CN112926654B - Pre-labeling model training and certificate pre-labeling method, device, equipment and medium - Google Patents

Pre-labeling model training and certificate pre-labeling method, device, equipment and medium Download PDF

Info

Publication number
CN112926654B
CN112926654B CN202110211382.2A CN202110211382A CN112926654B CN 112926654 B CN112926654 B CN 112926654B CN 202110211382 A CN202110211382 A CN 202110211382A CN 112926654 B CN112926654 B CN 112926654B
Authority
CN
China
Prior art keywords
model
target
migrated
labeling
certificate
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110211382.2A
Other languages
Chinese (zh)
Other versions
CN112926654A (en
Inventor
王晟宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Bank Co Ltd
Original Assignee
Ping An Bank Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Bank Co Ltd filed Critical Ping An Bank Co Ltd
Priority to CN202110211382.2A priority Critical patent/CN112926654B/en
Publication of CN112926654A publication Critical patent/CN112926654A/en
Application granted granted Critical
Publication of CN112926654B publication Critical patent/CN112926654B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to the field of classification models of artificial intelligence, and provides a method, a device, equipment and a medium for training a pre-labeling model and pre-labeling a certificate, wherein the method comprises the following steps: obtaining a target annotation category, a target description, a model performance parameter and an image sample set; crawling the category to be migrated in the target classification and identification library by using a text similarity technology; searching a model to be migrated from a target classification recognition library and recognizing target areas of all image samples by a simulated target recognition technology; performing target fine adjustment to obtain a fine adjustment region, and inputting an image sample, the fine adjustment region and a target labeling category into a model to be migrated; obtaining a marked target marking area by using a transfer learning technology; determining a loss value according to the target labeling area and the fine tuning area; training the model to be migrated until training is completed to obtain the pre-labeling model. The invention realizes the automatic training of the image sample set of the zero mark annotation, obtains the pre-marking model, and reduces the manual marking time and the workload.

Description

Pre-labeling model training and certificate pre-labeling method, device, equipment and medium
Technical Field
The invention relates to the field of classification models of artificial intelligence, in particular to a pre-labeling model training method, a certificate pre-labeling method, a device, computer equipment and a storage medium.
Background
The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions. Among them, computer Vision (CV) is a science of how to "look" a machine, and generally includes technologies such as image processing, image recognition, image semantic understanding, image retrieval, and optical character recognition (OCR, optical Character Recognition).
As artificial intelligence technology is mature, image recognition technology is increasingly widely applied to daily life. In order to obtain an image recognition model with higher recognition accuracy, a large number of marked samples are needed to train the image recognition model, and in the prior art, manual input and other manual marking modes are generally adopted to realize the training of the sample, so that labor cost is consumed, the marking efficiency of the sample is greatly reduced, and great difficulty is brought to model training.
Disclosure of Invention
The invention provides a pre-labeling model training method, a pre-labeling device, computer equipment and a storage medium, which realize automatic training of zero-labeling image sample sets through a text similarity technology, a crawling technology, a simulation target recognition technology and a transfer learning technology, reduce manual labeling time and workload, improve labeling efficiency, save input cost and improve pre-labeling accuracy.
A pre-labeling model training method comprises the following steps:
acquiring a target annotation category, a target description corresponding to the target annotation category, a model performance parameter and an image sample set; the set of image samples includes at least one image sample; the image sample corresponds to the target annotation class;
the method comprises the steps of crawling historical categories similar to target descriptions in a target classification recognition library by using a text similarity technology, and determining the crawled historical categories as categories to be migrated;
searching a model to be migrated matched with the category to be migrated and the model performance parameter from the target classification recognition library through a simulation target recognition technology, and recognizing target areas of the image samples through the model to be migrated;
Performing target fine adjustment on all the target areas to obtain fine adjustment areas corresponding to the image samples, and inputting the image samples, the fine adjustment areas corresponding to the image samples and the target annotation categories into the model to be migrated; the model to be migrated comprises migration parameters;
obtaining a target labeling area corresponding to the image sample, which is labeled by the model to be migrated, through extracting the self-adaptive learning target category characteristics of the model to be migrated by using a migration learning technology;
determining a loss value according to the target labeling area and the fine tuning area corresponding to the image sample;
and when the loss value does not reach a preset convergence condition, iteratively updating the migration parameters of the model to be migrated, and recording the model to be migrated after convergence as a pre-labeled model after training is completed when the loss value reaches the preset convergence condition.
A method for pre-labeling a document, comprising:
receiving an image annotation instruction, and acquiring a certificate image in the image annotation instruction;
inputting the certificate image into a certificate pre-labeling model trained by the pre-labeling model training method, extracting learned certificate characteristics through the certificate pre-labeling model, and obtaining a labeling result output by the certificate pre-labeling model according to the certificate characteristics; the certificate pre-labeling model is a model trained by taking a certificate as a target labeling category, a certificate description as a target description, preset certificate model performance parameters and a certificate photo collected in a history as an image sample set; and marking the region containing the certificate in the certificate image by the marking result.
A pre-labeling model training device, comprising:
the acquisition module is used for acquiring a target annotation category, a target description corresponding to the target annotation category, a model performance parameter and an image sample set; the set of image samples includes at least one image sample; the image sample corresponds to the target annotation class;
the crawling module is used for crawling historical categories similar to the target description in the target classification recognition library by using a text similarity technology, and determining the crawled historical categories as categories to be migrated;
the identification module is used for searching a model to be migrated, which is matched with the category to be migrated and the model performance parameter, from the target classification identification library through a simulated target identification technology, and identifying target areas of the image samples through the model to be migrated;
the fine adjustment module is used for carrying out target fine adjustment on all the target areas to obtain fine adjustment areas corresponding to the image samples, and inputting the image samples, the fine adjustment areas corresponding to the image samples and the target annotation categories into the model to be migrated; the model to be migrated comprises migration parameters;
The migration module is used for obtaining a target labeling area, corresponding to the image sample, of the model to be migrated through extracting the self-adaptive learning target category characteristics of the model to be migrated by using a migration learning technology;
the loss module is used for determining a loss value according to the target labeling area and the fine tuning area corresponding to the image sample;
and the training module is used for iteratively updating the migration parameters of the model to be migrated when the loss value does not reach the preset convergence condition, and recording the model to be migrated after convergence as a pre-labeled model after training is completed when the loss value reaches the preset convergence condition.
A document pre-labeling apparatus comprising:
the receiving module is used for receiving the image annotation instruction and acquiring the certificate image in the image annotation instruction;
the marking module is used for inputting the certificate image into the certificate pre-marking model trained by the pre-marking model training method, extracting the learned certificate characteristics through the certificate pre-marking model, and obtaining a marking result output by the certificate pre-marking model according to the certificate characteristics; the certificate pre-labeling model is a model trained by taking a certificate as a target labeling category, a certificate description as a target description, preset certificate model performance parameters and a certificate photo collected in a history as an image sample set; and marking the region containing the certificate in the certificate image by the marking result.
A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the pre-labeling model training method described above when the computer program is executed or the processor implementing the steps of the certificate pre-labeling method described above when the computer program is executed.
A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps of the pre-labeling model training method described above, or the computer program when executed by a processor implements the steps of the certificate pre-labeling method described above.
According to the pre-labeling model training method, the pre-labeling model training device, the computer equipment and the storage medium, a target labeling category, a target description corresponding to the target labeling category, a model performance parameter and an image sample set are obtained; the method comprises the steps of crawling historical categories similar to target descriptions in a target classification recognition library by using a text similarity technology, and determining the crawled historical categories as categories to be migrated; searching a model to be migrated matched with the category to be migrated and the model performance parameter from the target classification recognition library through a simulation target recognition technology, and recognizing target areas of the image samples through the model to be migrated; performing target fine adjustment on all the target areas to obtain fine adjustment areas corresponding to the image samples, and inputting the image samples, the fine adjustment areas corresponding to the image samples and the target annotation categories into the model to be migrated; obtaining a target labeling area marked by the model to be migrated through extracting the class characteristics of the target to be migrated in a self-adaptive learning manner by using a migration learning technology; determining a loss value according to the target labeling area and the fine tuning area; when the loss value does not reach the preset convergence condition, iteratively updating the migration parameters of the model to be migrated until the loss value reaches the preset convergence condition, recording the model to be migrated after convergence as a pre-labeling model after training, thus realizing automatic training of an image sample set with zero labeling through a text similarity technology, a crawling technology and an image sample set, reducing the labor labeling time and the labor cost, improving the efficiency, saving the investment cost, and improving the accuracy of pre-labeling by crawling the text similarity technology, the crawling technology and the crawling technology, crawling the class from a target classification recognition library, searching the best matched model to be migrated from the target classification recognition library through a simulated target recognition technology, obtaining a fine tuning region through a target fine tuning technology, adaptively learning the characteristics of the class characteristics of the target and recognizing the target labeling region, and continuously iteratively updating the model to be migrated through the loss value.
According to the certificate pre-marking method, device, computer equipment and storage medium, the certificate pre-marking model is obtained by training through the training mode of the pre-marking model under the condition that a small number of certificate photos are obtained, so that the marking time of manual certificate photos can be reduced, automatic pre-marking of certificates is realized, the cost is saved, the accuracy of pre-marking is improved, and the identification reliability is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of an application environment of a pre-labeling model training method or a certificate pre-labeling method according to an embodiment of the present invention;
FIG. 2 is a flow chart of a pre-labeling model training method in an embodiment of the invention;
FIG. 3 is a flow chart of a method for pre-labeling credentials in accordance with one embodiment of the present invention;
FIG. 4 is a functional block diagram of a pre-labeling model training apparatus in an embodiment of the invention;
FIG. 5 is a functional block diagram of a document pre-labeling apparatus in accordance with one embodiment of the present invention;
FIG. 6 is a schematic diagram of a computer device in accordance with an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The pre-labeling model training method provided by the invention can be applied to an application environment as shown in fig. 1, wherein a client (computer equipment) communicates with a server through a network. Among them, clients (computer devices) include, but are not limited to, personal computers, notebook computers, smartphones, tablet computers, cameras, and portable wearable devices. The server may be implemented as a stand-alone server or as a server cluster composed of a plurality of servers.
In one embodiment, as shown in fig. 2, a pre-labeling model training method is provided, and the technical scheme mainly includes the following steps S10-S70:
S10, acquiring a target annotation category, a target description corresponding to the target annotation category, a model performance parameter and an image sample set; the set of image samples includes at least one image sample; and the image sample corresponds to the target annotation category.
It may be appreciated that after a certain number of images of a class are collected, pre-labeling is required for these images, a model capable of identifying the class is built for subsequent training, the collected images of the class are huge, in order to be able to automatically label the required region of the class in each image, to trigger a labeling request, the target labeling class, the target description, the model performance parameter and the image sample set are obtained from the labeling request, the target labeling class is a collected class name belonging to the same class, the target description is a description of the relevant feature of the target labeling class, the model performance parameter is a relevant performance parameter of an automatic pre-labeled model preset according to an application scenario, the model performance parameter includes preset time consumption and preset capacity, and the application scenario may be set according to requirements, for example: the application scene can be a face recognition scene, the performance requirement under the application scene is that the recognition is fast, and the capacity of the model is not considered; the application scene can be a scene identified by an identity card, and the performance requirement under the application scene is that the application scene can be transplanted into mobile equipment to run, so that the requirement on the capacity of a required model is small, the identification speed is not considered, and the like, the image sample set is a collection of collected image samples, the image samples are historically collected images or randomly extract a small number of images from the historically collected images, the image samples comprise target objects which need to be marked and belong to the target marking category, the image samples are in one-to-one correspondence with the target marking category, namely, the image samples belong to the target marking category, for example: the collected 1000 images of the identity cards need to be marked, then the images of the 1000 identity cards are randomly extracted from the images of the 1000 identity cards to be determined as an image sample set, the target marking category is determined to be the identity card, the target is described as a rectangular certificate containing a face head portrait, a national badge and 18-bit characters, the preset time consumption in the model performance parameters is less than 1 second, the preset accuracy is higher than 98 percent, and the preset capacity is less than 1.5M, so that the marking request is triggered.
And S20, crawling a history category similar to the target description in a target classification recognition library by using a text similarity technology, and determining the crawled history category as a category to be migrated.
The crawling process is a process of crawling out related descriptions of the target labeling category by using a web crawler technology, summarizing the crawled related descriptions to extract keywords and occurrence times of each keyword, using the related descriptions as network descriptions, performing keyword aggregation on the network descriptions and the target descriptions, namely, giving keywords consistent with the network descriptions to the target descriptions, weighting the keywords consistent with the network descriptions, and performing important word important comparison on the target descriptions, and comparing the history descriptions under each history category with the weighted target descriptions in the target classification recognition library by using the text similarity technology, and determining the history category with the highest similarity value after comparison as the history category similar to the target description, for example: the target labeling category is an identity card, the target description is a rectangular certificate containing a face head portrait, a national badge and 18-bit characters, the face head portrait, the 18-bit characters and the rectangle are weighted, the history description of which the history category is a photo frame is an image of a rectangular photo frame containing a landscape of a person through the application of a text similarity technology in the target classification recognition library, the similarity value of the image and the weighted target description is the highest, and the photo frame is determined to be the history category similar to the identity card.
The method comprises the steps of storing all history categories which are trained by the history and are endowed with history descriptions in a target classification recognition library, wherein the history descriptions are descriptions of relevant characteristics of the history categories, a text similarity technology is a technology for comparing the similarity degree between two texts by using a text similarity algorithm, the text similarity algorithm is a technology for performing word embedding (word embedding) conversion processing on the two texts, performing equal weight conversion according to keywords with weights in the word embedding conversion processing process, performing similarity calculation on the processed two texts to obtain similarity values between the two texts, and the web crawler technology is a technology for automatically capturing programs or scripts of web information according to a certain rule so as to obtain required information.
In an embodiment, in the step S20, that is, the step of using the text similarity technology to crawl a history category similar to the target description in a target classification recognition library, determining the crawled history category as a category to be migrated includes:
s201, crawling the network description matched with the target annotation category by using a web crawler technology.
It is to be understood that the web crawler technology is a technology for automatically capturing a program or script of web information according to a certain rule to obtain required information, crawling descriptions related to the target annotation category from the internet, summarizing and refining all the crawled descriptions, extracting keywords by using a TF-IDF algorithm, and determining the occurrence times of each keyword of the extracted keywords as a network description.
Wherein the TF-IDF algorithm is a weighted technique for information retrieval (information retrieval) and text mining (text mining), and the TF-IDF algorithm is a statistical method for evaluating the importance of a word in a text, the importance of a word or word increasing in proportion to the number of times it appears in the document, but decreasing in inverse proportion to the frequency of its occurrence in the corpus.
S202, keyword weighting is carried out on the target description according to the network description, and focusing description is obtained.
Understandably, words conforming to keywords in the network description are found in the target description, and the found words are weighted according to the occurrence times of the keywords, so that the focusing description is obtained.
S203, comparing the history description under each history category with the focusing description in the target classification recognition library by using a text similarity technology to obtain a similarity value corresponding to each history description.
The text similarity technology is a technology for comparing the similarity degree between two texts by using a text similarity algorithm, the text similarity algorithm is a technology for performing word embedding (word parts) conversion processing on the two texts, performing equal weight conversion according to keywords with weights in the word embedding conversion processing process, performing similarity calculation on the two texts after processing to obtain a similarity value between the two texts, the word embedding (word parts) conversion processing is also called word2vec, namely, converting a word into a vector (vector) to represent, assigning equal weights to the vector converted by the keywords with weights in the conversion process, comparing the historical description under each historical category with the focus description by using the text similarity algorithm in the target classification recognition library, and increasing the equal weights to the vector corresponding to the keywords with weights in the output similarity value, namely, having higher overall weight than the influence on the similarity value of the vector corresponding to the keywords with weights.
And S204, determining the history category corresponding to the history description corresponding to the maximum similarity value as the category to be migrated.
The invention realizes that the network description matched with the target annotation category is crawled by applying the web crawler technology; according to the network description, keyword weighting is carried out on the target description, and focusing description is obtained; comparing the history description under each history category with the focusing description in the target classification recognition library by using a text similarity technology to obtain a similarity value corresponding to each history description; the historical category corresponding to the historical description corresponding to the maximum similarity value is determined to be the category to be migrated, so that the category to be migrated is automatically identified from the target classification identification library by applying the web crawler technology, the keyword weighting technology and the text similarity technology, the cost of manual identification is reduced, the category to be migrated can be quickly and accurately found, and the identification accuracy and reliability are improved.
S30, searching a model to be migrated matched with the category to be migrated and the model performance parameter from the target classification recognition library through a simulation target recognition technology, and recognizing target areas of the image samples through the model to be migrated.
The object classification recognition library may further store object detection depth models including object detection depth models set under each history category, object detection depth models of different network structures are further included under the history category, the object detection depth models of different network structures are models conforming to different performance parameters, because complexity and hierarchical structures of the network structures directly affect the performance parameters, the object detection depth model set includes at least one object detection depth model, the object recognition technology is to input each image sample into each object detection depth model, object recognition is performed through each object detection depth model, the object region identified by each object detection depth model is obtained, and the performance parameters are predicted, according to all the object regions and the predicted performance parameters, a technology of a model to be migrated matching with the performance parameters of the model is determined, by applying the object recognition technology, the model to be migrated matching with each object classification and the performance parameters of the model can be found, and each image sample to be recognized has the characteristics of the object region of the image sample to be migrated.
The target region identifies the region of the image sample conforming to the category to be migrated for the target detection depth model, and the prediction performance parameter identifies the time-consuming mean value of the image sample for the target detection depth model and predicts the capacity of the model after learning.
In an embodiment, in the step S30, the searching, by using a simulated target recognition technology, the model to be migrated matching the class to be migrated and the model performance parameter from the target classification recognition library, and the target area of each image sample identified by the model to be migrated includes:
s301, searching a target detection depth model set associated with the category to be migrated in the target classification recognition library; the set of target detection depth models includes at least one target detection depth model.
It is understood that in the object classification recognition library, one of the classes to be migrated is associated with one of the object detection depth model sets, which includes object detection depth models of various network structures, which are embodied as models conforming to different performance parameters, because the complexity and hierarchy of the network structures directly affect the performance parameters.
S302, inputting each image sample into each target detection depth model, carrying out target recognition on each image sample through each target detection depth model, obtaining each target detection depth model to recognize the target area corresponding to each image sample, and outputting prediction performance parameters corresponding to each target area one by one.
It is understood that each of the image samples is input into each of the target detection depth models, each of the target detection depth models is a model which is trained and used for identifying a target area having characteristics related to the category to be migrated, the network structure of the target detection depth models can be VGG19, resnet51, SVM, YOLO, or the like, the targets are identified as extracting characteristics related to the category to be migrated from the input image, and identifying areas according to the extracted characteristics, thereby obtaining target areas, and the process of obtaining the prediction performance parameters by counting time consuming time for inputting the image samples to output target areas and capacity of the model is performed, wherein one of the image samples for one of the target detection depth models identifies the target area and the prediction performance parameters for the image samples.
S303, determining the target detection depth model matched with the model performance parameters according to all the target areas and the predicted performance parameters, and determining the target detection depth model matched with the model performance parameters as a model to be migrated.
Understandably, according to all the target areas and the predicted performance parameters, the model to be migrated matched with the model performance parameters can be comprehensively considered, that is, all the target areas and the predicted performance parameters are combined with the model performance parameters, the winning coefficients of each target detection depth model are calculated, that is, according to the differences between all the target areas and the predicted performance parameters and the model performance parameters, the winning coefficients of each target detection depth model are determined, the winning coefficients can measure the differences between each target detection depth model and the model performance parameters, the target detection depth model which is most suitable for the model performance parameters can be identified through the winning coefficients, and the target detection depth model is determined as the model to be migrated.
The method realizes that the target detection depth model set associated with the category to be migrated is found in the target classification recognition library; inputting each image sample into each target detection depth model, and carrying out target recognition on each image sample through each target detection depth model to obtain a target region which is recognized by each target detection depth model and corresponds to each image sample and a prediction performance parameter which is recognized by each target detection depth model and corresponds to each target region; according to all the target areas and the predicted performance parameters, determining the target detection depth model matched with the model performance parameters and determining the target detection depth model as a model to be migrated, so that the target areas and the predicted performance parameters which are output by target identification according to a plurality of target detection depth models associated with the category to be migrated are matched, the cost of manual matching is reduced, and the appropriate model to be migrated is matched through two dimensions of the identified target areas and the performance parameters, thereby improving the accuracy and reliability of the identification of the follow-up labels.
In an embodiment, in the step S303, that is, the determining, according to all the target areas and the predicted performance parameters, the target detection depth model matching the model performance parameters, and determining the target detection depth model matching the model performance parameters as a model to be migrated, includes:
s3031, performing intersection analysis on all the target areas corresponding to the same image sample to obtain an intersection area corresponding to the image sample, and determining the winning rate of each target detection depth model corresponding to each image sample according to the intersection area corresponding to each image sample and each target area.
It is to be understood that an intersection analysis is performed on all the target areas corresponding to the same image sample, the intersection analysis is an analysis process in which the target areas identified by the target detection depth models to which the same image sample is input are intersected, the target area having the largest area ratio of the area including the intersection is determined as the intersection area, the intersection area is the target area corresponding to the largest area ratio of the area of the intersection of all the target areas corresponding to the same image sample, the area ratio of the intersection area corresponding to one image sample and the target area output by the target detection depth model to which the image sample is input is determined as the winning ratio of the target detection depth model to the image sample, that is, the winning ratio of the area of the intersection area to the target area is determined as the winning ratio of the image sample to each target detection depth model.
S3032, determining a target index weight corresponding to the target detection depth model according to the winning rate of the same target detection depth model, and predicting a final prediction performance parameter corresponding to the target detection depth model according to the prediction performance parameter output by the same target detection depth model.
As can be appreciated, by applying an inverse distance weighted interpolation method, the larger the difference between the first distances is, the smaller the weight is given, and the distance difference between the second winning rate and the first winning rate corresponding to the same target detection depth model is given to the weight of each winning rate, wherein the weight of each winning rate can be obtained by a weight function, and the weight function is:
λ (i，j) ＝1/(1-1/ln(1-x (i，j) ))
wherein lambda is (i，j) To give weight to the winning bid rate of the ith image sample in the set of target detection depth models in the jth target detection depth model, x (i，j) For the winning bid rate of the ith image sample in the jth target detection depth model in the target detection depth model set in the image sample set, according to the winning bid rate corresponding to the same target detection depth model and the weight corresponding to the winning bid rate, namely, the winning bid corresponding to the same target detection depth model And multiplying the rate and the weight corresponding to the rate, summing, and taking an average value to obtain the target index weight corresponding to the target detection depth model.
S3033, obtaining winning coefficients corresponding to the target detection depth models according to the model performance parameters, the target index weights corresponding to the target detection depth models and the final prediction performance parameters.
Understandably, the model performance parameter is divided by the final predicted performance parameter corresponding to each target detection depth model to obtain the performance ratio corresponding to each target detection depth model, the performance ratio is a measure index of a gap between the final predicted performance parameter of each target detection depth model and the model performance parameter, the performance ratio corresponding to the same target detection depth model and the target index weight are multiplied to obtain the winning coefficient corresponding to each target detection depth model, and the winning coefficient is a measure index of matching the target detection depth model and the model performance parameter in combination with the target index weight.
In an embodiment, in the step S3033, the obtaining, according to the model performance parameter, the target index weight corresponding to each target detection depth model, and the final prediction performance parameter, a winning bid coefficient corresponding to each target detection depth model includes:
S30331, dividing the model performance parameter and the final predicted performance parameter corresponding to each target detection depth model to obtain a performance ratio corresponding to each target detection depth model.
S30332, multiplying the performance ratio corresponding to the same target detection depth model by the target index weight to obtain the winning bid coefficient corresponding to each target detection depth model.
The invention realizes that the performance ratio corresponding to each target detection depth model is obtained by dividing the model performance parameter and the final prediction performance parameter corresponding to each target detection depth model; and multiplying the performance ratio corresponding to the same target detection depth model by the target index weight to obtain the winning bid coefficient corresponding to each target detection depth model, thereby providing a scientific winning bid coefficient calculation method and improving the accuracy of selecting the model to be migrated.
And S3034, determining the target detection depth model corresponding to the maximum winning bid coefficient as the model to be migrated.
Understandably, the largest winning bid coefficient is found in all the winning bid coefficients, the target detection depth model corresponding to the winning bid coefficient is recorded as the model to be migrated, the parameter of the target detection depth model is determined as the migration parameter of the model to be migrated, and the model to be migrated is the target detection depth model of the area which needs to be migrated and learned for identification and the target labeling category under the category to be migrated.
The invention realizes that intersection analysis is carried out on all the target areas corresponding to the same image sample to obtain an intersection area corresponding to the image sample, and the winning rate of each target detection depth model corresponding to each image sample is determined according to the intersection area corresponding to each image sample and each target area; determining a target index weight corresponding to each target detection depth model according to all the winning rates; predicting a final predicted performance parameter corresponding to the target detection depth model according to the predicted performance parameter output to the same target detection depth model; obtaining a winning bid coefficient corresponding to each target detection depth model according to the model performance parameters, the target index weight corresponding to each target detection depth model and the final prediction performance parameters; the target detection depth model corresponding to the largest winning bid coefficient is determined to be the model to be migrated, so that an intersection area is analyzed through intersection, the winning bid rate of each target detection depth model corresponding to each image sample is identified, the target index weight corresponding to each target detection depth model is determined, the winning bid coefficient corresponding to each target detection depth model is calculated, and finally the model to be migrated is matched, therefore, the intersection area is automatically identified through intersection analysis, the target index weight is calculated, the winning bid coefficient is calculated, the model to be migrated is matched by combining the model performance parameter, the target index weight and the model performance parameter, the model to be migrated can be scientifically and accurately matched, and accuracy and reliability are improved for subsequent labeling.
S40, performing target fine adjustment on all the target areas to obtain fine adjustment areas corresponding to the image samples, and inputting the image samples, the fine adjustment areas corresponding to the image samples and the target labeling categories into the model to be migrated; the model to be migrated includes migration parameters.
The process of target fine tuning includes performing edge segmentation on an image in a preset range adjacent to an edge of the target region output by the to-be-migrated model corresponding to the image sample, identifying an edge line, performing an edge reduction adjustment process on the target region along the edge line, obtaining a fine tuning region corresponding to each image sample by the target fine tuning method, taking the image sample as a training sample input by the to-be-migrated model, taking the fine tuning region corresponding to the image sample and the target labeling category as labels of the image sample, and inputting the image sample, the fine tuning region corresponding to the image sample and the target labeling category to the to-be-migrated model.
In an embodiment, in the step S40, that is, performing the target trimming on all the target areas, a trimming area corresponding to each image sample is obtained, including:
S401, edge segmentation is carried out in a preset range adjacent to the target area corresponding to the image sample, and edge lines are identified.
Understandably, the edge segmentation is a process of identifying edges in the target area, identifying pixels with color difference values of pixels between adjacent pixels being larger than a preset color difference value, and performing closed line segmentation on the identified pixels, namely segmenting lines capable of being enclosed into closed areas to segment edge lines, wherein the edge lines are lines capable of being enclosed into a closed area by the pixels with color difference values being larger than the preset color difference value.
And S402, carrying out edge reduction on the target area according to the edge line to obtain the fine adjustment area corresponding to the image sample.
Understandably, the edge of the target area is reduced according to the area surrounded by the edge line, the edge is reduced to a process that the target area is reduced to be overlapped with the edge line and can cover the area surrounded by the edge line, and finally, the reduced area is determined as the fine adjustment area, and one image sample corresponds to one fine adjustment area.
The invention realizes that the edge line is identified by carrying out edge segmentation in the preset range adjacent to the target area corresponding to the image sample; according to the edge line, the target area is subjected to edge reduction, and the fine adjustment area corresponding to the image sample is obtained, so that automatic fine adjustment of the target area is realized, more accurate targets are close, and the accuracy of pre-marking is improved.
In an embodiment, the inputting the image sample, the trimming area corresponding to the image sample, and the target labeling category into the model to be migrated in step S40 includes:
s403, carrying out random data enhancement on the image samples to generate a plurality of enhanced images corresponding to the image samples.
Understandably, the random data enhancement includes enhancement processing procedures of random rotation, random saturation, random contrast and random brightness, and a plurality of enhancement images can be generated through the random data enhancement method, so that samples for training the model to be migrated are increased, and the recognition accuracy of the model to be migrated is increased.
S404, associating the enhanced image with the fine tuning area corresponding to the image sample, and inputting the enhanced image into the model to be migrated.
It is understandable that the enhanced image and the fine tuning area corresponding to the image sample are associated, and in the case that the enhanced image is obtained by random rotation, the fine tuning area associated with the enhanced image is also associated by corresponding rotation, and the enhanced image and the fine tuning area corresponding to the image sample are input to the model to be migrated after being associated.
And S405, taking the target annotation category as a category result in the model to be migrated.
Understandably, the model to be migrated characterizes a model for identifying the target annotation class, and the target annotation class is taken as the class result.
The invention realizes that a plurality of enhanced images corresponding to the image samples are generated by carrying out random data enhancement on the image samples; associating the enhanced image with the fine tuning area corresponding to the image sample, and inputting the enhanced image and the fine tuning area into the model to be migrated; the target labeling category is used as a category result in the model to be migrated, so that the enhancement image is generated through random data enhancement, and is associated with the fine adjustment area and is input into the model to be migrated, the model to be migrated can be prevented from being over-fitted, and the accuracy and quality of labeling are improved.
S50, obtaining a target labeling area corresponding to the image sample by the model to be migrated through extracting the target class characteristics of the model to be migrated in a self-adaptive learning mode by using a migration learning technology.
The self-adaptive learning target class features are related hidden features for automatically adjusting and learning the target annotation class, namely learning hidden features suitable for the target annotation class among the originally learned features for identifying the class to be migrated, and the self-adaptive learning target class features are extracted for identification, so that the hidden features for identifying the target annotation class can be continuously learned, and a target annotation region with the features of the target annotation class can be output, wherein the target annotation region is a region which is learned and output by the model to be migrated and contains the hidden features of the target annotation class.
And S60, determining a loss value according to the target labeling area and the fine tuning area corresponding to the image sample.
The target labeling area and the trimming area are input into a loss function in the model to be migrated, the loss value corresponding to the image sample is calculated, the loss function can be set according to requirements, for example, the loss function is a cross entropy loss function, and the loss function is the logarithm of the difference value between the target labeling area and the trimming area, and indicates the difference between the target labeling area and the trimming area.
And S70, when the loss value does not reach a preset convergence condition, iteratively updating the migration parameters of the model to be migrated, and recording the model to be migrated after convergence as a pre-labeled model after training is completed until the loss value reaches the preset convergence condition.
Understandably, the convergence condition may be a condition that the value of the loss value is small and will not fall down after 3000 times of calculation, that is, when the value of the loss value is small and will not fall down again after 3000 times of calculation, training is stopped, and the model to be migrated after convergence is recorded as a pre-labeled model after training is completed; the convergence condition may also be a condition that the loss value is smaller than a set threshold, that is, when the loss value is smaller than the set threshold, training is stopped, and the model to be migrated after convergence is recorded as a pre-labeled model after training is completed, so when the loss value does not reach the preset convergence condition, the migration parameters of the model to be migrated are continuously adjusted, and the migration learning technology is triggered to be applied, through the extraction of the self-adaptive learning target class characteristics of the model to be migrated, the step that the target labeled area corresponding to the image sample is marked by the model to be migrated is obtained, the accurate result can be continuously drawn together, and the recognition accuracy is higher. Therefore, the identification of the pre-marking can be optimized, and the accuracy and the reliability of the pre-marking are improved.
Thus, the invention realizes the realization of the target annotation category, the target description corresponding to the target annotation category, the model performance parameter and the image sample set; the method comprises the steps of crawling historical categories similar to target descriptions in a target classification recognition library by using a text similarity technology, and determining the crawled historical categories as categories to be migrated; searching a model to be migrated matched with the category to be migrated and the model performance parameter from the target classification recognition library through a simulation target recognition technology, and recognizing target areas of the image samples through the model to be migrated; performing target fine adjustment on all the target areas to obtain fine adjustment areas corresponding to the image samples, and inputting the image samples, the fine adjustment areas corresponding to the image samples and the target annotation categories into the model to be migrated; obtaining a target labeling area marked by the model to be migrated through extracting the class characteristics of the target to be migrated in a self-adaptive learning manner by using a migration learning technology; determining a loss value according to the target labeling area and the fine tuning area; when the loss value does not reach the preset convergence condition, iteratively updating the migration parameters of the model to be migrated until the loss value reaches the preset convergence condition, recording the model to be migrated after convergence as a pre-labeling model after training, thus realizing automatic training of an image sample set with zero labeling through a text similarity technology, a crawling technology and an image sample set, reducing the labor labeling time and the labor cost, improving the efficiency, saving the investment cost, and improving the accuracy of pre-labeling by crawling the text similarity technology, the crawling technology and the crawling technology, crawling the class from a target classification recognition library, searching the best matched model to be migrated from the target classification recognition library through a simulated target recognition technology, obtaining a fine tuning region through a target fine tuning technology, adaptively learning the characteristics of the class characteristics of the target and recognizing the target labeling region, and continuously iteratively updating the model to be migrated through the loss value.
The certificate pre-labeling method provided by the invention can be applied to an application environment as shown in fig. 1, wherein a client (computer equipment) communicates with a server through a network. Among them, clients (computer devices) include, but are not limited to, personal computers, notebook computers, smartphones, tablet computers, cameras, and portable wearable devices. The server may be implemented as a stand-alone server or as a server cluster composed of a plurality of servers.
In one embodiment, as shown in fig. 3, a method for pre-labeling certificates is provided, which mainly includes the following steps S100-S200:
s100, receiving an image labeling instruction, and acquiring a certificate image in the image labeling instruction.
The method for training the pre-labeling model of the certificate comprises the steps of triggering an image labeling instruction when the collected certificate images need to be pre-labeled, wherein the image labeling instruction comprises the certificate images, and the certificate images are images containing certificates, wherein the certificates can be set according to requirements, such as the certificates are identity certificates, because the difficulty of collecting photos of the identity certificates is high, the confidentiality and the privacy are related, the quantity of the collected images in the confidentiality or privacy categories is limited, and the method for training the pre-labeling model of the certificate, which is suitable for labeling the regions of the certificate, can be used for training a small quantity of certificate images by using the method for training the pre-labeling model of the certificate.
S200, inputting the certificate image into a certificate pre-labeling model trained by the pre-labeling model training method, extracting learned certificate characteristics through the certificate pre-labeling model, and obtaining a labeling result output by the certificate pre-labeling model according to the certificate characteristics; the certificate pre-labeling model is a model trained by taking a certificate as a target labeling category, a certificate description as a target description, preset certificate model performance parameters and a certificate photo collected in a history as an image sample set; and marking the region containing the certificate in the certificate image by the marking result.
The method comprises the steps of inputting a certificate image into a certificate pre-labeling model trained by the pre-labeling model training method, wherein the certificate pre-labeling model is a model trained by taking a certificate as a target labeling category, taking a certificate description as the target description, taking a preset certificate model performance parameter and taking a historically collected certificate photo as an image sample set, the certificate model performance parameter is a preset relevant parameter applied to a model for identifying the certificate, extracting and learning the certificate characteristic by the certificate pre-labeling model, taking the certificate characteristic as a self-adaptive learning target category characteristic in the training process of the pre-labeling model, and therefore obtaining the labeling result identified by the trained certificate pre-labeling model, and labeling the region containing the certificate in the certificate image by the labeling result. The method and the device realize that the certificate pre-marking model is obtained by training through the training mode of the pre-marking model under the condition that a small number of certificate photos are obtained, can reduce the marking time of manual certificate photos, and improve the accuracy of pre-marking.
In an embodiment, a pre-labeling model training device is provided, where the pre-labeling model training device corresponds to the pre-labeling model training method in the foregoing embodiment one by one. As shown in fig. 4, the pre-labeling model training device includes an acquisition module 11, a crawling module 12, an identification module 13, a fine adjustment module 14, a migration module 15, a loss module 16 and a training module 17. The functional modules are described in detail as follows:
the acquisition module 11 is used for acquiring a target annotation category, a target description corresponding to the target annotation category, a model performance parameter and an image sample set; the set of image samples includes at least one image sample; the image sample corresponds to the target annotation class;
a crawling module 12, configured to crawl a history category similar to the target description in a target classification recognition library by using a text similarity technology, and determine the crawled history category as a category to be migrated;
the identifying module 13 is configured to find, by using a simulated target identifying technique, a model to be migrated that matches the category to be migrated and the model performance parameter from the target classification identifying library, and target areas of the image samples identified by the model to be migrated;
The fine adjustment module 14 is configured to perform target fine adjustment on all the target areas, obtain fine adjustment areas corresponding to the image samples, and input the image samples, the fine adjustment areas corresponding to the image samples, and the target labeling category into the model to be migrated; the model to be migrated comprises migration parameters;
the migration module 15 is configured to obtain, by using a migration learning technology, a target labeling area corresponding to the image sample from the model to be migrated through extracting a target class feature of the model to be migrated in a self-adaptive learning manner;
a loss module 16, configured to determine a loss value according to the target labeling area and the trimming area corresponding to the image sample;
and the training module 17 is configured to iteratively update the migration parameters of the model to be migrated when the loss value does not reach a preset convergence condition, until the loss value reaches the preset convergence condition, and record the model to be migrated after convergence as a pre-labeled model after training is completed.
For specific limitations of the pre-labeling model training apparatus, reference may be made to the above limitation of the pre-labeling model training method, and no further description is given here. The modules in the pre-labeling model training device can be realized in whole or in part by software, hardware and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a certificate pre-marking device is provided, and the certificate pre-marking device corresponds to the certificate pre-marking method in the embodiment one by one. As shown in fig. 5, the certificate pre-labeling apparatus includes a receiving module 101 and a labeling module 102. The functional modules are described in detail as follows:
the receiving module 101 is configured to receive an image labeling instruction, and obtain a certificate image in the image labeling instruction;
the labeling module 102 is used for inputting the certificate image into the certificate pre-labeling model trained by the pre-labeling model training method, extracting the learned certificate characteristics through the certificate pre-labeling model, and obtaining a labeling result output by the certificate pre-labeling model according to the certificate characteristics; the certificate pre-labeling model is a model trained by taking a certificate as a target labeling category, a certificate description as a target description, preset certificate model performance parameters and a certificate photo collected in a history as an image sample set; and marking the region containing the certificate in the certificate image by the marking result.
For specific limitations on the document pre-labeling device, reference may be made to the above limitations on the document pre-labeling method, and no further description is given here. The various modules in the certificate pre-labeling apparatus described above may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 6. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by the processor, implements a pre-labeling model training method, or a certificate pre-labeling method.
In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the pre-labeling model training method in the above embodiment when executing the computer program, or implements the certificate pre-labeling method in the above embodiment when executing the computer program.
In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, which when executed by a processor implements the pre-labeling model training method in the above embodiment, or which when executed by a processor implements the certificate pre-labeling method in the above embodiment.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions.
The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention, and are intended to be included in the scope of the present invention.

Claims (10)

1. The pre-labeling model training method is characterized by comprising the following steps of:
acquiring a target annotation category, a target description corresponding to the target annotation category, a model performance parameter and an image sample set; the set of image samples includes at least one image sample; the image sample corresponds to the target annotation class;
The method comprises the steps of crawling historical categories similar to target descriptions in a target classification recognition library by using a text similarity technology, and determining the crawled historical categories as categories to be migrated;
searching a model to be migrated matched with the category to be migrated and the model performance parameter from the target classification recognition library through a simulation target recognition technology, and recognizing target areas of the image samples through the model to be migrated;
performing target fine adjustment on all the target areas to obtain fine adjustment areas corresponding to the image samples, and inputting the image samples, the fine adjustment areas corresponding to the image samples and the target annotation categories into the model to be migrated; the model to be migrated comprises migration parameters;
obtaining a target labeling area corresponding to the image sample, which is labeled by the model to be migrated, through extracting the self-adaptive learning target category characteristics of the model to be migrated by using a migration learning technology;
determining a loss value according to the target labeling area and the fine tuning area corresponding to the image sample;
and when the loss value does not reach a preset convergence condition, iteratively updating the migration parameters of the model to be migrated, and recording the model to be migrated after convergence as a pre-labeled model after training is completed when the loss value reaches the preset convergence condition.
2. The method for training a pre-labeling model according to claim 1, wherein the searching the model to be migrated matching the class to be migrated and the model performance parameter from the target classification recognition library by using a simulated target recognition technology, and the target area of each image sample recognized by the model to be migrated, comprises:
searching a target detection depth model set associated with the category to be migrated in the target classification recognition library; the set of target detection depth models includes at least one target detection depth model;
inputting each image sample into each target detection depth model, carrying out target recognition on each image sample through each target detection depth model, obtaining each target detection depth model to recognize the target area corresponding to each image sample, and outputting prediction performance parameters corresponding to each target area one by one;
and determining the target detection depth model matched with the model performance parameters according to all the target areas and the predicted performance parameters, and determining the target detection depth model matched with the model performance parameters as a model to be migrated.
3. The method for training a pre-labeling model according to claim 2, wherein determining the target detection depth model matching the model performance parameter according to all the target areas and the predicted performance parameters, and determining the target detection depth model matching the model performance parameter as a model to be migrated comprises:
performing intersection analysis on all the target areas corresponding to the same image sample to obtain an intersection area corresponding to the image sample, and determining the winning rate of each target detection depth model corresponding to each image sample according to the intersection area corresponding to each image sample and each target area;
determining a target index weight corresponding to the target detection depth model according to the winning rate of the same target detection depth model, and predicting a final prediction performance parameter corresponding to the target detection depth model according to the prediction performance parameter output by the same target detection depth model;
obtaining a winning bid coefficient corresponding to each target detection depth model according to the model performance parameters, the target index weight corresponding to each target detection depth model and the final prediction performance parameters;
And determining the target detection depth model corresponding to the maximum winning bid coefficient as the model to be migrated.
4. The method of pre-labeling model training according to claim 3, wherein the obtaining the winning coefficients corresponding to each of the target detection depth models according to the model performance parameters, the target index weights corresponding to each of the target detection depth models, and the final prediction performance parameters comprises:
dividing the model performance parameter and the final predicted performance parameter corresponding to each target detection depth model to obtain a performance ratio corresponding to each target detection depth model;
and multiplying the performance ratio corresponding to the same target detection depth model by the target index weight to obtain the winning coefficient corresponding to each target detection depth model.
5. The method of claim 1, wherein the inputting the image sample, the trimming area corresponding to the image sample, and the target annotation class into the model to be migrated comprises:
carrying out random data enhancement on the image samples to generate a plurality of enhanced images corresponding to the image samples;
Associating the enhanced image with the fine tuning area corresponding to the image sample, and inputting the enhanced image and the fine tuning area into the model to be migrated;
and taking the target labeling category as a category result in the model to be migrated.
6. A method for pre-labeling a document, comprising:
receiving an image annotation instruction, and acquiring a certificate image in the image annotation instruction;
inputting the certificate image into a certificate pre-labeling model trained by the pre-labeling model training method according to any one of claims 1 to 5, extracting learned certificate characteristics through the certificate pre-labeling model, and obtaining a labeling result output by the certificate pre-labeling model according to the certificate characteristics; the certificate pre-labeling model is a model trained by taking a certificate as a target labeling category, a certificate description as a target description, preset certificate model performance parameters and a certificate photo collected in a history as an image sample set; and marking the region containing the certificate in the certificate image by the marking result.
7. A pre-labeling model training device, comprising:
the acquisition module is used for acquiring a target annotation category, a target description corresponding to the target annotation category, a model performance parameter and an image sample set; the set of image samples includes at least one image sample; the image sample corresponds to the target annotation class;
The crawling module is used for crawling historical categories similar to the target description in the target classification recognition library by using a text similarity technology, and determining the crawled historical categories as categories to be migrated;
the identification module is used for searching a model to be migrated, which is matched with the category to be migrated and the model performance parameter, from the target classification identification library through a simulated target identification technology, and identifying target areas of the image samples through the model to be migrated;
the fine adjustment module is used for carrying out target fine adjustment on all the target areas to obtain fine adjustment areas corresponding to the image samples, and inputting the image samples, the fine adjustment areas corresponding to the image samples and the target annotation categories into the model to be migrated; the model to be migrated comprises migration parameters;
the migration module is used for obtaining a target labeling area, corresponding to the image sample, of the model to be migrated through extracting the self-adaptive learning target category characteristics of the model to be migrated by using a migration learning technology;
the loss module is used for determining a loss value according to the target labeling area and the fine tuning area corresponding to the image sample;
And the training module is used for iteratively updating the migration parameters of the model to be migrated when the loss value does not reach the preset convergence condition, and recording the model to be migrated after convergence as a pre-labeled model after training is completed when the loss value reaches the preset convergence condition.
8. A document pre-labeling apparatus, comprising:
the receiving module is used for receiving the image annotation instruction and acquiring the certificate image in the image annotation instruction;
the marking module is used for inputting the certificate image into the certificate pre-marking model trained by the pre-marking model training method according to any one of claims 1 to 5, extracting the learned certificate characteristics through the certificate pre-marking model, and obtaining a marking result output by the certificate pre-marking model according to the certificate characteristics; the certificate pre-labeling model is a model trained by taking a certificate as a target labeling category, a certificate description as a target description, preset certificate model performance parameters and a certificate photo collected in a history as an image sample set; and marking the region containing the certificate in the certificate image by the marking result.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the pre-labeling model training method of any of claims 1 to 5 when executing the computer program or the certificate pre-labeling method of claim 6 when the processor executes the computer program.
10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the pre-labeling model training method of any of claims 1 to 5, or wherein the processor when executing the computer program implements the certificate pre-labeling method of claim 6.
CN202110211382.2A 2021-02-25 2021-02-25 Pre-labeling model training and certificate pre-labeling method, device, equipment and medium Active CN112926654B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110211382.2A CN112926654B (en) 2021-02-25 2021-02-25 Pre-labeling model training and certificate pre-labeling method, device, equipment and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110211382.2A CN112926654B (en) 2021-02-25 2021-02-25 Pre-labeling model training and certificate pre-labeling method, device, equipment and medium

Publications (2)

Publication Number Publication Date
CN112926654A CN112926654A (en) 2021-06-08
CN112926654B true CN112926654B (en) 2023-08-01

Family

ID=76171813

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110211382.2A Active CN112926654B (en) 2021-02-25 2021-02-25 Pre-labeling model training and certificate pre-labeling method, device, equipment and medium

Country Status (1)

Country Link
CN (1) CN112926654B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113469176B (en) * 2021-06-28 2023-06-02 北京有竹居网络技术有限公司 Target detection model training method, target detection method and related equipment thereof
CN113435463B (en) * 2021-07-27 2023-06-06 平安科技（深圳）有限公司 Object image labeling method, system, equipment and storage medium
CN113705677A (en) * 2021-08-27 2021-11-26 北京三快在线科技有限公司 Image recognition method and device, electronic equipment and storage medium
CN113806613B (en) * 2021-09-29 2023-07-25 中国平安人寿保险股份有限公司 Training image set generation method, training image set generation device, computer equipment and storage medium
CN114049499A (en) * 2021-11-18 2022-02-15 南昌黑鲨科技有限公司 Target object detection method, apparatus and storage medium for continuous contour
CN114972883B (en) * 2022-06-17 2024-05-10 平安科技（深圳）有限公司 Target detection sample generation method based on artificial intelligence and related equipment
CN117671678A (en) * 2022-08-29 2024-03-08 华为技术有限公司 Image labeling method and device
CN116167010B (en) * 2023-04-25 2023-12-08 南方电网数字电网研究院有限公司 Rapid identification method for abnormal events of power system with intelligent transfer learning capability

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019200748A1 (en) * 2018-04-17 2019-10-24 平安科技（深圳）有限公司 Transfer learning method, device, computer device, and storage medium
CN110659646A (en) * 2019-08-21 2020-01-07 北京三快在线科技有限公司 Automatic multitask certificate image processing method, device, equipment and readable storage medium
CN111476268A (en) * 2020-03-04 2020-07-31 中国平安人寿保险股份有限公司 Method, device, equipment and medium for training reproduction recognition model and image recognition
WO2021017261A1 (en) * 2019-08-01 2021-02-04 平安科技（深圳）有限公司 Recognition model training method and apparatus, image recognition method and apparatus, and device and medium

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019200748A1 (en) * 2018-04-17 2019-10-24 平安科技（深圳）有限公司 Transfer learning method, device, computer device, and storage medium
WO2021017261A1 (en) * 2019-08-01 2021-02-04 平安科技（深圳）有限公司 Recognition model training method and apparatus, image recognition method and apparatus, and device and medium
CN110659646A (en) * 2019-08-21 2020-01-07 北京三快在线科技有限公司 Automatic multitask certificate image processing method, device, equipment and readable storage medium
CN111476268A (en) * 2020-03-04 2020-07-31 中国平安人寿保险股份有限公司 Method, device, equipment and medium for training reproduction recognition model and image recognition

Also Published As

Publication number Publication date
CN112926654A (en) 2021-06-08

Similar Documents

Publication Publication Date Title
CN112926654B (en) Pre-labeling model training and certificate pre-labeling method, device, equipment and medium
CN109492643B (en) Certificate identification method and device based on OCR, computer equipment and storage medium
CN109241904B (en) Character recognition model training, character recognition method, device, equipment and medium
CN108288078B (en) Method, device and medium for recognizing characters in image
CN111950329A (en) Target detection and model training method and device, computer equipment and storage medium
CN110444198B (en) Retrieval method, retrieval device, computer equipment and storage medium
US8762383B2 (en) Search engine and method for image searching
WO2022042123A1 (en) Image recognition model generation method and apparatus, computer device and storage medium
CN109886719B (en) Data mining processing method and device based on grid and computer equipment
WO2022134805A1 (en) Document classification prediction method and apparatus, and computer device and storage medium
CN113204660B (en) Multimedia data processing method, tag identification device and electronic equipment
CN112001932A (en) Face recognition method and device, computer equipment and storage medium
CN113806613B (en) Training image set generation method, training image set generation device, computer equipment and storage medium
CN111832581A (en) Lung feature recognition method and device, computer equipment and storage medium
CN111666932B (en) Document auditing method, device, computer equipment and storage medium
CN112528022A (en) Method for extracting characteristic words corresponding to theme categories and identifying text theme categories
CN113408897A (en) Data resource sharing method applied to big data service and big data server
CN115482436B (en) Training method and device for image screening model and image screening method
CN116129881A (en) Voice task processing method and device, electronic equipment and storage medium
CN113283388B (en) Training method, device, equipment and storage medium of living body face detection model
Hettiarachchi Analysis of different face detection andrecognition models for Android
CN115374278A (en) Text processing model distillation method, device, computer equipment and medium
CN112733645B (en) Handwritten signature verification method, handwritten signature verification device, computer equipment and storage medium
CN115840817A (en) Information clustering processing method and device based on contrast learning and computer equipment
CN115359468A (en) Target website identification method, device, equipment and medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant