Automatic iterative training method, system, device and storage medium for recognition model

Abstract

The present disclosure provides an automatic iterative training method, system, device, and computer-readable storage medium for an identification model. Through the technical solution proposed by the present disclosure, it is possible to automatically generate the training data required for the iteration of the identification model based on the identification results and verification results obtained between different training rounds, and use this training data to automatically update and iterate the identification model, so as to avoid the situation that the identification model ages and cannot meet the identification requirements, while saving the labor cost for the operation and maintenance upgrade of the identification model, and has the value of popularization.

Description

Technical Field

[0001] The present disclosure relates to the technical field of model training. Specifically, an automatic iterative training method, system, device, and storage medium for an identification model are disclosed. Background Art

[0002] In the field of model identification technology, a trained identification model is usually used to identify a target object to be identified to obtain an identification result. With the continuous improvement of the requirement for model identification accuracy and the continuous update and iteration of the types of identification objects, it is often necessary to retrain the obtained identification model to meet the growing change of identification requirements. Summary of the Invention

[0003] The present disclosure provides an automatic iterative training method, system, device, and storage medium for an identification model. Specifically, in a first aspect of the present disclosure, an automatic iterative training method for an identification model is provided, including the following steps:

[0004] Perform a training operation on the identification model according to the training data required for the current training round;

[0005] Perform an inference operation on the identification model that has completed the current training round to obtain a first identification result including pre-annotation information;

[0006] Verify the first identification result to obtain a second identification result including verification information;

[0007] Compare the first identification result of the previous training round of the identification model with the second identification result of the current training round to generate the training data required for the next training round;

[0008] Repeat the above steps to achieve automatic iterative training of the identification model.

[0009] In a possible implementation of the above first aspect, a preset model iteration time is set between two adjacent training rounds.

[0010] In a possible implementation of the above first aspect, the identification model is used to identify a to-be-identified image including a target object to obtain identification information of the target object;

[0011] The training data includes the to-be-identified image.

[0012] In a possible implementation of the above first aspect, performing a training operation on the identification model includes the following steps:

[0013] Divide the to-be-identified image to obtain a corresponding foreground region;

[0014] Perform a first preset process on the foreground region to obtain a minimum circumscribed rectangle marking region including the target object;

[0015] Perform a second preset process on the image to be recognized to obtain a preferred training set including a minimum bounding rectangle marked area;

[0016] Train the recognition model with the preferred training set to generate a pre-trained model;

[0017] Perform an optimization process on the pre-trained model to use the pre-trained model that meets the preset evaluation conditions as the recognition model for completing the current training round.

[0018] In a possible implementation of the foregoing first aspect, performing the first preset process on the foreground area includes the following steps:

[0019] Obtain the edge line of the foreground object in the foreground area to obtain the angle at which the foreground object is located in the image to be recognized;

[0020] Rotate the foreground area according to the angle at which it is located so that the foreground object is in the vertical or horizontal direction relative to the image to be recognized;

[0021] Obtain a bounding rectangle marking frame including the target object in the rotated foreground area as the minimum bounding rectangle marked area.

[0022] In a possible implementation of the foregoing first aspect, performing the second preset process on the image to be recognized includes the following steps:

[0023] Perform sliding window cropping on the image to be recognized according to a preset window size to obtain a number of cropped images that have overlapping areas with the minimum bounding rectangle marked area;

[0024] Perform data augmentation processing on the cropped images to generate a preferred training set.

[0025] In a possible implementation of the foregoing first aspect, performing the optimization process on the pre-trained model includes the following steps:

[0026] Perform post-processing on the recognition results of the pre-trained model by using non-maximum suppression.

[0027] The second aspect of the present disclosure provides an automatic iterative training system for a recognition model, which is applied to the automatic iterative training method for a recognition model provided in the foregoing first aspect;

[0028] The automatic iterative training system for a recognition model includes:

[0029] A training unit, configured to perform a training operation on the recognition model according to the training data required for the current training round;

[0030] An inference unit, configured to perform an inference operation on the recognition model that has completed the current training round to obtain a first recognition result including pre-annotation information;

[0031] A verification unit, configured to verify the first recognition result to obtain a second recognition result including verification information;

[0032] A generation unit, configured to compare the first recognition result of the previous training round of the recognition model with the second recognition result of the current training round to generate training data required for the next training round;

[0033] An iteration unit, configured to repeat the above steps to implement automatic iterative training of the recognition model.

[0034] The third aspect of the present disclosure provides an automatic iterative training device for a recognition model, including:

[0035] A memory, configured to store a computer program;

[0036] A processor, configured to implement the automatic iterative training method of the recognition model provided in the foregoing first aspect when executing the computer program.

[0037] The fourth aspect of the present disclosure provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the automatic iterative training method of the recognition model provided in the foregoing first aspect is implemented.

[0038] Compared with the prior art, the present disclosure has the following beneficial effects:

[0039] Through the technical solution proposed by the present disclosure, it is possible to automatically generate the training data required for the iteration of the recognition model based on the recognition results and verification results obtained between different training rounds, and use these training data for automatic update and iteration of the recognition model, so as to avoid the situation that the recognition model ages and cannot meet the recognition requirements, and at the same time save the labor cost for operation and maintenance and upgrade of the recognition model, and has the value of popularization. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objects, and advantages of the present invention will become more apparent:

[0041] Figure 1 According to an embodiment of the present disclosure, a flowchart showing a method for automatic iterative training of a recognition model is shown;

[0042] Figure 2 According to an embodiment of the present disclosure, a flowchart showing a process of performing a training operation on a recognition model is shown;

[0043] Figure 3 According to an embodiment of the present disclosure, a flowchart showing a process of performing a first preset process on a foreground area is shown;

[0044] Figure 4 According to an embodiment of the present disclosure, a schematic diagram of a second preset processing flow for an image to be recognized is shown;

[0045] Figure 5 According to an embodiment of the present disclosure, a schematic structural diagram of an automatic iterative training system for an identification model is shown;

[0046] Figure 6 According to an embodiment of the present disclosure, a schematic structural diagram of an electronic device is shown;

[0047] Figure 7 According to an embodiment of the present disclosure, a schematic structural diagram of a computer-readable storage medium is shown. Specific implementation method

[0048] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that those of ordinary skill in the art can make several changes and improvements without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0049] As used herein, the term "including" and its variants represent open inclusion, that is, "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "at least regionally based on". The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc. may refer to different or the same objects. There may also be other explicit and implicit definitions below.

[0050] Based on the relevant descriptions in the background art, it can be understood that the existing identification model training methods are prone to the problem of identification model aging during the identification process. It is often necessary to repeat the training multiple times to adapt to the continuous improvement of the model identification accuracy requirements and the continuous update and iteration of the types of identification objects. This will consume a large amount of human and time costs for continuous monitoring and timely updating of model aging. To overcome the above technical problems, in some embodiments provided by the present disclosure, Figure 1 A schematic diagram of a process of an automatic iterative training method for an identification model is shown. The automatic iterative training method for the identification model includes the following steps:

[0051] Step 101: Perform a training operation on the identification model according to the training data required for the current training round.

[0052] Step 102: Perform an inference operation on the recognition model that has completed the current training round to obtain a first recognition result containing pre-annotation information.

[0053] Step 103: Verify the first recognition result to obtain a second recognition result containing verification information.

[0054] Step 104: Compare the first recognition result of the previous training round of the recognition model with the second recognition result of the current training round to generate the training data required for the next training round.

[0055] Step 105: Repeat the above steps 101 to 104 to achieve the automatic iterative training of the recognition model. Among them, a preset model iteration time can be set between two adjacent training rounds.

[0056] It can be understood that after each training round is completed, inference can be performed based on the current optimal model, and the pre-annotation result (i.e., the above-mentioned first recognition result) can be generated by the machine, and then the review personnel can perform manual review and fine-tuning based on the annotation result (the result after review and fine-tuning is the above-mentioned second recognition result).

[0057] In the above embodiment, the technical solution provided by the present disclosure can use a database implemented based on Pandas (a numerical calculation extension tool based on the Python computer programming language, which incorporates a large number of libraries and some standard data models and provides the tools required to efficiently operate large datasets), and utilize its related characteristics to quickly and flexibly create the required training data, thereby greatly reducing the maintenance and usage costs; this database can also save the annotation result and the manual review result at the same time and record the differences between the two. If the training performed to optimize the training model each time a new training image is obtained is called a round of training, then the recognition model can be optimized to generate a new training set for corresponding update and iteration according to the difference between the current recognition result and the previous round of annotation result. In the specific implementation process, the model iteration time can be set so that the model can automatically iterate and update itself.

[0058] Furthermore, on the basis of realizing the automatic iterative training of the recognition model, the training process of the model can be further optimized: It can be understood that in the prior art, the quality of the recognition model is closely related to the quality of the manually annotated training data, and there are many problems with manually annotated training data, such as high labor costs, large workload, and inconsistent annotation quality, resulting in obvious differences in the effect of the automatic iterative training of the recognition model. In addition, existing image recognition models are often only applied to general technical fields such as pedestrian detection and face recognition, and cannot meet the needs of ultra-high resolution defect detection and model adaptive iterative upgrade in specific technical fields.

[0059] To overcome the above technical problems, in some embodiments of the present application, Figure 2 shows a schematic flowchart of performing a training operation on an identification model. As Figure 2 shown, it may specifically include the following steps:

[0060] Train the identification model with a preferred training set to generate a pre-trained model;

[0061] Optimize the pre-trained model to use the pre-trained model that meets the preset evaluation conditions as the identification model for completing the current training round

[0062] Step 201: Divide the image to be recognized to obtain the foreground region. It can be understood that by dividing the foreground region from the background region, the consumption of computing resources can be reduced, and at the same time, the interference of the complex background region to the foreground can be reduced. In the above step 201, a foreground segmentation model can be used to implement the division operation, and those skilled in the art can select a suitable division method according to actual needs, which is not limited here.

[0063] Step 202: Perform a first preset process on the foreground region to obtain the minimum bounding rectangle marking region containing the target object. The specific implementation of the first preset process will be elaborated and explained in detail later.

[0064] Step 203: Perform a second preset process on the image to be recognized to obtain a preferred training set containing the minimum bounding rectangle marking region. The specific implementation of the second preset process will be elaborated and explained in detail later.

[0065] Step 204: Train a preset identification model with the preferred training set to generate a pre-trained model.

[0066] Step 205: Optimize the pre-trained model to use the pre-trained model that meets the preset evaluation conditions as the identification model for completing the current training round.

[0067] On the one hand, the automatic iterative training method of the identification model provided in the foregoing steps 201 to 205 can overcome the problems of high labor cost, large workload, and inconsistent annotation quality existing in manual annotation. On the other hand, it can be applied to various specific application scenarios: In an application scenario of the present disclosure, the automatic iterative training method of the above identification model can be applied to the inspection image recognition of wind turbine blades. It can be understood that in the process of inspecting wind turbine blades using image recognition technology, a drone can be used to fly around the wind turbine blade and take images of the surface of the wind turbine blade during the flight. Subsequently, image recognition can be performed on the captured images to obtain whether there are defects on the surface of the wind turbine blade and the specific types corresponding to the defects.

[0068] In the process of inspecting wind turbine blades, the to-be-identified images in the above embodiments may include the inspection captured images containing the wind turbine blades. The foreground region may include the region occupied by the wind turbine blades in the inspection captured images, and the target objects include the defects existing on the surface of the wind turbine blades. Hereinafter, the specific implementation of the automatic iterative training method of the above identification model will be explained and illustrated by taking the application in the identification of wind turbine blade defects as an example.

[0069] In some embodiments of the present disclosure, further, before performing the first preset processing on the foreground region, the following steps may further be included:

[0070] Judge whether the edge line of the foreground object in the foreground region can be recognized and conforms to the preset features: If so, then continue to perform the first preset processing; if not, directly obtain the bounding box containing the target object in the to-be-identified image to generate training data.

[0071] It can be understood that in the process of identifying wind turbine blade defects, if the most primitive blade foreground segmentation model makes a misjudgment on a certain image, it is impossible to obtain the minimum circumscribed rectangle marking region through the subsequent first preset processing. Therefore, it is necessary to pre-judge whether the edge line of the foreground object in the foreground region can be recognized and whether the edge line of the foreground object matches the external features of the foreground object after the foreground division is completed. If any of the conditions does not match, it is impossible to obtain an accurate minimum circumscribed rectangle marking region. At this time, the bounding box containing the target object can be directly obtained in the original to-be-identified image, and then the corresponding training data can be generated.

[0072] In some embodiments of the present disclosure, further, Figure 3 shows a schematic flowchart of performing the first preset processing on the foreground region. As Figure 3 shown, specifically, it may include:

[0073] Step 201: Obtain the edge line of the foreground object in the foreground region to obtain the angle at which the foreground object is located in the to-be-identified image. Taking the identification of wind turbine blade defects as an example, when the foreground region is correctly divided, the position of the blade can be intuitively obtained through the edge line of the wind turbine blade, and then the approximate angle at which the wind turbine blade is located in the to-be-identified image can be calculated by extracting the edge line.

[0074] Step 302: Rotate the foreground region according to the obtained angle so that the foreground object is in the vertical direction or the horizontal direction.

[0075] Step 303: Obtain the circumscribed rectangle bounding box containing the target object in the rotated foreground region as the minimum circumscribed rectangle marking region.

[0076] It can be understood that since most of the defects on the wind turbine blade are distributed along the direction of the wind turbine blade, most of them are in a long and narrow, inclined state; and when the wind turbine blade is in an inclined state in the image to be recognized, when facing long and narrow, inclined defects, if the bounding box is to completely enclose the defective area on the blade, a larger rectangular box must be used to achieve this. In the above step 202, the foreground area can be rotated by using the mask information corresponding to the foreground area, so that the wind turbine blade is in the vertical direction or the horizontal direction. At this time, these long and narrow, inclined defects are rotated to the vertical direction or the horizontal direction. The minimum circumscribed rectangle marking area obtained by marking the circumscribed rectangle at this time can further increase the proportion of the defective foreground compared with directly marking the defect recognition box.

[0077] In some embodiments of the present disclosure, further, Figure 4 shows a schematic diagram of a second preset processing flow for the image to be recognized. As Figure 4 shown, it may specifically include:

[0078] Step 401: Perform sliding window cropping on the image to be recognized according to a preset window size to obtain a number of cropped images that have overlapping regions with the minimum circumscribed rectangle marking area. It can be understood that in the usual artificial intelligence image recognition algorithm, the resolution of the input image can be 1333*800 pixels. In the application of wind turbine blade defect recognition, the resolution of the captured pictures obtained during the inspection process exceeds 20 million pixels. Such a huge captured picture is obviously not suitable for direct input as an image and needs to be preprocessed by means of sliding window cropping.

[0079] During the execution of the sliding window cropping, multiple fixed-size sliding windows can be set in the image to be recognized, and then the overlap degree of each sliding window with the minimum circumscribed rectangle marking area is calculated one by one. If the overlap degree is higher than the preset threshold, the cropped image corresponding to the sliding window is used as qualified training data.

[0080] Step 402: Perform data augmentation processing on the cropped images to generate training data. Among them, the data augmentation processing can use the mosaic algorithm to augment the data, which is equivalent to increasing the size of the batch of images by 4 times, effectively saving computing resources. At the same time, based on the characteristics of the mosaic algorithm itself, operations such as random left-right reversal, random color conversion, random size scaling, random affine transformation, and random rotation are performed during the data augmentation process, effectively improving the generalization ability of the model. In the above embodiment, finally, the four enhanced cropped images can be spliced together to obtain the training data required for training.

[0081] In some embodiments of the present disclosure, further, in the process of training the pre-trained model based on the training data, a method well-known to those skilled in the art is adopted: that is, the training data is input into the pre-built framework model, and the training data here is the spliced image obtained after data augmentation in the above embodiments. In this embodiment, the pre-built framework model can be developed based on the open-source framework mmdectection (a deep learning object detection toolbox implemented by an open-source Python machine learning library). The open-source framework has basic generality, but it is not perfect for all scenarios. Therefore, some local adjustments need to be made to the open-source framework to meet the actual training requirements and the rapid iteration and deployment requirements of subsequent possible requirements. Those skilled in the art can complete the model construction according to the knowledge they have and meet the training requirements of relevant application scenarios, which will not be limited here.

[0082] In some embodiments of the present disclosure, further, the optimization process of the pre-trained model includes the following steps: the recognition results of the pre-trained model are post-processed by using non-maximum merging.

[0083] It can be understood that in the automatic iterative training process of a general recognition model, the post-processing method of non-maximum suppression (NMS) is usually adopted. However, in the above embodiments of the present disclosure, the post-processing method of non-maximum merging is adopted to fit the actual application requirements of wind turbine blade defect recognition. In wind turbine blade defect recognition, it is required that the defect recognition bounding box obtained after image recognition can completely contain the area where the defect is located. Usually, the non-maximum suppression post-processing method may select a better item from multiple eligible candidate boxes as the only retained option, thus abandoning other recognition boxes, and may result in the finally selected recognition box still unable to meet the requirement of completely covering a single recognized defect on the surface of the wind turbine blade; instead, considering the maximum external rectangle corresponding to the union of multiple candidate boxes, the defect part to be recognized can be completely surrounded. Therefore, the post-processing of the recognition results of the pre-trained model by using non-maximum merging is a customized design for the above specific application field.

[0084] In some embodiments of the present disclosure, further, the preset evaluation condition includes that the recall rate of the preferred recognition model is greater than the preset threshold. It can be understood that in the process of evaluating an image recognition model, the evaluation criteria often include two dimensions: recall rate and accuracy. Among them, the accuracy refers to the probability of recognizing the target in the accurately recognized pictures, and the recall rate refers to the ratio of the number of accurately recognized targets to the number of targets in the training set.

[0085] In the application scenario of wind turbine blade defect recognition, the recall rate requirement reaches 99% or even higher. In the above embodiments, the optimization process for achieving a high recall rate includes:

[0086] Ensuring that there is a sufficient number of positive samples for each type of defect that is expected to be recognized during the model training process; and

[0087] By adjusting the judgment threshold for the detection results of positive targets during the recognition judgment process, or the judgment sensitivity.

[0088] Among them, during the process of ensuring a sufficient number of positive samples, more data augmentation work needs to be performed on some rare but high-severity defect samples, so that these defect types with relatively small original sample sizes will not be affected or overwhelmed by those defect types with particularly large sample sizes during the training process; and during the process of adjusting the judgment threshold, the probability distribution of the defect types in the predicted bounding boxes of the model can be viewed, and then the judgment threshold for the true positive prediction results can be appropriately reduced. However, it should be noted that reducing the prediction threshold will directly lead to more false positive predictions in the model, thereby affecting the evaluation index of accuracy. In the application scenario of wind turbine blade defect recognition, the ideal evaluation criteria for recall rate and accuracy are that the recall rate is greater than 99% and the accuracy is greater than 45%.

[0089] In some embodiments of the present disclosure, Figure 5 A kind of automatic iterative training system for an identification model is shown, which is applied to the automatic iterative training method of the identification model provided in the foregoing embodiments. Specifically, as Figure 5 shown, this kind of automatic iterative training system for an identification model may include:

[0090] A training unit 001, configured to perform a training operation on the identification model according to the training data required for the current training round.

[0091] An inference unit 002, configured to perform an inference operation on the identification model that has completed the current training round to obtain a first identification result including pre-annotation information.

[0092] A verification unit 003, configured to verify the first identification result to obtain a second identification result including verification information.

[0093] A generation unit 004, configured to compare the first identification result of the previous training round of the identification model with the second identification result of the current training round to generate the training data required for the next training round.

[0094] An iteration unit 005, configured to sequentially repeat the function implementation of the above functional modules to achieve the automatic iterative training of the identification model.

[0095] It can be understood that the functions implemented by the training unit 001 to the iteration unit 005 in the above functional modules correspond one by one to the operations performed in the foregoing steps 101 to 105, and will not be elaborated herein.

[0096] It can be understood that various aspects of the technical solution of the present disclosure can be implemented as a system, a method, or a program product. Therefore, various aspects of the technical solution of the present disclosure can be specifically implemented in the following forms, namely, a complete hardware implementation method, a complete software implementation method (including firmware, microcode, etc.), or an implementation method combining hardware and software aspects, which can be collectively referred to herein as "circuit", "module", or "platform".

[0097] Figure 6 According to some embodiments of the present disclosure, a schematic structural diagram of an electronic device is shown. This electronic device is used to implement the automatic iteration method in the foregoing embodiments. The following refers to Figure 6 to describe in detail the electronic device 600 implemented according to the implementation method in this embodiment. Figure 6 The displayed electronic device 600 is only an example and should not impose any restrictions on the functions and usage scope of any embodiment of the technical solution of the present disclosure.

[0098] As Figure 6 shown, the electronic device 600 is presented in the form of a general-purpose computing device. The components of the electronic device 600 may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different platform components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0099] Among them, the storage unit stores program code, and the program code can be executed by the processing unit 610, so that the processing unit 610 implements each functional module in the automatic iteration training system of the recognition model in this embodiment.

[0100] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access unit (RAM) 6201 and / or a cache storage unit 6202, and may further include a read-only storage unit (ROM) 6203.

[0101] The storage unit 620 may further include a program / utilities 6204 having a set (at least one) of program modules 6205. Such program modules 6205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. The implementation of a network environment may be included in each or some combination of these examples.

[0102] The bus 630 may represent one or more of several types of bus architectures, including a memory bus or memory controller, a peripheral bus, an image acceleration port, a processing unit, or a local bus using any of the various bus architectures.

[0103] The audio - video signal synchronization processing device 600 may also communicate with one or more external devices 700 (such as a keyboard, a pointing device, a Bluetooth device, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or may communicate with any device that enables the electronic device to communicate with one or more other computing devices (such as a router, a modem, etc.). Such communication may be carried out through the input / output (I / O) interface 650. Moreover, the electronic device 600 may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 through the bus 630. It should be understood that, although Figure 6 not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms, etc.

[0104] In some embodiments of the present disclosure, there is also provided a computer - readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it can implement the realization of each functional module in the automatic iteration system in the above - mentioned disclosure.

[0105] Although other specific implementation manners are not exhaustively listed in this embodiment, in some possible implementation manners, each aspect of the technical solution described in the present disclosure can also be implemented in the form of a program product, which includes program code. When the program product runs on a terminal device, the program code is used to cause the terminal device to execute the steps described in the automatic iteration training method according to the implementation manners in various embodiments of the present disclosure.

[0106] Figure 7 According to some embodiments of the present disclosure, a schematic structural diagram of a computer - readable storage medium is shown. As Figure 7As shown, a program product 800 for implementing the above method in an embodiment according to the technical solution of the present disclosure is described. It may be a portable compact disc read-only memory (CD-ROM) and includes program code, and can run on a terminal device, such as a personal computer. Of course, the program product generated according to this embodiment is not limited thereto. In the technical solution of the present disclosure, a readable storage medium may be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device.

[0107] The program product may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may, for example, be but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0108] The computer-readable storage medium may include a data signal propagated in a baseband or as a carrier wave in a region, which carries the readable program code. Such a propagated data signal may take various forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable storage medium may also be any readable medium other than the readable storage medium, and this readable medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted by any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0109] Program code for performing the operations of the technical solution of the present disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and also including conventional procedural programming languages such as C language or similar programming languages. The program code can be executed entirely on the user's computing device, executed locally on the user's device, executed as a stand-alone software package, executed locally on the user's computing device and remotely on a remote computing device, or executed entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., by connecting through the Internet using an Internet service provider).

[0110] In summary, through the technical solution proposed by the present disclosure, it is possible to automatically generate the training data required for the iteration of the recognition model based on the recognition results and verification results obtained between different training rounds, and use these training data for the automatic update and iteration of the recognition model, so as to avoid the situation that the recognition model ages and cannot meet the recognition requirements, and at the same time save the labor cost for the operation and maintenance upgrade of the recognition model, and has the value of popularization.

[0111] The above description is only a description of the preferred embodiments of the technical solution of the present disclosure, and does not limit the scope of the technical solution of the present disclosure in any way. Any changes and modifications made by those of ordinary skill in the technical field of the present disclosure based on the above disclosure are within the protection scope of the claims.

Claims

1. An automatic iterative training method for an identification model, characterized in that It includes the following steps: Perform a training operation on the recognition model according to the training data required for the current training round; Perform an inference operation on the recognition model that has completed the current training round to obtain a first recognition result containing pre-annotation information; Verify the first recognition result to obtain a second recognition result containing verification information; Compare the first recognition result of the previous training round of the recognition model with the second recognition result of the current training round, and generate the training data required for the next training round according to the difference between the second recognition result and the first recognition result; Repeat the above steps to achieve automatic iterative training of the recognition model; Among them, the recognition model is used to recognize a to-be-recognized image containing a target object to obtain the recognition information of the target object; the training data includes the to-be-recognized image; The performing a training operation on the recognition model includes the following steps: Divide the to-be-recognized image to obtain a corresponding foreground region; Perform a first preset process on the foreground region to obtain a minimum circumscribed rectangle marking region containing the target object; Perform a second preset process on the to-be-recognized image to obtain a preferred training set containing the minimum circumscribed rectangle marking region; Train the recognition model through the preferred training set to generate a pre-trained model; Perform an optimization process on the pre-trained model to use the pre-trained model that meets the preset evaluation conditions as the recognition model that has completed the current training round; The performing a first preset process on the foreground region includes the following steps: Obtain the edge line of the foreground object in the foreground region to obtain the angle at which the foreground object is located in the to-be-recognized image; Rotate the foreground region according to the angle at which it is located so that the foreground object is in the vertical or horizontal direction relative to the to-be-recognized image; Obtain a circumscribed rectangle marking frame containing the target object in the rotated foreground region as the minimum circumscribed rectangle marking region.

2. The automatic iterative training method of the recognition model according to claim 1, characterized in that, There is a preset model iteration time interval between two adjacent training rounds.

3. The automatic iterative training method of the recognition model according to claim 1, characterized in that The performing a second preset process on the to-be-recognized image includes the following steps: Perform sliding window cropping on the to-be-recognized image according to a preset window size to obtain a number of cropped images that have an overlapping region with the minimum circumscribed rectangle marking region; Perform data augmentation processing on the cropped images to generate the preferred training set.

4. The automatic iterative training method of the recognition model according to claim 1, characterized in that, The performing an optimization process on the pre-trained model includes the following steps: Perform post-processing on the recognition result of the pre-trained model by using non-maximum suppression.

5. An automatic iterative training system for an identification model, characterized in that, It is applied to the automatic iterative training method of the recognition model described in any one of claims 1 to 4; The automatic iterative training system of the recognition model includes: A training unit, configured to perform a training operation on the recognition model according to the training data required for the current training round; An inference unit, configured to perform an inference operation on the recognition model that has completed the current training round to obtain a first recognition result containing pre-annotation information; A verification unit, configured to verify the first recognition result to obtain a second recognition result containing verification information; A generating unit, configured to compare the first recognition result of the previous training round of the recognition model with the second recognition result of the current training round, and generate training data required for the next training round according to the difference between the second recognition result and the first recognition result; An iterative unit, configured to repeat the above steps to implement automatic iterative training of the recognition model; wherein, the recognition model is used to recognize a to-be-recognized image containing a target object to obtain recognition information of the target object; the training data includes the to-be-recognized image; The training unit is further configured to: Divide the to-be-recognized image to obtain a corresponding foreground region; Perform a first preset process on the foreground region to obtain a minimum circumscribed rectangle marking region containing the target object; Perform a second preset process on the to-be-recognized image to obtain a preferred training set containing the minimum circumscribed rectangle marking region; Train the recognition model through the preferred training set to generate a pre-trained model; Perform an optimization process on the pre-trained model, and use the pre-trained model that meets the preset evaluation conditions as the recognition model for completing the current training round.

6. An automatic iterative training device for an identification model, characterized in that, Comprising: A memory, configured to store a computer program; A processor, configured to implement the automatic iterative training method of the recognition model according to any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the automatic iterative training method of the recognition model according to any one of claims 1 to 4 is implemented.

Images