is determined; and the field extraction features corresponding to the target image are determined according to the risk information. Thus, the field extraction features corresponding to the target image can be accurately determined according to the field extraction method.

[0051] The field extraction method can be an information processing method in the information extraction process, which can mainly include image recognition processing, feature extraction processing, and semantic prediction processing. Each field extraction method corresponds to a field extraction feature. Risk information can be used to indicate the probability that each field extraction method may have errors in the corresponding business scenario.

[0052] Specifically, the field extraction method corresponding to the target image can be determined first according to the business scenario; then the probability of errors in each field extraction method under the business scenario can be determined; if the probability of errors in a certain field extraction method under the business scenario is higher than a preset probability threshold, the field extraction feature corresponding to the field extraction method is used as the field extraction feature corresponding to the target image.

[0053] Exemplarily, if the business scenario is a medical scenario, it can be determined first that the field extraction method corresponding to the diagnosis form may include image recognition processing, feature extraction processing, and semantic prediction processing. If the probability of error in each of the above field extraction methods in this business scenario is 85%, 90%, and 60%, and the preset probability threshold is 80%, then it can be determined that the image recognition processing and feature extraction processing have a higher probability of error. The field extraction features corresponding to the image recognition processing are OCR features and grouping features, and the field extraction features corresponding to the feature extraction processing are entity boundary features, that is, the grouping features, OCR features, and entity boundary features are used as the field extraction features corresponding to the target image.

[0054] S103, traverse multiple feature classifiers to determine the target feature classifier corresponding to the field extraction feature.

[0055] Among them, the multiple feature classifiers may include classifiers corresponding to different field extraction scenarios, which are used to judge the field information in the field extraction result to determine whether the field extraction result is correct. The target feature classifier is used to classify the field extraction result to determine the confidence score of each field information. Different target feature classifiers have different confidence scores for the same field information.

[0056] In some embodiments, a preset feature classifier and a preset feature data set are obtained, wherein the preset feature data set includes multiple field extraction features; the multiple field extraction features are classified to obtain category information of each field extraction feature; and the preset feature classifier is iteratively trained according to the category information of each field extraction feature to obtain a feature classifier. In this way, a feature classifier for classifying different features can be obtained by training according to actual needs.

[0057] Among them, the preset feature classifier is an untrained feature classifier. The preset feature data set may include multiple field extraction features of different categories. The category information of the field extraction feature is used to represent the categories of different features, such as grouping processing category, OCR processing category, normalization processing category, boundary extraction category, etc. The field extraction features of different category information are used to train feature classifiers for classifying corresponding features.

[0058] Exemplarily, a preset feature classifier and a preset feature data set are first obtained, and the preset feature data set includes multiple field extraction features; the multiple field extraction features are classified, and the category information of each field extraction feature can include a grouping processing category, an OCR processing category, a normalization processing category, and a boundary extraction category. Then, the preset feature classifier can be iteratively trained using the field extraction features of the grouping processing category to obtain a feature classifier for classifying grouping features; the preset feature classifier can also be iteratively trained using the field extraction features of the OCR processing category to obtain a feature classifier for classifying OCR features; the preset feature classifier can also be iteratively trained using the field extraction features of the normalization processing category to obtain a feature classifier for classifying normalization features; the preset feature classifier can also be iteratively trained using the field extraction features of the boundary extraction processing category to obtain a feature classifier for classifying boundary extraction features.

[0059] Among them, in order to optimize the accuracy of the feature classifier, it is necessary to analyze the arguments that lead to positive examples and negative examples, and then use the arguments as field extraction features to train the feature classifier. By training the feature classifier iteratively for multiple times, the optimization efficiency of the feature classifier can be improved.

[0060] Exemplarily, multiple field extraction features in a preset feature data set can be analyzed to find the most widely influential and direct features as the field extraction features of the preset feature classifier; secondly, based on the negative examples output by the preset feature classifier, analyze why the feature classifier misclassifies, thereby analyzing the lower-frequency field extraction features, and then retrain the feature classifier based on the new field extraction features. Through repeated iterative training, the classification effect of the feature classifier will be continuously optimized.

[0061] In some embodiments, the field extraction features are parsed to obtain the field extraction logic; the field extraction scenario corresponding to the field extraction logic is determined, and the similarity between the field extraction scenario and the preset field extraction scenario of each feature classifier is determined; if the similarity between the field extraction scenario and the preset field extraction scenario of the feature classifier exceeds a preset similarity threshold, the feature classifier corresponding to the preset field extraction scenario is used as the target feature classifier. In this way, the target feature classifier can be accurately determined to classify the field extraction results.

[0062] Among them, the field extraction logic can be used to represent the extraction logic information in the field extraction process. The field extraction scenario can be various information processing scenarios such as image recognition scenarios, feature extraction scenarios, etc., corresponding to the category information of the field extraction feature. The preset similarity threshold can be any similarity, which is not specifically limited here. The preset field extraction scenario can be a grouping processing scenario, an OCR processing scenario, a normalization processing scenario, a boundary extraction scenario, etc.

[0063] Exemplarily, the field extraction features can be analyzed to obtain the field extraction logic corresponding to the image recognition logic, and the field extraction scenario corresponding to the field extraction logic can be the image recognition scenario. If the similarity threshold is 80%, and the similarities between the image recognition scenario and the grouping processing scenario, the OCR processing scenario, the normalization processing scenario, and the boundary extraction scenario are determined to be 20%, 90%, 40%, and 30%, respectively, and the similarity between the image recognition scene and the OCR processing scene exceeds the similarity threshold, the feature classifier corresponding to the OCR processing scene is used as the target feature classifier. It should be noted that the feature classifier corresponding to the OCR processing scenario is the feature classifier corresponding to the OCR feature.

[0064] S104: Based on the target feature classifier, the field extraction results are classified and processed to obtain the confidence score of each field information in the field extraction results.

[0065] The confidence score may include extraction score, recognition score, frequency score, normalization score, entity boundary score, etc. The confidence score may be determined by the confidence level. If the confidence level is 0.6, the corresponding confidence score may be 60 points.

[0066] Exemplarily, if the target feature classifier includes a feature classifier corresponding to the OCR feature and a feature classifier corresponding to the entity boundary feature, the field extraction result is first classified based on the feature classifier corresponding to the OCR feature to obtain the classification probability of each field information in the field extraction result, thereby determining the recognition score; then, based on the feature classifier corresponding to the entity boundary feature, the field extraction result is classified to obtain the classification probability of each field information in the field extraction result, thereby determining the entity boundary score.

[0067] Among them, the classification probability predicted by the feature classifier can reflect the confidence of the prediction result. The greater the probability of positive examples, the higher the classification accuracy. However, there is a certain difference between the score and the actual accuracy. That is, when the probability of positive examples is 0.6, it does not mean that the feature classifier has 60% accuracy. Therefore, the classification probability predicted by the feature classifier needs to be calibrated later.

[0068] It should be noted that if the target feature classifier includes multiple feature classifiers, there are corresponding confidence scores generated by multiple feature classifiers.

[0069] S105 . Bucket the field information in the field extraction result according to the confidence score to obtain a target array bucket.

[0070] The target array bucket is an array bucket composed of field information with high confidence.

[0071] In some embodiments, each field information in the field extraction result is bucketed according to the confidence score to obtain multiple array buckets, and the multiple array buckets are screened to obtain the target array bucket. In this way, the confidence score can be accurately bucketed and the target array bucket can be screened.

[0072] Among them, the array bucket can be composed of field information with different confidence score ranges.

[0073] Specifically, the confidence mean can be determined according to the extraction score, recognition score, frequency score, normalization score and entity boundary score; and each field information in the field extraction result is bucketed according to the confidence mean to obtain multiple array buckets.

[0074] The confidence mean may be the mean of the extraction score, the recognition score, the frequency score, the normalization score, and the entity boundary score.

[0075] Exemplarily, the interval of the confidence mean [0, 1] can be divided into n buckets in equal proportions, and the field information belongs to the bucket in which the confidence mean falls.

[0076] For example, assume that the interval of the confidence mean [0, 1] is divided into n buckets in equal proportions, and the interval range of each bucket is 0.1. If the extraction score, recognition score, frequency score, normalization score, and entity boundary score of field information A are 90, 80, 85, 95, and 70 points respectively, then the average value of the calculated confidence score is 84 points, that is, the confidence mean of field information A is 0.84; if the extraction score, recognition score, frequency score, normalization score, and entity boundary score of field information B are 90, 80, 85, 95, and 70 points respectively, then the average value of the calculated confidence score is 84 points, that is, the confidence mean of field information A is 0.84; if the extraction score, recognition score, frequency score, normalization score, and entity boundary score of field information B are 90, 80, 85, 95, and 70 points respectively, then the average value of the calculated confidence score is 84 points, that is, the confidence mean of field information A is 0.84; The extraction score, recognition score, frequency score, normalization score and entity boundary score of field information B are 80 points, 85 points, 75 points, 75 points and 80 points respectively. The calculated average confidence score is 79 points, that is, the average confidence of field information B is 0.79; therefore, field information A will be divided into the array bucket of [0.8, 0.9], and field information B will be divided into the array bucket of [0.7, 0.8], and so on. Each field information in the field extraction result is bucketed to assign each field information to the corresponding array bucket, thereby obtaining multiple array buckets.

[0077] It should be noted that the confidence score may include one or more of an extraction score, a recognition score, a frequency score, a normalized score, and an entity boundary score, which may be determined based on actual conditions and is not specifically limited here.

[0078] Specifically, the array bucket whose confidence mean is higher than the confidence threshold is used as the target array bucket.

[0079] The confidence threshold may be any confidence level and is not specifically limited here.

[0080] For example, if the confidence threshold is 0.8, the field information in the array bucket [0.8, 0.9] and the array bucket [0.9, 1] will both be higher than the confidence threshold, and the array bucket [0.8, 0.9] and the array bucket [0.9, 1] can be used as the target array bucket.

[0081] S106: If the accuracy of the target array bucket is higher than a preset accuracy threshold, the field information in the target array bucket is used as the target field information.

[0082] The preset accuracy threshold may be any accuracy, which is not specifically limited here. The target field information may be field information that meets the confidence requirement and the accuracy requirement, that is, the target field information can guarantee a high confidence and accuracy without manual verification.

[0083] Specifically, the accuracy of the target array bucket is calculated. If the accuracy of the target array bucket is higher than the preset accuracy threshold, the field information in the target array bucket is used as the target field information; if the accuracy of the target array bucket is not higher than the preset accuracy threshold, it is determined that the target field information does not exist in the field extraction result, that is, the field extraction result needs to be manually verified.

[0084] For example, the accuracy of the target array bucket can be calculated as 95% by the number of positive examples in the target array bucket and the number of bucket samples. If the preset accuracy threshold is 90%, the field information in the target array bucket is used as the target field information. Since the field information of the target array bucket meets the confidence requirements and the accuracy requirements, that is, the target field information can guarantee a high confidence and accuracy, and no manual verification is required.

[0085] The field information extraction method disclosed in the embodiment of the present application obtains a target image, extracts information from the text in the target image, and obtains a field extraction result; determines the field extraction feature corresponding to the target image according to the field extraction result; traverses multiple feature classifiers to determine the target feature classifier corresponding to the field extraction feature; classifies the field extraction feature based on the target feature classifier to obtain the confidence score of each field information in the field extraction result; buckets the field information in the field extraction result according to the confidence score to obtain a target array bucket; if the accuracy of the target array bucket is higher than the preset accuracy threshold, the field information in the target array bucket is used as the target field information. In this way, the confidence determination method can be introduced to improve the accuracy of the field extraction result, without the need for manual intervention to verify the field extraction result, thereby improving the efficiency and intelligence of field extraction.

[0086] See also Figure 3 , Figure 3 It is a schematic block diagram of a field information extraction device provided in an embodiment of the present application. The field information extraction device can be configured in a server to execute the aforementioned field information extraction method.

[0087] like Figure 3 As shown, the field information extraction device 200 includes: a field extraction module 201, a feature determination module, a classifier determination module 203, a confidence determination module 204, an information bucketing module 205 and an information screening module 206.

[0088] The field extraction module 201 is used to obtain a target image, extract information from the text in the target image, and obtain a field extraction result;

[0089] A feature determination module 202 is used to determine the field extraction feature corresponding to the target image according to the field extraction result;

[0090] A classifier determination module 203 is used to traverse multiple feature classifiers and determine a target feature classifier corresponding to the field extraction feature;

[0091] A confidence determination module 204, configured to classify the field extraction result based on the target feature classifier to obtain a confidence score of each field information in the field extraction result;

[0092] An information bucketing module 205 is used to bucket the field information in the field extraction result according to the confidence score to obtain a target array bucket;

[0093] The information screening module 206 is configured to use the field information in the target array bucket as the target field information if the accuracy of the target array bucket is higher than a preset accuracy threshold.

[0094] In some embodiments, the feature determination module 202 is also used to parse the field extraction results to obtain field semantic information; determine the business scenario corresponding to the target image based on the field semantic information; and determine the field extraction features corresponding to the target image based on the business scenario.

[0095] In some embodiments, the feature determination module 202 is also used to determine the field extraction method corresponding to the target image according to the business scenario; determine the risk information corresponding to the field extraction method; and determine the field extraction feature corresponding to the target image according to the risk information.

[0096] In some embodiments, the field information extraction device 200 also includes a classifier generation module 207, which is used to obtain a preset feature classifier and a preset feature data set, wherein the preset feature data set includes multiple field extraction features; classify the multiple field extraction features to obtain category information of each of the field extraction features; and iteratively train the preset feature classifier according to the category information of each of the field extraction features to obtain the feature classifier.

[0097] In some embodiments, the classifier determination module 203 is also used to parse the field extraction features to obtain field extraction logic; determine the field extraction scenario corresponding to the field extraction logic, and determine the similarity between the field extraction scenario and the preset field extraction scenario of each feature classifier; if the similarity between the field extraction scenario and the preset field extraction scenario of the feature classifier exceeds a preset similarity threshold, the feature classifier corresponding to the preset field extraction scenario is used as the target feature classifier.

[0098] In some embodiments, the information bucketing module 205 is further used to bucket each field information in the field extraction result according to the confidence score to obtain a plurality of array buckets; and filter the plurality of array buckets to obtain a target array bucket.

[0099] In some embodiments, the information bucketing module 205 is also used to determine a confidence mean based on the extraction score, recognition score, frequency score, normalization score and entity boundary score; bucket each field information in the field extraction result according to the confidence mean to obtain multiple array buckets; and use the array bucket whose confidence mean is higher than the confidence threshold as the target array bucket.

[0100] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0101] The method and apparatus of the present application can be used in many general or special computing system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer terminal devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices.

[0102] Exemplarily, the above method and apparatus may be implemented in the form of a computer program. The computer program may be implemented in Figure 4 Runs on the computer device shown.

[0103] See also Figure 4 , Figure 4 1 is a schematic diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0104] like Figure 4 As shown, the computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory may include a volatile storage medium, a non-volatile storage medium and an internal memory.

[0105] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any field information extraction method.

[0106] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0107] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any field information extraction method.

[0108] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that the structure of the computer device is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0109] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0110] Among them, in some embodiments, the processor is used to run a computer program stored in the memory to implement the following steps: obtain a target image, extract information from the text in the target image, and obtain a field extraction result; determine the field extraction feature corresponding to the target image according to the field extraction result; traverse multiple feature classifiers to determine the target feature classifier corresponding to the field extraction feature; based on the target feature classifier, classify the field extraction result to obtain the confidence score of each field information in the field extraction result; bucket the field information in the field extraction result according to the confidence score to obtain a target array bucket; if the accuracy of the target array bucket is higher than a preset accuracy threshold, the field information in the target array bucket is used as the target field information.

[0111] In some embodiments, the processor is further used to parse the field extraction result to obtain field semantic information; determine the business scenario corresponding to the target image based on the field semantic information; and determine the field extraction features corresponding to the target image based on the business scenario.

[0112] In some embodiments, the processor is further used to determine a field extraction method corresponding to the target image according to the business scenario; determine risk information corresponding to the field extraction method; and determine field extraction features corresponding to the target image according to the risk information.

[0113] In some embodiments, the processor is also used to obtain a preset feature classifier and a preset feature data set, wherein the preset feature data set includes multiple field extraction features; classify the multiple field extraction features to obtain category information of each of the field extraction features; and iteratively train the preset feature classifier based on the category information of each of the field extraction features to obtain the feature classifier.

[0114] In some embodiments, the processor is also used to parse the field extraction features to obtain field extraction logic; determine the field extraction scenario corresponding to the field extraction logic, and determine the similarity between the field extraction scenario and the preset field extraction scenario of each feature classifier; if the similarity between the field extraction scenario and the preset field extraction scenario of the feature classifier exceeds a preset similarity threshold, the feature classifier corresponding to the preset field extraction scenario is used as the target feature classifier.

[0115] In some implementations, the processor is further configured to perform bucket processing on each field information in the field extraction result according to the confidence score to obtain a plurality of array buckets; and screen the plurality of array buckets to obtain a target array bucket.

[0116] In some embodiments, the processor is further used to determine a confidence mean based on the extraction score, recognition score, frequency score, normalization score, and entity boundary score; bucket each field information in the field extraction result according to the confidence mean to obtain multiple array buckets; and use the array bucket whose confidence mean is higher than the confidence threshold as the target array bucket.

[0117] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions, and when the program instructions are executed, any one of the field information extraction methods provided in the embodiments of the present application is implemented.

[0118] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.

[0119] Furthermore, the computer-readable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the blockchain node, etc.

[0120] The present invention refers to a new application model of computer technologies such as storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. of the blockchain language model. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the blockchain underlying platform, platform product service layer, and application service layer.

[0121] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for extracting field information, characterized in that: The method comprises: Obtain a target image, extract information from the text in the target image, and obtain a field extraction result; Determine the field extraction feature corresponding to the target image according to the field extraction result; Traversing multiple feature classifiers to determine a target feature classifier corresponding to the field extraction feature; Based on the target feature classifier, the field extraction result is classified to obtain the confidence score of each field information in the field extraction result; Bucketing the field information in the field extraction result according to the confidence score to obtain a target array bucket; If the accuracy of the target array bucket is higher than a preset accuracy threshold, the field information in the target array bucket is used as the target field information.

2. The method according to claim 1, characterized in that The determining, according to the field extraction result, the field extraction feature corresponding to the target image includes: Parsing the field extraction result to obtain field semantic information; Determining the business scenario corresponding to the target image according to the field semantic information; Determine the field extraction features corresponding to the target image according to the business scenario.

3. The method according to claim 2, characterized in that The step of determining the field extraction feature corresponding to the target image according to the business scenario includes: Determine a field extraction method corresponding to the target image according to the business scenario; Determining risk information corresponding to the field extraction method; Determine a field extraction feature corresponding to the target image according to the risk information.

4. The method according to claim 1, characterized in that The method further comprises: Obtaining a preset feature classifier and a preset feature data set, wherein the preset feature data set includes a plurality of field extraction features; Classifying the plurality of field extraction features to obtain category information of each of the field extraction features; According to the category information of the features extracted from each of the fields, a preset feature classifier is iteratively trained to obtain the feature classifier.

5. The method according to claim 1, characterized in that The traversing multiple feature classifiers to determine a target feature classifier corresponding to the field extraction feature includes: Parsing the field extraction features to obtain field extraction logic; Determining a field extraction scenario corresponding to the field extraction logic, and determining a similarity between the field extraction scenario and a preset field extraction scenario of each of the feature classifiers; If the similarity between the field extraction scenario and the preset field extraction scenario of the feature classifier exceeds a preset similarity threshold, the feature classifier corresponding to the preset field extraction scenario is used as the target feature classifier.

6. The method according to claim 1, characterized in that The step of performing bucketing processing on the field information in the field extraction result according to the confidence score to obtain a target array bucket includes: Bucketing each field information in the field extraction result according to the confidence score to obtain a plurality of array buckets; The multiple array buckets are screened to obtain a target array bucket.

7. The method according to claim 6, characterized in that The confidence score includes an extraction score, a recognition score, a frequency score, a normalization score, and an entity boundary score; and the field information in the field extraction result is bucketed according to the confidence score to obtain a plurality of array buckets, including: Determine the confidence mean based on the extraction score, recognition score, frequency score, normalization score and entity boundary score; Performing bucket processing on each field information in the field extraction result according to the confidence mean value to obtain a plurality of array buckets; The screening of the plurality of array buckets to obtain a target array bucket includes: The array bucket whose confidence mean is higher than the confidence threshold is used as the target array bucket.

8. A field information extraction device, characterized in that: The field information extraction device comprises: A field extraction module is used to obtain a target image, extract information from the text in the target image, and obtain a field extraction result; A feature determination module, used to determine the field extraction feature corresponding to the target image according to the field extraction result; A classifier determination module, used to traverse multiple feature classifiers and determine a target feature classifier corresponding to the field extraction feature; A confidence determination module, used to classify the field extraction results based on the target feature classifier to obtain the confidence score of each field information in the field extraction results; An information bucketing module, used to bucket the field information in the field extraction result according to the confidence score to obtain a target array bucket; The information screening module is used to use the field information in the target array bucket as the target field information if the accuracy of the target array bucket is higher than a preset accuracy threshold.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and implement, when executing the computer program: The field information extraction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the field information extraction method according to any one of claims 1 to 7.

Images