CN116740496A - Training method, device, equipment and storage medium of bank image recognition model - Google Patents
Training method, device, equipment and storage medium of bank image recognition model Download PDFInfo
- Publication number
- CN116740496A CN116740496A CN202310661352.0A CN202310661352A CN116740496A CN 116740496 A CN116740496 A CN 116740496A CN 202310661352 A CN202310661352 A CN 202310661352A CN 116740496 A CN116740496 A CN 116740496A
- Authority
- CN
- China
- Prior art keywords
- image data
- bank image
- training
- bank
- model
- 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.)
- Pending
Links
- 238000012549 training Methods 0.000 title claims abstract description 169
- 238000000034 method Methods 0.000 title claims abstract description 56
- 230000005540 biological transmission Effects 0.000 claims abstract description 8
- 238000002372 labelling Methods 0.000 claims description 90
- 230000006870 function Effects 0.000 claims description 34
- 238000004590 computer program Methods 0.000 claims description 10
- 238000012937 correction Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 9
- 239000000463 material Substances 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 238000012360 testing method Methods 0.000 description 6
- 238000012795 verification Methods 0.000 description 3
- 230000001360 synchronised effect Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the application discloses a training method, a training device, training equipment and a training storage medium for a bank image recognition model, wherein the training method comprises the following steps: receiving first bank image data marked manually and second bank image data marked in multiple modes; wherein the multiple models are a single set of at least 2 models that complete training; the first bank image data and the second bank image data are proportionally generated into a training sample set, and the training sample set is input into a bank image recognition model; the bank image recognition model separates the training sample set from the first bank image data and the second bank image data for forward transmission for training; and training until the bank image recognition model converges, and obtaining the bank image recognition model. The method provided by the application improves the data image recognition efficiency and has high model recognition accuracy.
Description
Technical Field
The application belongs to the technical field of image recognition, and particularly relates to a training method, device and equipment of a bank image recognition model and a storage medium.
Background
The image is used as a certificate for archiving records, and is widely applied to the operation of various industries, such as banking business. However, the amount of image data is large, the variety is many, the workload is large when the image data is arranged, checked and archived, and more work difficulties exist when the image data is arranged, checked and archived by manpower.
The algorithm is used for identifying the images, so that arrangement, auditing and archiving are realized, and one of effective practical modes is realized. However, the image identification is performed by using an algorithm, so that high accuracy is realized, and manual labeling is still required. However, in a big data scene, the manual labeling cost is high and the efficiency is low. Therefore, the semi-supervision becomes a choice with higher cost performance, the existing semi-supervision is a UDA algorithm, the method belongs to online training, and the method requires larger resources of the GPU and has high resource consumption while improving the recognition effect of the algorithm.
Disclosure of Invention
The embodiment of the application provides a training method, device, equipment and storage medium for a bank image recognition model, which solve the problems of high manual labeling cost, low efficiency and high resource consumption in the conventional data image recognition.
The embodiment of the application provides a training method of a bank image recognition model, which comprises the following steps:
receiving first bank image data marked manually and second bank image data marked in multiple modes; wherein the multiple models are a single set of at least 2 models that complete training;
the first bank image data and the second bank image data are proportionally generated into a training sample set, and the training sample set is input into a bank image recognition model;
the bank image recognition model separates the training sample set from the first bank image data and the second bank image data for forward transmission for training;
and training until the bank image recognition model converges, and obtaining the bank image recognition model.
Further, the second bank image data of the multimode annotation specifically includes:
inputting unlabeled bank image data one by one into at least 2 single models which are trained;
the single model after training marks the input unlabeled bank image data, and the single model after training outputs marking results respectively;
and acquiring second bank image data of the multi-model label through all the labeling results.
Further, the obtaining the second bank image data of the multimode annotation according to all the annotation results specifically includes:
and carrying out normalization processing on all the labeling results to obtain second bank image data of the multi-model labeling.
Further, the method further comprises:
the method comprises the steps of obtaining a single model for completing training, wherein the single model comprises the following specific steps of:
inputting unlabeled bank image data one by one into at least 2 initial models of single models, and labeling the unlabeled bank image data by the initial models of the single models to obtain a single model labeling data result;
and training all the single models according to the single model labeling data result and the first bank image data labeled by the person to obtain a single model after training.
Further, training all the single models according to the single model labeling data result and the first bank image data labeled by the person to obtain a trained single model, including:
obtaining a labeling difficult sample and a labeling non-difficult sample in an initial model labeling data result of a single model;
correcting the marked difficult sample through manual marking to obtain a correction result;
and respectively inputting the first bank image data marked by the person, the correction result and the marking non-difficult sample into initial models of each single model for training, and obtaining the single models which are trained.
Further, the obtaining the labeling difficult sample and the labeling non-difficult sample in the initial model labeling data result of the single model includes:
obtaining labeling results respectively output by the same bank image data through all single models;
obtaining a prediction success rate through the ratio of the number of the marked correct single models to the total number of all the single models in marked results of all the single models;
comparing the predicted success rate with a preset probability threshold;
if the prediction success rate is greater than or equal to the preset probability threshold, the bank image data is a marked non-difficult sample; if the prediction success rate is smaller than the preset probability threshold, the bank image data is a labeling difficulty sample.
Further, the method further comprises:
and judging different image data according to the naming of the bank image data.
Further, the method further comprises:
the method comprises the steps of obtaining an initial model of a single model, wherein the initial model is specifically as follows:
selecting at least 2 different backbone networks;
and training the different backbone networks through the first bank image data marked by the manpower to obtain an initial model of the single model.
Further, the different backbone networks are backbone networks with different stacking modes, depths, widths, modules and levels.
Further, the training sample set is trained by the bank image recognition model according to the forward propagation of the first bank image data and the second bank image data, including:
forward transmitting the first bank image data to obtain a first loss function; forward transmitting the second bank image data to obtain a second loss function;
multiplying the first loss function with a first duty ratio corresponding to the first loss function to obtain a first multiplication result; multiplying the second loss function with a second duty ratio corresponding to the second loss function to obtain a second multiplying result; the sum of the first and second duty cycles is 1;
and adding the first multiplication result and the second multiplication result to obtain a loss function of the bank image recognition model.
The embodiment of the application also provides a training device of the bank image recognition model, which comprises:
the receiving module is used for receiving the first bank image data marked manually and the second bank image data marked in multiple modes; wherein the multiple models are at least 2 single model sets that have been trained;
the training sample set generation module is used for proportionally generating the first bank image data and the second bank image data into a training sample set and inputting the training sample set into a bank image recognition model;
the training module is used for the bank image recognition model to separate the training sample set into the first bank image data and the second bank image data for forward propagation for training;
and the acquisition module is used for acquiring the bank image recognition model when training until the bank image recognition model converges.
The embodiment of the application also provides training equipment of the bank image recognition model, which comprises a memory and a processor, wherein the memory stores a computer program, and the computer program when executed by the processor causes the processor to execute the following steps:
receiving first bank image data marked manually and second bank image data marked in multiple modes; wherein the multiple models are a single set of at least 2 models that complete training;
the first bank image data and the second bank image data are proportionally generated into a training sample set, and the training sample set is input into a bank image recognition model;
the bank image recognition model separates the training sample set from the first bank image data and the second bank image data for forward transmission for training;
and training until the bank image recognition model converges, and obtaining the bank image recognition model.
The embodiment of the application also provides a computer readable storage medium storing a computer program, which when being executed by a processor, causes the processor to execute the following steps:
receiving first bank image data marked manually and second bank image data marked in multiple modes; wherein the multiple models are a single set of at least 2 models that complete training;
the first bank image data and the second bank image data are proportionally generated into a training sample set, and the training sample set is input into a bank image recognition model;
the bank image recognition model separates the training sample set from the first bank image data and the second bank image data for forward transmission for training;
and training until the bank image recognition model converges, and obtaining the bank image recognition model.
According to the training method for the bank image recognition model, the multiple models are formed through the difference among the trained single models, automatic labeling of image data is achieved, then the bank image recognition model is trained by combining manual labeling, the manual labeling cost and the resource consumption in data image recognition are reduced, the data image recognition efficiency is improved, and the model recognition accuracy is high.
Drawings
In order to more clearly illustrate the embodiments of the application or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Wherein:
FIG. 1 is a diagram of an environment for training a bank image recognition model according to an embodiment;
fig. 2 is a flow chart of a training method of a bank image recognition model according to an embodiment of the present application;
FIG. 3 is a flowchart of a second image data of a second bank with multiple model labels according to an embodiment of the present application;
FIG. 4 is a schematic diagram of a process of training a bank image recognition model by separating forward propagation according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a training device for a bank image recognition model according to an embodiment of the present application;
fig. 6 is a schematic structural diagram of a training device for a bank image recognition model according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of a computer readable storage medium according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Fig. 1 is an application environment diagram of a training method of a bank image recognition model according to an embodiment. Referring to fig. 1, the training method of the bank image recognition model is applied to a bank image recognition system. The bank image recognition system includes a server 200 and a terminal device cluster, where the terminal device cluster may include one or more terminal devices, and in this embodiment, the number of terminal devices is not limited. As shown in fig. 1, the plurality of terminal devices may specifically include a terminal device 1, a terminal device 2, a terminal device 3, …, a terminal device n; the terminal device 1, the terminal device 2, the terminal devices 3, … and the terminal device n are all connected with the server 200 through the network 300, so that each terminal device can perform data interaction with the server 200 through the network 300. Terminal device 1, terminal device 2, terminal devices 3, …, terminal device n may be: intelligent terminals such as smart phones, tablet computers, notebook computers, desktop computers, intelligent televisions and the like. The server 200 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, basic cloud computing services such as big data and artificial intelligence platforms, and the like.
The terminal device 1, the terminal device 2, the terminal devices 3, …, and the terminal device n may be used to collect image data generated in banking business and send the collected image data to the server 200. The server 200 is configured to receive first bank image data that has been manually marked and second bank image data that has been marked in multiple models; wherein the multiple models are a single set of at least 2 models that complete training; the first bank image data and the second bank image data are proportionally generated into a training sample set, and the training sample set is input into a bank image recognition model; the bank image recognition model separates the training sample set from the first bank image data and the second bank image data for forward transmission for training; and training until the bank image recognition model converges, and obtaining the bank image recognition model. The first manually noted bank image data may be noted in the terminal device 1, the terminal device 2, the terminal devices 3, …, and the terminal device n, or may be noted in the server 200.
By adopting the method provided by the embodiment of the application, the multiple models are formed through the difference among the single models which are trained, the automatic labeling of the image data is realized, then the manual labeling is combined, the bank image recognition model is trained, the manual labeling cost and the resource consumption in the data image recognition are reduced, the data image recognition efficiency is improved, and the model recognition accuracy is high.
As shown in fig. 2, a flowchart of a training method of a bank image recognition model according to an embodiment of the present application is provided, and the method can be applied to a terminal device or a server, and this embodiment is exemplified by a server 200. The method comprises the following steps:
step S101, receiving first bank image data marked manually and second bank image data marked in multiple modes; wherein the multiple models are a single set of at least 2 models that complete training;
specifically, in this embodiment, first, multiple kinds of bank image data are collected through a terminal device, and a part of the collected multiple kinds of bank image data are manually marked, so as to obtain first manually marked bank image data; and a part of the images are input into a single model set which is trained to form a plurality of models for marking, and second bank image data with the plurality of models marked are obtained.
It should be noted that, in the present application, the single model is different backbone networks, and at least two different backbone networks may be selected according to actual requirements, where the backbone networks may be ResNet, SENet, inceptionNet, efficientNet, convNeXt, vision Transformer, etc.
In this embodiment, when labeling various collected bank image data, the collected bank image data is labeled according to actual business of the bank image data, that is, the business types of the bank image data are labeled according to liability business, loan business, securities investment business, financial business, transaction clearing business, payment settlement business, bank card business, agency business, escrow business, electronic banking business, and the like. And marking the bank image data in each business type according to materials required in the handling process, for example, marking the collected bank image data in loan business according to materials such as 'agreement', 'and' commitment, identity card or effective identity card submitted by borrowing applicant, household book, temporary or effective residence certificate, retirement certificate, effective marital relation certificate, pay-by-sale invoice (receipt), academic position certificate, title certificate, own housing certificate, lease housing certificate, right-of-service certificate, and 'borrowing electric request list'. And the bank image data after labeling is arranged, checked and archived according to different service types and material types in the service types, so that the bank image data management is facilitated.
Step S102, a training sample set is generated by the first bank image data and the second bank image data according to a proportion and is input into a bank image recognition model;
specifically, the bank image recognition model may be an existing model selected according to the corresponding requirement, or may be constructed according to the actual requirement, and in this embodiment, a neural network composed of a convolution layer, a bn layer and an activation layer is used as the bank image recognition model.
And generating a training sample set by marking the manual and multi-model according to the service type and the material type in the service type, wherein the training sample set comprises a training set, a testing set and a verification set. Then inputting the training set into a bank image recognition model for training, wherein in the training process, the batch_size is set to be 32, each batch consists of first bank image data and second bank image data with multimode labels, and the proportion is 1:3. after training, testing and optimizing are carried out through the testing set, and after testing and optimizing, further optimizing is carried out through the verification set.
Step S103, the bank image recognition model separates the training sample set according to the first bank image data and the second bank image data for forward propagation for training;
specifically, the manual labeling result and the multi-model labeling result obtained in the step S101 are obtained according to the proportion in the step S102, and then the manual labeling result and the multi-model labeling result can be transmitted forward separately, so that the bank image recognition model is trained, tested and optimized.
Step S104, training to obtain the bank image recognition model when the bank image recognition model converges.
Specifically, in this embodiment, when training is performed until convergence according to step S103, a bank image recognition model is obtained.
According to the training method for the bank image recognition model, the multiple models are formed through the difference among the trained single models, automatic labeling of image data is achieved, then the bank image recognition model is trained by combining manual labeling, the manual labeling cost and the resource consumption in data image recognition are reduced, the data image recognition efficiency is improved, and the model recognition accuracy is high.
In some embodiments, as shown in fig. 3, a flow chart of the second bank image data with multi-model labeling provided in the embodiment of the present application specifically includes:
step S201, inputting unlabeled bank image data one by one into at least 2 single models which are trained;
step S202, the single model after training marks the input unlabeled bank image data, and the single model after training outputs marking results respectively;
and step S203, obtaining second bank image data of the multi-model label through all the labeling results.
Specifically, in this embodiment, the video data of the bank without any label is input into at least 2 single models for completing training, and labeling of the banking business type and the material type is performed on the video data of the bank through all the single models, and each single model outputs a labeling result. And then acquiring second bank image data of the multi-model annotation from the annotation result of each model.
In some embodiments, the obtaining the second bank image data of the multi-model label according to all the labeling results is specifically:
and carrying out normalization processing on all the labeling results to obtain second bank image data of the multi-model labeling.
In some embodiments, the method further comprises: and judging different image data according to the naming of the bank image data.
Specifically, in this embodiment, normalization processing is performed on labeling results output by all the single models, the results predicted by different single models of the same image data are obtained, and then the final labeling result of the image data is obtained through processing.
In this embodiment, the same bank image data is determined by naming the bank data, and the same bank image data is input to different single models and the prediction result of the bank image data by the different single models is obtained. And then normalizing the labeling results obtained by the same bank image data through different single models, and in the embodiment, selecting softmax to process the labeling results of different models to obtain the final second bank image data with multi-model labeling.
In some embodiments, the method further comprises: the method comprises the steps of obtaining a single model for completing training, wherein the single model comprises the following specific steps of:
inputting unlabeled bank image data one by one into at least 2 initial models of single models, and labeling the unlabeled bank image data by the initial models of the single models to obtain a single model labeling data result;
and training all the single models according to the single model labeling data result and the first bank image data labeled by the person to obtain a single model after training.
Specifically, in this embodiment, firstly, the untagged bank image data is input into the initial model of the single model, the single model tagging data result including the service type and the material type is obtained, then, a training sample is generated by manually tagging the bank image data of the service type and the material type and the single model tagging data result, the single model is trained, tested and optimized, and the model training can be stopped until the error of the tagging result of the initial model of the single model is less than 20%, and the single model with the training completed is obtained.
In some embodiments, the training all the single models according to the single model labeling data result and the first bank image data labeled by the person to obtain a single model with complete training includes:
obtaining a labeling difficult sample and a labeling non-difficult sample in an initial model labeling data result of a single model;
correcting the marked difficult sample through manual marking to obtain a correction result;
and respectively inputting the first bank image data marked by the person, the correction result and the marking non-difficult sample into initial models of each single model for training, and obtaining the single models which are trained.
Preferably, the obtaining the labeling difficult sample and the labeling non-difficult sample in the initial model labeling data result of the single model includes:
obtaining labeling results respectively output by the same bank image data through all single models;
obtaining a prediction success rate through the ratio of the number of the marked correct single models to the total number of all the single models in marked results of all the single models;
comparing the predicted success rate with a preset probability threshold;
if the prediction success rate is greater than or equal to the preset probability threshold, the bank image data is a marked non-difficult sample; if the prediction success rate is smaller than the preset probability threshold, the bank image data is a labeling difficulty sample.
Specifically, in this embodiment, a labeling difficult sample and a labeling non-difficult sample are obtained from a single model labeling data result output from an initial model of a single model, a preset probability threshold is set to 0.5, and a prediction success rate is greater than or equal to 0.5 and is divided into the labeling non-difficult samples; and dividing the prediction success rate smaller than 0.5 into labeling difficult samples. And then manually labeling all or part of the difficult-labeling samples, and in the embodiment, manually correcting 10% of the difficult-labeling samples to obtain a correction result. And then generating a training sample set through the first bank image data marked manually, the correction result and the marked non-difficult sample, and inputting the training sample set into an initial model of the single model for training to obtain the single model which is trained.
In some embodiments, the method further comprises:
the method comprises the steps of obtaining an initial model of a single model, wherein the initial model is specifically as follows:
selecting at least 2 different backbone networks;
preferably, the different backbone networks are backbone networks comprising different stacking modes, depths, widths, modules and levels.
And training the different backbone networks through the first bank image data marked by the manpower to obtain an initial model of the single model.
Specifically, in this embodiment, the manually marked bank image data is divided into a training set, a verification set and a test set according to the ratio of 3:1:1, and the training set and the test set are input into different backbone networks, and training is performed on each backbone network until each backbone network converges, so as to obtain an initial model of a single model. The method comprises the steps of setting Cross Entropy as a training loss function, evaluating indexes as average correct rate, training different backbone networks, setting parameters of batch_size as 32, training round as 90, enhancing images as colors, rotating and coding in a small range. And respectively fine-tuning the learning rate, the image resolution and other parameters to perform initial model training of the single model. Where correct is the ratio of correct recognition to total recognition.
In some embodiments, as shown in fig. 4, a flow chart for training a bank image recognition model by separating forward propagation is provided in the embodiment of the present application, and step S103 includes:
step S1031, propagating the first bank image data forward, and obtaining a first loss function; forward transmitting the second bank image data to obtain a second loss function;
step S1032, the first loss function is multiplied by a first duty ratio corresponding to the first loss function, so as to obtain a first multiplication result; multiplying the second loss function with a second duty ratio corresponding to the second loss function to obtain a second multiplying result; the sum of the first and second duty cycles is 1;
and step S1033, adding the first multiplication result and the second multiplication result to obtain a loss function of the bank image recognition model.
Specifically, in this embodiment, a first Loss function in the forward propagation of the manual label is defined as a CEloss1, a second Loss function in the forward propagation of the second bank image data is defined as a CEloss2, and then the Loss function Loss of the bank image recognition model is:
Loss＝α×CEloss1+β×CEloss2
the first duty ratio corresponding to the first loss function is α, and the second duty ratio corresponding to the second loss function is β, in this embodiment α=0.8, and β=0.2.
It should be noted that, the first duty ratio α and the second duty ratio β may be adjusted according to different banking types, and may be dynamically changed, that is, the first duty ratio α and the second duty ratio β may be dynamically adjusted according to the banking types.
Fig. 5 is a schematic structural diagram of a training device for a bank image recognition model according to an embodiment of the present application, where the training device includes:
the receiving module 501 is configured to receive first bank image data that has been manually marked and second bank image data that has been marked in multiple models; wherein the multiple models are at least 2 single model sets that have been trained;
the training sample set generating module 502 is configured to generate a training sample set from the first bank image data and the second bank image data according to a proportion, and input the training sample set into a bank image recognition model;
the training module 503 is configured to perform training by using the bank image recognition model to separate the training sample set according to the first bank image data and the second bank image data and forward propagate the first bank image data and the second bank image data;
and the obtaining module 504 is configured to obtain the bank image recognition model after training until the bank image recognition model converges.
In some embodiments, the receiving module 501 includes:
the input model is used for inputting unlabeled bank image data one by one into at least 2 single models which are trained;
the marking module is used for marking the input untagged bank image data by the single model after training, and the single model after training respectively outputs marking results;
and the second bank image data acquisition module is used for acquiring the second bank image data of the multi-model label through all the labeling results.
In some embodiments, the second bank image data obtaining module is further configured to normalize all the labeling results to obtain the second bank image data with the multimode labels.
In some embodiments, the apparatus further comprises:
the single model acquisition module is used for acquiring a single model for completing training, and specifically comprises the following steps:
inputting unlabeled bank image data one by one into at least 2 initial models of single models, and labeling the unlabeled bank image data by the initial models of the single models to obtain a single model labeling data result;
and training all the single models according to the single model labeling data result and the first bank image data labeled by the person to obtain a single model after training.
In some embodiments, the single model acquisition module is further configured to acquire a labeling difficult sample and a labeling non-difficult sample in an initial model labeling data result of the single model;
correcting the marked difficult sample through manual marking to obtain a correction result;
and respectively inputting the first bank image data marked by the person, the correction result and the marking non-difficult sample into initial models of each single model for training, and obtaining the single models which are trained.
In some embodiments, the single model obtaining module is further configured to obtain labeling results that are respectively output by all single models from the same image data of the bank;
obtaining a prediction success rate through the ratio of the number of the marked correct single models to the total number of all the single models in marked results of all the single models;
comparing the predicted success rate with a preset probability threshold;
if the prediction success rate is greater than or equal to the preset probability threshold, the bank image data is a marked non-difficult sample; if the prediction success rate is smaller than the preset probability threshold, the bank image data is a labeling difficulty sample.
In some embodiments, the apparatus further comprises:
and the judging module is used for judging different image data according to the naming of the bank image data.
In some embodiments, the apparatus further comprises:
the initial model obtaining module of the single model is used for obtaining the initial model of the single model, and specifically comprises the following steps:
selecting at least 2 different backbone networks;
and training the different backbone networks through the first bank image data marked by the manpower to obtain an initial model of the single model.
The different backbone networks are backbone networks with different stacking modes, depths, widths, modules and levels.
In some embodiments, training module 503 includes:
the first acquisition module is used for forward transmitting the first bank image data to acquire a first loss function; forward transmitting the second bank image data to obtain a second loss function;
the multiplication module is used for multiplying the first loss function with the corresponding first duty ratio to obtain a first multiplication result; multiplying the second loss function with a second duty ratio corresponding to the second loss function to obtain a second multiplying result; the sum of the first and second duty cycles is 1;
and the second acquisition module is used for adding the first multiplication result and the second multiplication result to acquire a loss function of the bank image recognition model.
For other details of implementing the above technical solution by each module in the training device of the bank image recognition model, reference may be made to the description in the training method of the bank image recognition model provided above, which is not repeated here.
In some embodiments, as shown in fig. 6, a schematic structural diagram of a training device for a bank image recognition model according to an embodiment of the present application includes a memory 601 and a processor 602, where the memory 601 stores a computer program, and when the computer program is executed by the processor 602, the processor 601 performs the following steps:
receiving first bank image data marked manually and second bank image data marked in multiple modes; wherein the multiple models are a single set of at least 2 models that complete training;
the first bank image data and the second bank image data are proportionally generated into a training sample set, and the training sample set is input into a bank image recognition model;
the bank image recognition model separates the training sample set from the first bank image data and the second bank image data for forward transmission for training;
and training until the bank image recognition model converges, and obtaining the bank image recognition model.
For further details of the implementation of the above technical solution by the processor 601 in the training device of the bank image recognition model, reference may be made to the description of the training method of the bank image recognition model provided above, which is not repeated here.
The processor 601 may also be called a CPU (Central Processing Unit ), and the processor 601 may be an integrated circuit chip with signal processing capability; the processor 601 may also be a general purpose processor, such as a microprocessor or the processor 601 may be any conventional processor, a DSP (Digital Signal Process, digital signal processor), ASIC (Application Specific Integrated Circuit ), FPGA (Field Programmable Gata Array, field programmable gate array) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like.
In some embodiments, as shown in fig. 7, a schematic structural diagram of a computer readable storage medium according to an embodiment of the present application is provided, where a readable computer program 701 is stored on the storage medium; wherein the computer program 701 may be stored in the above-mentioned storage medium in the form of a software product, comprising instructions for causing a computer device (which may be a personal computer, a service machine, or a network device, etc.) or a processor (processor) to perform the following steps:
receiving first bank image data marked manually and second bank image data marked in multiple modes; wherein the multiple models are a single set of at least 2 models that complete training;
the first bank image data and the second bank image data are proportionally generated into a training sample set, and the training sample set is input into a bank image recognition model;
the bank image recognition model separates the training sample set from the first bank image data and the second bank image data for forward transmission for training;
and training until the bank image recognition model converges, and obtaining the bank image recognition model.
And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a magnetic or optical disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), or a terminal device such as a computer, a service machine, a mobile phone, or a tablet.
Those skilled in the art will appreciate that all or part of the processes in the methods of the above embodiments may be implemented by a computer program for instructing relevant hardware, where the program may be stored in a non-volatile computer readable storage medium, and where the program, when executed, may include processes in 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.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The foregoing examples illustrate only a few embodiments of the application and are described in detail herein without thereby limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of protection of the present application is to be determined by the appended claims.
Claims (13)
1. A training method for a bank image recognition model, the method comprising:
receiving first bank image data marked manually and second bank image data marked in multiple modes; wherein the multiple models are a single set of at least 2 models that complete training;
the first bank image data and the second bank image data are proportionally generated into a training sample set, and the training sample set is input into a bank image recognition model;
the bank image recognition model separates the training sample set from the first bank image data and the second bank image data for forward transmission for training;
and training until the bank image recognition model converges, and obtaining the bank image recognition model.
2. The method for training a bank image recognition model according to claim 1, wherein the second bank image data of the multimode annotation specifically comprises:
inputting unlabeled bank image data one by one into at least 2 single models which are trained;
the single model after training marks the input unlabeled bank image data, and the single model after training outputs marking results respectively;
and acquiring second bank image data of the multi-model label through all the labeling results.
3. The method for training a bank image recognition model according to claim 2, wherein the obtaining the second bank image data of the multimode annotation by using all the annotation results specifically comprises:
and carrying out normalization processing on all the labeling results to obtain second bank image data of the multi-model labeling.
4. The method of claim 1, further comprising:
the method comprises the steps of obtaining a single model for completing training, wherein the single model comprises the following specific steps of:
inputting unlabeled bank image data one by one into at least 2 initial models of single models, and labeling the unlabeled bank image data by the initial models of the single models to obtain a single model labeling data result;
and training all the single models according to the single model labeling data result and the first bank image data labeled by the person to obtain a single model after training.
5. The method according to claim 4, wherein the training all the individual models according to the single model labeling data result and the first manually labeled bank image data to obtain a trained individual model comprises:
obtaining a labeling difficult sample and a labeling non-difficult sample in an initial model labeling data result of a single model;
correcting the marked difficult sample through manual marking to obtain a correction result;
and respectively inputting the first bank image data marked by the person, the correction result and the marking non-difficult sample into initial models of each single model for training, and obtaining the single models which are trained.
6. The method for training a bank image recognition model according to claim 5, wherein the obtaining the labeling difficult sample and the labeling non-difficult sample in the initial model labeling data result of the single model comprises:
obtaining labeling results respectively output by the same bank image data through all single models;
obtaining a prediction success rate through the ratio of the number of the marked correct single models to the total number of all the single models in marked results of all the single models;
comparing the predicted success rate with a preset probability threshold;
if the prediction success rate is greater than or equal to the preset probability threshold, the bank image data is a marked non-difficult sample; if the prediction success rate is smaller than the preset probability threshold, the bank image data is a labeling difficulty sample.
7. The method of claim 1, further comprising:
and judging different image data according to the naming of the bank image data.
8. The method of claim 1, further comprising:
the method comprises the steps of obtaining an initial model of a single model, wherein the initial model is specifically as follows:
selecting at least 2 different backbone networks;
and training the different backbone networks through the first bank image data marked by the manpower to obtain an initial model of the single model.
9. The method of claim 8, wherein the different backbone networks are backbone networks with different stacking modes, depths, widths, modules and levels.
10. The method of claim 1, wherein the training the bank image recognition model to train the training sample set as separate forward propagates of the first bank image data and the second bank image data comprises:
forward transmitting the first bank image data to obtain a first loss function; forward transmitting the second bank image data to obtain a second loss function;
multiplying the first loss function with a first duty ratio corresponding to the first loss function to obtain a first multiplication result; multiplying the second loss function with a second duty ratio corresponding to the second loss function to obtain a second multiplying result; the sum of the first and second duty cycles is 1;
and adding the first multiplication result and the second multiplication result to obtain a loss function of the bank image recognition model.
11. A training device for a bank image recognition model, the device comprising:
the receiving module is used for receiving the first bank image data marked manually and the second bank image data marked in multiple modes; wherein the multiple models are at least 2 single model sets that have been trained;
the training sample set generation module is used for proportionally generating the first bank image data and the second bank image data into a training sample set and inputting the training sample set into a bank image recognition model;
the training module is used for the bank image recognition model to separate the training sample set into the first bank image data and the second bank image data for forward propagation for training;
and the acquisition module is used for acquiring the bank image recognition model when training until the bank image recognition model converges.
12. A training device for a bank image recognition model, comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the method of any one of claims 1 to 10.
13. A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method of any one of claims 1 to 10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310661352.0A CN116740496A (en) | 2023-06-05 | 2023-06-05 | Training method, device, equipment and storage medium of bank image recognition model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310661352.0A CN116740496A (en) | 2023-06-05 | 2023-06-05 | Training method, device, equipment and storage medium of bank image recognition model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116740496A true CN116740496A (en) | 2023-09-12 |
Family
ID=87903784
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310661352.0A Pending CN116740496A (en) | 2023-06-05 | 2023-06-05 | Training method, device, equipment and storage medium of bank image recognition model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116740496A (en) |
-
2023
- 2023-06-05 CN CN202310661352.0A patent/CN116740496A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2022252363A1 (en) | Data processing method, computer device and readable storage medium | |
| CN109783730A (en) | Products Show method, apparatus, computer equipment and storage medium | |
| CN109816534B (en) | Financing lease product recommendation method, financing lease product recommendation device, computer equipment and storage medium | |
| CN112990281A (en) | Abnormal bid identification model training method, abnormal bid identification method and abnormal bid identification device | |
| CN113239879B (en) | Federal model training and certificate detection methods, devices, equipment and media | |
| CN108062423A (en) | Information-pushing method and device | |
| CN116304352A (en) | Message pushing method, device, equipment and storage medium | |
| CN113343577B (en) | Parameter optimization method, device, equipment and medium based on machine learning | |
| CN116740496A (en) | Training method, device, equipment and storage medium of bank image recognition model | |
| CN114186831B (en) | Personal credit risk prediction method and system by applying transfer learning | |
| CN114529309A (en) | Information auditing method and device, electronic equipment and computer readable medium | |
| CN109409922A (en) | Data aggregate modeling method, device, computer equipment and storage medium | |
| CN115017362A (en) | Data processing method, electronic device and storage medium | |
| CN113468142A (en) | Data transfer request processing method, device, equipment and storage medium | |
| CN112580916B (en) | Data evaluation method, device, computer equipment and storage medium | |
| CN113592549A (en) | Conversion rate prediction method and device and electronic equipment | |
| CN113627514A (en) | Data processing method and device of knowledge graph, electronic equipment and storage medium | |
| CN113052509A (en) | Model evaluation method, model evaluation apparatus, electronic device, and storage medium | |
| US11861384B2 (en) | Methods and systems for generating recommendations based on explainable decision trees for users of a software application | |
| CN115577093B (en) | AI analysis method and system for financial information | |
| CN115034327B (en) | External data application, user identification method, device and equipment | |
| CN115102852B (en) | Internet of things service opening method and device, electronic equipment and computer medium | |
| CN118152874A (en) | Customer classification method, device, equipment and storage medium | |
| CN118172156A (en) | Liability collection method, liability collection device, computer readable storage medium and electronic equipment | |
| CN117853241A (en) | Risk service provider identification method, apparatus, device and storage medium thereof |
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 |