Method and device for determining image recognition model

Abstract

The present invention discloses a method and device for determining an image recognition model. The method comprises the following steps: obtaining an annotated initial data set, and configuring a model to be trained using a training scheme rule set based on sample information of training samples in the initial data set; training the model to be trained using the initial data set to obtain an original model; obtaining an extended data set, determining a preset mining algorithm based on a mining algorithm rule set, and using the original model and the preset mining algorithm to filter out a specified number of optimized samples from the extended data set; after the optimized samples are annotated, retraining the original model using the annotated optimized samples to obtain an optimized image recognition model. The image recognition model generated by the present invention can filter out valid data from massive amounts of unlabeled data or highly repetitive redundant data, thereby reducing the cost of manual annotation and improving development efficiency.

Description

Technical Field

[0001] The present invention relates to the field of computer science and technology, and in particular to a method and device for determining an image recognition model. Background Art

[0002] With the development and advancement of artificial intelligence (AI), the commercialization of AI has reached a stage of maturity in terms of computing power, algorithms, and technology, greatly facilitating people's work and lives. The implementation of AI applications relies heavily on high-quality, valid data. However, in real-world scenarios, there are massive amounts of unlabeled or highly repetitive redundant data that need to be processed.

[0003] Currently, labeling of this massive amount of unlabeled or highly repetitive data is primarily outsourced to third-party annotation systems, but this approach carries significant labor costs. Even with existing image recognition models, the training and mining processes often require the choice of deep learning algorithm frameworks. This requires a deep understanding of various frameworks to select the appropriate one based on the characteristics of the dataset. Users lacking AI expertise often face difficulties in selecting an algorithm framework. Summary of the Invention

[0004] Therefore, in order to solve the problem of high labor costs in the current screening process of massive unlabeled image data or highly repetitive image data, the present invention provides a method and device for determining an image recognition model.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] An embodiment of the present invention provides a method for determining an image recognition model, comprising: obtaining an annotated initial data set, and configuring a model to be trained through a training scheme rule set based on sample information of training samples in the initial data set; training the model to be trained through the initial data set to obtain an original model; obtaining an extended data set, determining a preset mining algorithm based on a mining algorithm rule set, and using the original model and the preset mining algorithm to screen out a specified number of optimized samples from the extended data set; after the optimized samples are annotated, training the original model again using the annotated optimized samples to obtain an optimized image recognition model.

[0007] The present invention obtains a labeled initial data set, configures a to-be-trained model based on the sample information of the training samples in the initial data set, and then trains the to-be-trained model with the initial data set to obtain an original model. The original model is then combined with the obtained extended data set and a predetermined number of optimized samples are screened from the extended data set using the original model and a preset mining algorithm. The optimized samples are finally labeled, and the original model is retrained using the labeled optimized samples to obtain an optimized image recognition model. In practical applications, the image recognition model obtained by the present invention can screen unlabeled image data and image data with high reproducibility to obtain high-quality, labeled, and low-reproducibility valid data. The valid data can then be used to develop artificial intelligence application models, thereby improving the development efficiency of artificial intelligence applications.

[0008] Optionally, the annotated initial data set is obtained in the following manner: collecting image data and performing deduplication processing on the image data using a preset deduplication rule set; annotating the deduplication-processed image data through the created annotation task to generate the annotated initial data set.

[0009] The present invention performs rule-set-based deduplication processing on the collected image data and labels the deduplication-processed image data, thereby reducing the initial data with high similarity in the image data and lowering the cost of subsequent training of the training model.

[0010] Optionally, the sample information of the training samples in the initial data set includes: the balance of positive samples and negative samples in the training samples, the size of the initial data set, and the proportion of each type of label in the initial data set.

[0011] Optionally, configuring the model to be trained based on the sample information of the training samples in the initial data set includes: configuring the hyperparameters, algorithm framework and number of backbone network layers of the model to be trained based on the sample information of the training samples in the initial data set.

[0012] The present invention can configure the hyperparameters of the model to be trained, the algorithm framework of the model to be trained, and the number of backbone network layers of the model to be trained by combining the balance of positive and negative samples of the training samples, the size of the initial data set, and the proportion of various labels in the initial data set. The present invention can more accurately complete the configuration of the hyperparameters, algorithm framework, and number of backbone network layers of the model to be trained, thereby improving the applicability of the model to be trained, by utilizing the balance of positive and negative samples of the training samples, the size of the initial data set, and the proportion of various labels in the initial data set.

[0013] Optionally, obtaining the extended data set includes: determining whether the original model after completing the current training round meets a preset training standard; when the original model does not meet the preset training standard, obtaining the extended data set.

[0014] Optionally, the original model does not meet the preset training standard including: when the number of iterative training of the original model does not reach a preset number, or the model accuracy of the original model does not reach a preset model accuracy, determining that the original model does not meet the preset training standard.

[0015] The present invention determines whether the original model obtained after training meets the preset conditions. When the original model does not meet the preset training standards, an extended data set is obtained and the extended data set is used to optimize the original model; when the original model meets the preset training standards, the original model can be directly used for image recognition.

[0016] Optionally, the preset mining algorithm is determined based on the size of the extended data set and the specified number; screening out a specified number of optimized samples from the extended data set includes: inputting each extended sample in the extended data set into the original model to obtain scoring information of each extended sample; using the preset mining algorithm, sorting each extended sample according to the scoring information, and screening out a specified number of optimized samples from the sorted extended samples.

[0017] The present invention can select a suitable mining algorithm according to the size of the extended data set and the specified number of optimized samples in the extended data set, so as to optimize the original model that does not meet the preset training standard, and obtain the score of each extended data set in the extended data set by inputting the extended data set into the original model for processing, and then sorting the extended samples in the extended data set in combination with the preset mining algorithm, and finally screen out a specified number of optimized samples from the sorted extended samples, and use the optimized samples to optimize the original model, so as to obtain an image recognition model with good effect and user satisfaction.

[0018] According to the second aspect, the present invention also discloses a method and device for determining an image recognition model, which includes: an acquisition configuration module, which is used to obtain an initial data set after annotation, and configure the model to be trained through a training scheme rule set based on the sample information of the training samples in the initial data set; a model training module, which is used to train the model to be trained through the initial data set to obtain the original model; an acquisition screening module, which is used to obtain an extended data set, determine a preset mining algorithm based on the mining algorithm rule set, and use the original model and the preset mining algorithm to screen out a specified number of optimized samples from the extended data set; an iterative training module, which is used to train the original model again using the annotated optimized samples after the optimized samples are annotated to obtain an optimized image recognition model.

[0019] According to the third aspect, an embodiment of the present invention also discloses an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method for determining the image recognition model as described in the first aspect or any optional embodiment of the first aspect.

[0020] According to the fourth aspect, an embodiment of the present invention further discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for determining the image recognition model as described in the first aspect or any optional embodiment of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 Schematic diagram of a flow chart of a method for determining an image recognition model in an embodiment of the present invention;

[0023] Figure 2 1 is a flow chart of a method for determining an image recognition model according to an embodiment of the present invention;

[0024] Figure 3 1 is another specific module flow diagram of the method for determining an image recognition model in an embodiment of the present invention;

[0025] Figure 41 is another specific module flow diagram of the method for determining an image recognition model in an embodiment of the present invention;

[0026] Figure 5 Schematic diagram of the structure of a device for determining an image recognition model in an embodiment of the present invention;

[0027] Figure 6 Schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0029] In the description of the present invention, it should be noted that the term "and / and / or" used in this application specification and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other. The embodiment of the present invention discloses a method for determining an image recognition model, such as Figure 1 As shown, the method includes the following steps:

[0031] Step 101: obtain a labeled initial data set, and configure a model to be trained according to sample information of training samples in the initial data set through a training scheme rule set.

[0032] Specifically, obtaining the annotated initial data set includes: collecting image data, deduplicating the image data using a preset deduplication rule set, and annotating the deduplicated image data through the created annotation task to generate the annotated initial data set.

[0033] As a specific implementation method, Figure 2As shown, in order to improve the efficiency of model training, the preset deduplication rule set can be used to deduplicate the collected image data. Specifically, the deduplication process can be to obtain a preset label system in the image data. Among them, the labels in the preset label system include but are not limited to: camera ID, timestamp, etc. Then, the preset deduplication rule set is used to deduplicate the image data to eliminate duplicate data with high similarity. Subsequently, by creating a labeling task, the third-party labeling system can be used to label the remaining image data that does not have labels, thereby generating an initial labeled data set.

[0034] The preset deduplication rule set is a set of data rule definitions, including but not limited to: Rule A: compare with the already retained image data to determine if the image data does not belong to image data captured with the same camera ID; Rule B: compare with the already retained image data to determine if the image data does not belong to image data captured with the same timestamp. The rules in the present invention can be Rule A and Rule B, or other image data screening rules selected based on actual circumstances, and the present invention is not limited to this.

[0035] When using the preset deduplication rule set to screen image data, the data is evaluated based on the rule logic of the preset deduplication rule set. When data 1 in the image data does not meet rule A, it is proved that data 1 is abnormal data and data 1 is deleted; when data 2 in the image data does not meet rule B, it is proved that data 2 is abnormal data and data 2 is deleted; when data 3 in the image data meets both A and B, data 3 is retained, and image data that meets the preset deduplication rule set is obtained. Then, the unlabeled image data is sent to a third-party labeling system for labeling through the created labeling task, and finally an initial data set with low repeatability and labels is obtained. Among them, the initial labeled data set contains faces as positive sample sets, and the data set without faces is negative sample sets.

[0036] Specifically, after obtaining an initial data set with low repeatability and annotations, the model to be trained is configured according to the sample information of the training samples in the initial data set through the training scheme rule set.

[0037] More specifically, after obtaining the sample information of the training samples of the initial data set, the hyperparameters, algorithm framework, and number of backbone network layers of the model to be trained can be configured based on the balance of positive and negative samples in the training samples, the size of the training samples in the initial data set, and the proportion of various labels in the initial data set, thereby completing the configuration process of the model to be trained.

[0038] The combination of hyperparameters, algorithm frameworks, and backbone network layers can serve as a training scheme. In practical applications, different training schemes can be mapped to different sample information, and this mapping relationship can serve as a training scheme rule set. Specifically, after identifying the sample information of the training samples in the initial dataset, the training scheme that maps to the sample information can be determined from the training scheme rule set. The model to be trained is then configured based on the hyperparameters, algorithm framework, and backbone network layers in the training scheme.

[0039] As a specific implementation method, the learning rate, regularization parameter, number of neural network layers, number of neurons in each hidden layer, number of learning rounds, encoding method of output neurons, choice of cost function, method of weight initialization, type of neuron activation function, etc. of the model to be trained can be configured according to the balance of positive and negative samples in the training samples, the size of the training samples in the initial data set, and the proportion of each type of label in the initial data set. In addition, the input layer, hidden layer, output layer, etc. of the model to be trained can be configured. Specifically, the configuration of the hidden layer includes but is not limited to the configuration of the convolution layer, pooling layer, spectral pooling layer, fully connected layer, etc. By configuring the hyperparameters, algorithm framework, and number of backbone network layers of the model to be trained, the model to be trained is finally obtained.

[0040] Step 102: Train the model to be trained using the initial data set to obtain an original model.

[0041] Specifically, the model to be trained is trained using an initial dataset. A forward calculation is performed based on the training samples in the initial dataset to obtain an error function, and a reverse gradient descent is performed. For example, the most commonly used neural network training method, the backpropagation algorithm, can be used for training. Other neural network training algorithms can also be used, and the present invention is not limited thereto.

[0042] Step 103 , obtaining an extended data set, determining a preset mining algorithm based on a mining algorithm rule set, and using the original model and the preset mining algorithm to screen out a specified number of optimized samples from the extended data set.

[0043] Given the limited number of training samples in the initial dataset, the trained model may only be applicable to certain scenarios. For example, if the training samples in the initial dataset are all facial images taken indoors, the trained model will have high face recognition accuracy in indoor scenes, but not high enough for outdoor scenes. Therefore, after training the original model, you can use an extended dataset with richer training samples to further optimize the original model and obtain a more accurate model.

[0044] Specifically, in one embodiment, after completing the current training round to obtain the original model, it can be determined whether the original model meets the preset training standards. When the original model does not meet the preset training standards, it indicates that the model accuracy or the number of training times has not yet reached the standard. At this time, an extended data set can be further obtained, and the original model can continue to be optimized and trained based on the extended data set; when the original model meets the preset training standards, it indicates that the model accuracy or the number of training times has reached the standard. At this time, the training can be stopped, and the original model can be output as an image recognition model.

[0045] It should be noted that in actual applications, the original model can be iteratively trained over multiple rounds. Within each round, the training data used to train the original model remains unchanged. Only after a round is a decision made based on the results of the training process to determine whether to continue optimizing the original model with the expanded dataset for the next round. Within a round, the original model is trained multiple times using each training sample in the training data. The multiple training cycles within a round do not involve determining whether the original model meets the pre-set training criteria.

[0046] Determining whether the original model meets the preset training criteria includes, but is not limited to, determining whether the number of iterative training iterations of the original model reaches a preset number, or determining whether the model accuracy of the original model reaches a preset model accuracy. The preset number of iterations here refers to the number of training rounds.

[0047] As a specific implementation method, when the number of iterative training of the original model does not reach the preset number, or the model accuracy of the original model does not reach the preset model accuracy, it is determined that the original model does not meet the preset training standard; when the number of iterative training of the original model reaches the preset number, and the model accuracy of the original model reaches the preset model accuracy, it can be determined that the original model meets the preset training standard.

[0048] When the original model does not meet the preset training criteria, an extended dataset is obtained, and each extended sample in the extended dataset is input into the original model to obtain scoring information for each extended sample in the extended dataset. A preset mining algorithm is then used to sort each extended sample according to the scoring information, and a specified number of optimized samples are screened from the sorted extended samples. In practical applications, the preset mining algorithm can be screened using a mining algorithm rule set. Specifically, the mining algorithm rule set can determine the actual selected preset mining algorithm based on the size of the extended dataset and the specified number of optimized samples to be screened from the extended dataset. In practical applications, the preset mining algorithms may include, for example, the C4.5 algorithm, the maximum expectation algorithm, the Apriori algorithm, etc. Depending on the actual size of the extended dataset and the number of optimized samples to be ultimately screened, the corresponding mining algorithm can be flexibly selected. Specifically, different preset mining algorithms have different mining efficiencies for the extended dataset. In practical applications, a mapping relationship can be established between the preset mining algorithm, the size of the extended dataset, and the specified number of optimized samples to be screened. This mapping relationship can follow the principle of maximizing mining efficiency. That is, while keeping the size of the extended dataset and the specified number of optimized samples unchanged, the preset mining algorithm with the highest mining efficiency is used as the mapped mining algorithm. In this way, by varying the size of the expanded dataset and the specified number of optimized samples, different preset mining algorithms can be mapped. Ultimately, by identifying the actual size of the expanded dataset and the specified number of optimized samples, the corresponding preset mining algorithm can be found from a mapping relationship that follows the principle of maximizing mining efficiency. This mapping relationship that follows the principle of maximizing mining efficiency can then serve as the actual mining algorithm rule set.

[0049] As a specific implementation method, when the original model does not meet the preset training criteria, an extended data set is obtained, and the extended data set is input into the original model. The original model scores the extended samples in the extended data set to obtain the scoring information of the extended samples in the extended data set. The extended samples are then sorted in combination with the data mining algorithm to screen out a specified number of optimized samples that are ranked high in the extended data set.

[0050] In real-world scenarios, there's often an excessive amount of unlabeled data. Manually labeling these images individually takes a long time, and quality assurance is difficult. This method uses the original model to score the extended dataset, recommending a mining algorithm based on the extended dataset's size and quality. It then determines which data are most beneficial for optimizing the original model and retains them. This approach significantly reduces the labeling workload, allowing for subsequent model training and ultimately yielding a user-satisfied model.

[0051] Step 104: After the optimized samples are labeled, the original model is trained again using the labeled optimized samples to obtain an optimized image recognition model.

[0052] After selecting a specified number of highly rated optimized samples from the expanded dataset, these samples are annotated by a third-party annotation system to obtain annotated optimized samples. The annotated optimized samples are then used to retrain the original model to obtain an optimized image recognition model. A determination is then made as to whether the optimized image recognition model meets the preset criteria. If the optimized image recognition model does not meet the preset criteria, the model is iteratively trained again until the optimal model is obtained, which serves as the image recognition model.

[0053] Among them, the preset standards include but are not limited to: during a certain iterative training, the image recognition effect of the model is lower than the image recognition effect of the model obtained by the previous training, which can avoid overfitting of the image recognition model training; the number of training times of the image recognition model reaches a preset number, and the preset number can be customized by the user; the model accuracy of the image recognition model reaches a preset accuracy, and the preset accuracy can be customized by the user.

[0054] For example, as an optional embodiment of the present invention, Figure 3 and Figure 4 As shown, image data is obtained, and the image data is input into a preset rule set. The image data is deduplicated using the preset rule set to obtain deduplicated image data. The deduplicated image data is then input into a third-party annotation system to complete the annotation of the screened image data and obtain an annotated initial data set A. Then, based on the annotated initial data set A, the hyperparameters, algorithm framework, and number of backbone network layers of the model to be trained are configured, and the model to be trained is trained based on the initial data set A to obtain Model 1.

[0055] Determine whether the model 1 meets the preset standard. If the model 1 meets the preset standard, output the model 1 as the image recognition model, and end the training process.

[0056] When the model 1 does not meet the preset standards, such as Figure 4 As shown, dataset B is imported and input into model 1 to obtain the scores of the extended samples in data B. Then, the mining algorithm is combined to screen out the top-ranked optimized samples in dataset B to generate dataset B1.

[0057] Furthermore, the generated dataset B1 is handed over to a third-party annotation system for annotation to obtain the annotated dataset B2, and then the dataset B2 is input into model 1 for iterative training of model 1 to obtain the optimized model 2. After multiple iterative trainings, the optimal model n is obtained, and the model n is output as the image recognition model.

[0058] In practical applications, the image recognition model obtained by the present invention can screen unlabeled image data and image data with high repetitiveness to obtain a high-quality valid data set with labels and low repetitiveness, thereby reducing the cost of manual labeling, and then using valid data to develop artificial intelligence application models, thereby improving the development efficiency of artificial intelligence applications.

[0059] The present invention also provides a device for determining an image recognition model, such as Figure 5 As shown, the device includes:

[0060] The acquisition configuration module 51 is used to obtain the annotated initial data set and configure the model to be trained according to the sample information of the training samples in the initial data set through the training scheme rule set. For details, please refer to step 101;

[0061] The model training module 52 is used to train the model to be trained using the initial data set to obtain an original model. For details, refer to step 101.

[0062] An acquisition and screening module 53 is configured to acquire an extended data set, determine a preset mining algorithm based on a mining algorithm rule set, and screen a specified number of optimized samples from the extended data set using the original model and the preset mining algorithm. For details, refer to step 101.

[0063] The iterative training module 54 is used to train the original model again using the annotated optimized samples after the optimized samples are annotated to obtain an optimized image recognition model. For details, please refer to step 101.

[0064] The embodiment of the present invention further provides an electronic device, such as Figure 6 As shown, the electronic device may include a processor 601 and a memory 602, wherein the processor 601 and the memory 602 may be connected via a bus or other means. Figure 6 The bus connection is taken as an example.

[0065] The processor 601 may be a central processing unit (CPU). The processor 601 may also be other general-purpose processors, digital signal processors (DAPs), application-specific integrated circuits (AAICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0066] Memory 602, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as the program instructions / modules corresponding to the key masking method of the apparatus for determining an image recognition model in the embodiments of the present invention. Processor 601 executes the non-transitory software programs, instructions, and modules stored in memory 602 to perform various processor functions and data processing, thereby implementing the image recognition model determination method in the aforementioned method embodiments.

[0067] The memory 602 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created by the processor 601, etc. In addition, the memory 602 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 602 may optionally include a memory remotely located relative to the processor 601, and these remote memories may be connected to the processor 601 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0068] The one or more modules are stored in the memory 602 and when executed by the processor 601, perform the following steps: Figure 1-4 The method for determining the image recognition model in the illustrated embodiment.

[0069] For details of the above electronic equipment, please refer to Figure 1-4 The corresponding descriptions and effects in the embodiments shown can be understood and will not be repeated here.

[0070] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (HDD) or a solid-state drive (SSD), etc. The storage medium can also include a combination of the above-mentioned types of memory.

[0071] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for determining an image recognition model, characterized in that: The method comprises: Obtaining a labeled initial data set, and configuring a model to be trained using a training scheme rule set based on sample information of training samples in the initial data set; the sample information of the training samples in the initial data set includes the balance of positive and negative samples in the training samples, the size of the initial data set, and the proportion of each type of label in the initial data set; the training scheme rule set includes a mapping relationship between training schemes and sample information. After identifying the sample information of the training samples in the initial data set, determining a training scheme mapped to the sample information from the training scheme rule set, and configuring the model to be trained based on the hyperparameters, algorithm framework, and number of backbone network layers in the training scheme; Training the model to be trained using the initial data set to obtain an original model; Acquire an extended data set, determine a preset mining algorithm based on a mining algorithm rule set, and use the original model and the preset mining algorithm to screen a specified number of optimized samples from the extended data set; After the optimized samples are labeled, the original model is trained again using the labeled optimized samples to obtain an optimized image recognition model.

2. The method according to claim 1, characterized in that The annotated initial dataset is obtained in the following manner: Collecting image data and performing deduplication processing on the image data using a preset deduplication rule set; Through the created annotation task, the deduplicated image data is annotated to generate an annotated initial data set.

3. The method according to claim 1, characterized in that Acquiring the extended data set includes: Determining whether the original model after completing the current training round meets the preset training criteria; When the original model does not meet the preset training standard, the extended data set is obtained.

4. The method according to claim 3, characterized in that The original model does not meet the preset training standard including: When the number of iterative trainings of the original model does not reach a preset number, or the model accuracy of the original model does not reach a preset model accuracy, it is determined that the original model does not meet the preset training standard.

5. The method according to claim 1, characterized in that The preset mining algorithm is determined based on the size of the extended data set and the specified number; Filter out a specified number of optimized samples from the extended data set, including: Inputting each extended sample in the extended data set into the original model to obtain scoring information of each extended sample; The preset mining algorithm is used to sort the extended samples according to the scoring information, and a specified number of optimized samples are screened out from the sorted extended samples.

6. A device for determining an image recognition model, characterized in that: The device comprises: An acquisition configuration module is used to obtain an annotated initial data set and configure the model to be trained through a training scheme rule set based on the sample information of the training samples in the initial data set; the sample information of the training samples in the initial data set includes the balance of positive and negative samples in the training samples, the size of the initial data set, and the proportion of each type of label in the initial data set; the training scheme rule set includes a mapping relationship between training schemes and sample information. After identifying the sample information of the training samples in the initial data set, a training scheme mapped to the sample information is determined from the training scheme rule set, and the model to be trained is configured based on the hyperparameters, algorithm framework, and number of backbone network layers in the training scheme; A model training module is used to train the model to be trained using the initial data set to obtain an original model; An acquisition and screening module is used to obtain an extended data set, determine a preset mining algorithm based on a mining algorithm rule set, and use the original model and the preset mining algorithm to screen a specified number of optimized samples from the extended data set; The iterative training module is used to train the original model again using the annotated optimized samples after the optimized samples are annotated to obtain an optimized image recognition model.

7. An electronic device, characterized in that: include: at least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method for determining the image recognition model as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for determining an image recognition model as described in any one of claims 1 to 5 are implemented.

Images