Sample automatic generation, algorithm self-evolution method and device and electronic equipment

Abstract

The present application relates to a kind of sample automatic generation, algorithm self-evolution method and device and electronic equipment.Multiple sample audit strategies are generated, and the multiple sample audit strategies are used to evaluate the first attribute of sample;Identified first sample image is obtained;First information of the first sample image is used to match target sample audit strategy from the multiple sample audit strategies, and the first attribute of the first sample image is evaluated according to the target sample audit strategy;Wherein, the identified first sample image is obtained according to first period, and the first attribute is used to be positive or negative type for sample.Automatic generation of labeled sample, and algorithm parameter is automatically updated, reduce the dependence on artificial, improve the accuracy of algorithm.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of image detection, and particularly to a method and apparatus for automatic sample generation in image detection algorithms, a method and apparatus for self-evolution of algorithms based on automatic sample generation, and an electronic device. Background Technology

[0002] With the further improvement of the accuracy of image detection and recognition algorithms, these algorithms are widely used in security, access control, shopping malls, and other fields. Before performing object detection and recognition tasks, these algorithms need to be trained with a large amount of sample data to improve their accuracy. Therefore, the quantity, richness, and labeling accuracy of the samples are all important factors affecting the algorithm's performance. In existing technologies, the training samples for algorithms are usually manually labeled beforehand, or they utilize readily available online databases of pre-labeled samples. This means that all built-in detection and recognition algorithms in cameras are trained using manually labeled samples, requiring technicians to periodically retrieve and import samples for training. The acquisition, labeling, and utilization of samples all rely on manual operation by technicians. Therefore, in existing image recognition algorithms, the generation, use, and updating of samples are highly dependent on manual intervention. Samples cannot be created and updated in a timely and automatic manner, leading to low algorithm accuracy.

[0003] Based on the above sample acquisition methods, when using samples for algorithm training, manual import is required before use. During use, recently updated samples also need to be manually imported for algorithm self-evolution, resulting in a low update frequency and a high degree of dependence on manual intervention.

[0004] Therefore, there is an urgent need for a sample automatic generation method and device that can generate and update samples in real time, an algorithm self-evolution method and device based on automatically generated samples, and electronic equipment. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problems in the prior art where samples cannot be automatically generated and updated, and algorithms cannot automatically evolve. It provides a method, apparatus, and electronic device for automatic sample generation and algorithm self-evolution, which realizes automatic generation, labeling, and updating of samples and automatic evolution of algorithms based on the samples, reducing reliance on manual intervention and improving the efficiency and accuracy of sample generation and algorithm self-evolution.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide an automatic sample generation method, the automatic sample generation method comprising:

[0007] Multiple sample review strategies are generated, and the multiple sample review strategies are used to evaluate the first attribute of the sample;

[0008] Obtain the image of the first identified sample;

[0009] Based on the first information of the first sample image, a target sample review strategy is matched from the plurality of sample review strategies, and the first attribute of the first sample image is evaluated according to the target sample review strategy.

[0010] The identified first sample image is acquired according to the first cycle, and the first attribute is used to indicate the positive or negative type of the sample.

[0011] Preferably, the first information of the first sample image includes at least a first recognition algorithm identifier and a first acquisition device identifier;

[0012] Each of the multiple sample review strategies has a strategy attribute, which includes at least a second identification algorithm identifier, a second acquisition device identifier, and a strategy confidence level.

[0013] Among them, the first and second recognition algorithm identifiers are used to characterize the image recognition algorithm, and the first and second acquisition device identifiers are used to characterize the camera and camera angle for acquiring images.

[0014] Preferably, matching a target sample review strategy from the plurality of sample review strategies based on the first information of the first sample image includes:

[0015] Match the first identification algorithm identifier and the first acquisition device identifier among multiple second identification algorithm identifiers and second acquisition device identifiers to determine the matched target review strategy. The second identification algorithm identifier and the second acquisition device identifier of the target review strategy are the same as the first identification algorithm identifier and the first acquisition device identifier of the first sample image.

[0016] Evaluating the first attribute of the first sample image according to the target sample review strategy includes: evaluating the first attribute of the first sample image according to the strategy confidence of the target sample review strategy, wherein the first attribute includes at least correct, incorrect, minor, and inappropriate.

[0017] Preferably, the strategy for generating multiple sample reviews includes:

[0018] The historical reviewed sample set is statistically analyzed. Based on the identification algorithm identifier and the data acquisition device identifier, the historical reviewed sample set is classified to obtain multiple historical reviewed sample classes. The confidence level of each historical reviewed sample class is statistically calculated. Thus, multiple sample review strategies are generated based on the identification algorithm identifier, data acquisition device identifier, and confidence level of each historical reviewed sample class.

[0019] Preferably, after evaluating the first attribute of the first sample image according to the target sample review strategy, the method further includes:

[0020] The first sample image is added to the historical reviewed sample set, and the first sample image contains at least the first attribute and the target sample review strategy information.

[0021] Preferably, before acquiring the identified first sample image, the method further includes:

[0022] Unidentified sample images are acquired, their Regions of Interest (ROIs) are extracted, and object recognition algorithms are used to identify the ROIs to obtain the first identified sample image.

[0023] Preferably, the acquisition of the ROI region of the unidentified sample image includes:

[0024] Based on the object recognition algorithm identifier and the acquisition device identifier of the unrecognized sample image, the target ROI region determination rule is matched from the ROI rule base, and the ROI region of the unrecognized sample image is extracted according to the target ROI region determination rule.

[0025] Preferably, the ROI rule base includes multiple ROI region determination rules, wherein each ROI region determination rule includes an identification attribute, and the identification attribute includes at least the object recognition algorithm identifier, the acquisition device identifier, and the range of the ROI region to be identified;

[0026] The method involves matching target ROI region determination rules from the ROI rule base based on partial information, including: matching the object recognition algorithm identifier and the acquisition device identifier of the unrecognized sample image with multiple ROI region determination rules in the ROI rule base to obtain target ROI region determination rules, and extracting the ROI region of the unrecognized sample image based on the range of the ROI region to be identified in the target ROI region determination rules.

[0027] Preferably, while generating multiple sample review strategies, it also includes statistical analysis of historical reviewed sample sets, classifying the historical reviewed sample sets according to the identification algorithm identifier and the acquisition device identifier to obtain multiple historical reviewed sample classes, and statistically analyzing the regions to be identified for each sample in each historical reviewed sample class, thereby generating multiple ROI region determination rules based on the identification algorithm identifier, acquisition device identifier, and regions to be identified for each historical reviewed sample class.

[0028] Preferably, the step of extracting the ROI region of the unidentified sample image according to the target ROI region determination rule specifically includes: extracting the initial ROI region of the unidentified sample image using a first ROI region extraction algorithm, and correcting the ROI region of the unidentified sample image according to the target ROI region determination rule.

[0029] Preferably, the first sample image is added to the training sample pool, and the algorithm reads the evaluated sample images in the training sample pool to update the algorithm's parameters.

[0030] The present invention also provides an automatic sample generation device, the automatic sample generation device comprising:

[0031] The strategy generation module is used to generate multiple sample review strategies, which are used to evaluate the first attribute of the sample.

[0032] The image acquisition module is used to acquire the first identified sample image;

[0033] An image evaluation module is used to match a target sample review strategy from the plurality of sample review strategies based on the first information of the first sample image, and evaluate the first attribute of the first sample image according to the target sample review strategy.

[0034] The identified first sample image is acquired according to the first cycle, and the first attribute is used to indicate the positive or negative type of the sample.

[0035] This invention also provides an algorithm self-evolution method, the algorithm self-evolution method comprising:

[0036] The algorithm is triggered according to the first preset trigger condition to read the labeled sample image set, and the algorithm is trained and updated according to the read labeled sample image set; wherein, the labeled sample image set includes multiple labeled sample images, and each labeled sample image is an evaluated first sample image obtained according to the sample automatic generation method described above.

[0037] The present invention also provides an algorithm self-evolution device, the algorithm self-evolution device including the sample automatic generation device as described above, and the algorithm self-evolution device further including:

[0038] The algorithm self-evolution module is used to trigger the algorithm to read the labeled sample image set according to the first preset trigger condition, train the algorithm based on the read labeled sample image set, and update the algorithm parameters.

[0039] The labeled sample image set includes multiple labeled sample images, each labeled sample image being an evaluated first sample image obtained by the automatic sample generation device.

[0040] The present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the algorithm self-evolution method described above.

[0041] Compared with the prior art, the automatic sample generation method and apparatus provided by the present invention achieve automatic sample extraction by setting a first cycle, ensuring that automatic sample generation is triggered. At the same time, it automatically labels sample images based on a sample review strategy, eliminating the dependence on manual labor in the sample generation process, and using a unified standard to evaluate the samples, thereby improving the efficiency of sample generation and the accuracy of sample labeling.

[0042] Based on the automatic sample generation method and apparatus, the algorithm self-evolution method and apparatus provided by the present invention utilize automatically generated samples as data sources to automatically trigger the algorithm to read and train samples, thereby realizing the automatic evolution of algorithm parameters. This further reduces the dependence on manual intervention during the algorithm self-evolution process and improves the efficiency and accuracy of the algorithm self-evolution.

[0043] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0044] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0045] Figure 1 A flowchart of the automatic sample generation method provided in this embodiment of the invention;

[0046] Figure 2 This is a schematic diagram of the automatic sample generation device provided in an embodiment of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the various embodiments of the present invention to facilitate a better understanding of this application. However, the technical solutions claimed in this application can be implemented even without these technical details and various changes and modifications based on the following embodiments. The division of the various embodiments below is for ease of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined with and referenced by each other without contradiction.

[0048] The first embodiment of the present invention relates to a method for automatically generating samples, such as... Figure 1As shown, the automatic sample generation method includes: generating multiple sample review strategies, the multiple sample review strategies being used to evaluate a first attribute of a sample; acquiring an identified first sample image; matching a target sample review strategy from the multiple sample review strategies based on first information of the first sample image; and evaluating the first attribute of the first sample image according to the target sample review strategy, wherein the identified first sample image is acquired according to a first cycle, and the first attribute is used to indicate the positive or negative type of the sample.

[0049] The automatic sample generation method provided by this invention automatically acquires the latest first sample image according to a first cycle and automatically evaluates the first sample image, thus realizing automatic sample extraction and automatic annotation. Compared with the prior art of manually pulling samples periodically, manually annotating them, or using ordinary sample databases that have already been annotated online, the automatic sample extraction and annotation method of this invention can get rid of the dependence on manual labor, and uses a unified standard to evaluate the samples. It can automatically build and update samples in a timely and fast manner, ensuring timely and accurate acquisition of samples with the first attribute that are usable by the algorithm. In this way, it provides the algorithm with accurately annotated and periodically automatically updated sample data, thereby improving the accuracy of the algorithm.

[0050] Furthermore, the first information of the first sample image includes at least a first recognition algorithm identifier and a first acquisition device identifier. Each of the multiple sample review strategies has a strategy attribute, which includes at least a second recognition algorithm identifier, a second acquisition device identifier, and a strategy confidence level. The first and second recognition algorithm identifiers characterize the image recognition algorithm, and the first and second acquisition device identifiers characterize the camera and camera angle used to acquire the image. Matching a target sample review strategy from the multiple sample review strategies based on the first information of the first sample image specifically includes: matching the first recognition algorithm identifier and the first acquisition device identifier among multiple second recognition algorithm identifiers and second acquisition device identifiers to determine the matched target review strategy. The second recognition algorithm identifier and the second acquisition device identifier of the target review strategy are the same as the first recognition algorithm identifier and the first acquisition device identifier of the first sample image. Evaluating the first attribute of the first sample image according to the target sample review strategy, specifically, evaluating the first attribute of the first sample image according to the strategy confidence level of the target sample review strategy. The first attribute includes at least correct, incorrect, minor, and inappropriate. The first and second acquisition device identifiers are specifically camera identifiers, but can also be camera and shooting angle identifiers.

[0051] The automatic sample generation method provided by this invention matches information from sample images with information from sample review strategies to select an evaluation method for the sample images, thereby achieving automatic sample image annotation. Different acquisition devices and different recognition algorithms employ different target sample review strategies. Defining the target sample review strategy from multiple dimensions improves the accuracy of sample annotation, thereby enhancing the accuracy of automatically generated samples.

[0052] The generation of multiple sample review strategies includes: statistically analyzing the historical reviewed sample set, classifying the historical reviewed sample set according to the identification algorithm identifier and the acquisition device identifier to obtain multiple historical reviewed sample classes, statistically calculating the confidence level of each historical reviewed sample class, and thereby generating multiple sample review strategies based on the identification algorithm identifier, acquisition device identifier, and confidence level of each historical reviewed sample class; furthermore, after evaluating the first attribute of the first sample image according to the target sample review strategy, the method further includes adding the first sample image to the historical reviewed sample set, wherein the first sample image contains at least the first attribute and the target sample review strategy information. For example, multiple historically reviewed sample classes include four categories: Algorithm A, acquisition device m; Algorithm B, acquisition device m; Algorithm A, acquisition device n; Algorithm B, acquisition device n. The number of samples and the accuracy of sample identification under each category are obtained respectively, and the confidence level of each category is calculated, thus forming four sample review strategies: Algorithm A, acquisition device m, confidence level 0.8; Algorithm B, acquisition device m, confidence level 0.9; Algorithm A, acquisition device n, confidence level 0.5; Algorithm B, acquisition device n, confidence level 0.99. The first attribute includes at least correct, incorrect, minor, and inappropriate. The first attribute of the first sample image is evaluated based on the strategy confidence level of the target sample review strategy, specifically including determining the first attribute of the first sample image based on the interval in which the strategy confidence level of the target sample review strategy falls. For example, if the strategy confidence level of the target sample review strategy is greater than or equal to a first threshold, then the first attribute of the first sample image is correct; if the strategy confidence level of the target sample review strategy is less than a second threshold, then the first attribute of the first sample image is incorrect.

[0053] This invention automatically labels samples based on a sample review strategy when evaluating identified samples, thereby eliminating reliance on manual labeling, standardizing labeling criteria, and improving the scientific rigor of labeling. Furthermore, to enhance the accuracy of labeling, this invention continuously updates the sample review strategy by supplementing and statistically analyzing historically reviewed samples, further improving the accuracy of sample labeling.

[0054] Furthermore, before obtaining the identified first sample image, the method further includes obtaining an unidentified sample image, extracting the ROI region of the unidentified sample image, and using an object recognition algorithm to identify the ROI region in order to obtain the identified first sample image.

[0055] The process of obtaining the ROI region of the unrecognized sample image includes matching the target ROI region determination rule from the ROI rule base according to the object recognition algorithm identifier and the acquisition device identifier of the unrecognized sample image, and extracting the ROI region of the unrecognized sample image according to the target ROI region determination rule.

[0056] The ROI rule base includes multiple ROI region determination rules. Each ROI region determination rule includes an identification attribute, which at least includes an object recognition algorithm identifier, an acquisition device identifier, and the range of the ROI region to be identified. The acquisition device identifier is specifically a camera identifier, but can also be an identifier for the camera and its shooting angle. When preparing to identify unidentified sample images, a preset object recognition algorithm is selected in advance to identify the ROI region. To balance recognition accuracy and efficiency, a target ROI region determination rule is matched from the ROI rule base based on partial information. Specifically, the object recognition algorithm identifier and the acquisition device identifier of the unidentified sample image are matched with multiple ROI region determination rules in the ROI rule base to obtain the target ROI region determination rule. The ROI region of the unidentified sample image is then extracted based on the range of the ROI region to be identified in the target ROI region determination rule. As an optional embodiment, a first ROI region extraction algorithm can be used to extract the initial ROI region of the unidentified sample image, and the ROI region of the unidentified sample image can be corrected according to the target ROI region determination rule.

[0057] The automatic sample generation method provided by this invention includes three processes: image acquisition, image recognition, and sample annotation. This invention improves the automation level, speed, and accuracy of the sample annotation process by utilizing a target sample review strategy, while simultaneously enhancing the speed and accuracy of the image recognition process by employing target ROI region determination rules. By automatically processing samples at multiple stages, it enables the automatic generation of image samples based on periodically acquired images, reducing reliance on manual intervention and improving the speed and quality of sample generation and updates. The target ROI region determination rules accurately locate the region to be identified, reducing the scope of image recognition and thus increasing the speed of sample recognition. Simultaneously, by identifying only objects within the designated area, interference from other areas is avoided, improving the accuracy of sample recognition.

[0058] Furthermore, in both image recognition and sample labeling processes, the present invention utilizes the ROI region identified by the image acquisition device as a medium to determine the rules and sample review strategies. The placement and application scenarios of different acquisition devices are different. By using the acquisition device markers, the automatic generation of samples can generate labeled samples that are suitable for the current application scenario, making the samples used for the final evolutionary algorithm parameters more targeted and further improving the accuracy of the algorithm.

[0059] Furthermore, while generating multiple sample review strategies, the process also includes statistical analysis of historical reviewed sample sets to generate multiple ROI region determination rules. Specifically, the historical reviewed sample sets are classified according to the recognition algorithm identifier and the acquisition device identifier to obtain multiple historical reviewed sample classes. The unidentified regions of each sample in each historical reviewed sample class are statistically analyzed, thereby generating multiple ROI region determination rules based on the recognition algorithm identifier, acquisition device identifier, and unidentified region of each historical reviewed sample class. Furthermore, after evaluating the first attribute of the first sample image according to the target sample review strategy, the process also includes adding the first sample image to the historical reviewed sample set. The first sample image carries at least the recognition algorithm identifier and the acquisition device identifier of the first sample image.

[0060] Furthermore, after evaluating the first attribute of the first sample image according to the target sample review strategy, the method further includes: adding the first sample image to the training sample pool; the algorithm reads the evaluated sample images in the training sample pool and updates the algorithm's parameters through training. Specifically, the algorithm can read all evaluated samples from the training sample pool according to a second preset period to train the algorithm and update its parameters; as another optional embodiment, the number of samples in the training sample pool is monitored, and if the number of samples in the training sample pool is greater than or equal to a preset sample number threshold, the algorithm is triggered to read all evaluated samples from the training sample pool to train the algorithm and update its parameters.

[0061] Existing technologies involve manually importing samples after selection for image detection and recognition algorithm training and periodic updates, resulting in low update frequency and high reliance on manual intervention. This invention, based on an automatic sample generation method, utilizes automatically labeled samples to automatically update algorithm parameters and periodically reads and updates the algorithm parameters from the labeled samples. This reduces the reliance on manual intervention in the algorithm's self-evolution process, accelerates the self-evolution speed, increases the frequency of self-evolution, and ultimately improves the accuracy of the image recognition algorithm.

[0062] Embodiment 2 of the present invention provides an automatic sample generation device, such as... Figure 2As shown, the automatic sample generation device includes: a strategy generation module for generating multiple sample review strategies, the multiple sample review strategies being used to evaluate a first attribute of a sample; an image acquisition module for acquiring an identified first sample image; and an image evaluation module for matching a target sample review strategy from the multiple sample review strategies based on first information of the first sample image, and evaluating the first attribute of the first sample image according to the target sample review strategy, wherein the identified first sample image is acquired according to a first cycle, and the first attribute is used to indicate the positive or negative type of the sample.

[0063] It is not difficult to see that this embodiment is a device embodiment corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0064] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by this invention; however, this does not mean that other units are absent from this embodiment.

[0065] Embodiment 3 of the present invention provides an algorithm self-evolution method, the algorithm self-evolution method comprising: triggering an algorithm to read a set of labeled sample images according to a first preset triggering condition, training an algorithm based on the read set of labeled sample images, and updating algorithm parameters; wherein, the set of labeled sample images includes multiple labeled sample images, each labeled sample image being an evaluated first sample image obtained according to the sample automatic generation method described in Embodiment 1.

[0066] The algorithm self-evolution method provided by this invention can trigger the algorithm to automatically update and iterate using a first preset trigger condition. Based on the automatic sample generation method, the algorithm self-evolution method of this invention does not require manual intervention. After the algorithm is deployed and applied, it can continuously acquire image data that conforms to the algorithm application scenario through an image acquisition device—a camera. By using the acquisition device to identify and continuously obtain ROI region determination rules and review strategies suitable for the current application scenario based on historically reviewed samples applicable to this type of application scenario, the algorithm can acquire, label, and use samples in a targeted manner, thereby improving the accuracy of the algorithm and the frequency of algorithm self-evolution.

[0067] Furthermore, the first preset trigger condition can be: whether the time since the last update has reached the second preset period; if so, the algorithm is triggered to read the labeled sample image set. As another optional embodiment, the first preset trigger condition can be: the number of unused labeled sample image sets is greater than or equal to a preset sample number threshold. Specifically, the number of unused labeled sample image sets is monitored, and if the number of unused labeled sample image sets is greater than or equal to the preset sample number threshold, the algorithm is triggered to read the labeled sample image set for training the algorithm to update the algorithm's parameters.

[0068] Embodiment 4 of the present invention provides an algorithm self-evolution device, which includes the sample automatic generation device as described in Embodiment 2, and further includes an algorithm self-evolution module, which is used to trigger the algorithm to read the labeled sample image set according to a first preset trigger condition, train the algorithm based on the read labeled sample image set, and update the algorithm parameters; wherein, the labeled sample image set includes multiple labeled sample images, and each labeled sample image is an evaluated first sample image obtained by the sample automatic generation device.

[0069] Furthermore, the first preset trigger condition can be: whether the time since the last update has reached the second preset period; if so, the algorithm is triggered to read the labeled sample image set. As another optional embodiment, the first preset trigger condition can be: the number of unused labeled sample image sets is greater than or equal to a preset sample number threshold. Specifically, the number of unused labeled sample image sets is monitored, and if the number of unused labeled sample image sets is greater than or equal to the preset sample number threshold, the algorithm is triggered to read the labeled sample image set for training the algorithm to update the algorithm's parameters.

[0070] The fifth embodiment of the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program for an algorithmic self-evolution method stored in the memory and executable on the processor, wherein:

[0071] The processor is used to call a computer program stored in memory to perform the following steps: generating multiple sample review strategies, the multiple sample review strategies being used to evaluate a first attribute of a sample; acquiring an identified first sample image; matching a target sample review strategy from the multiple sample review strategies based on first information of the first sample image; evaluating the first attribute of the first sample image according to the target sample review strategy, wherein the identified first sample image is acquired according to a first cycle, and the first attribute is used to indicate the positive or negative type of the sample.

[0072] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0073] Those skilled in the art will understand that the above embodiments are specific examples of implementing the present invention, and in practical applications, various changes in form and detail may be made without departing from the spirit and scope of the present invention.

Claims

1. A method for automatically generating samples, characterized in that, The automatic sample generation method includes: Multiple sample review strategies are generated, and the multiple sample review strategies are used to evaluate the first attribute of the sample; Obtain the image of the first identified sample; Based on the first information of the first sample image, a target sample review strategy is matched from the plurality of sample review strategies, and the first attribute of the first sample image is evaluated according to the target sample review strategy. The process involves acquiring the identified first sample image according to a first cycle, where the first attribute is used to indicate the positive or negative type of the sample, and the first information of the first sample image includes at least a first recognition algorithm identifier and a first acquisition device identifier; each of the multiple sample review strategies has a strategy attribute, which includes at least a second recognition algorithm identifier, a second acquisition device identifier, and a strategy confidence level; the first and second recognition algorithm identifiers are used to characterize the image recognition algorithm, and the first and second acquisition device identifiers are used to characterize the camera and camera angle used to acquire the image; Matching a target sample review strategy from the plurality of sample review strategies based on the first information of the first sample image includes: Match the first identification algorithm identifier and the first acquisition device identifier among multiple second identification algorithm identifiers and second acquisition device identifiers to determine the matched target review strategy. The second identification algorithm identifier and the second acquisition device identifier of the target review strategy are the same as the first identification algorithm identifier and the first acquisition device identifier of the first sample image. Evaluating the first attribute of the first sample image according to the target sample review strategy includes: evaluating the first attribute of the first sample image according to the strategy confidence of the target sample review strategy, wherein the first attribute includes at least correct, incorrect, minor, and inappropriate.

2. The automatic sample generation method according to claim 1, characterized in that, The strategy for generating multiple sample reviews includes: The historical reviewed sample set is statistically analyzed. Based on the identification algorithm identifier and the data acquisition device identifier, the historical reviewed sample set is classified to obtain multiple historical reviewed sample classes. The confidence level of each historical reviewed sample class is statistically calculated. Thus, multiple sample review strategies are generated based on the identification algorithm identifier, data acquisition device identifier, and confidence level of each historical reviewed sample class.

3. The automatic sample generation method according to claim 1, characterized in that, After evaluating the first attribute of the first sample image according to the target sample review strategy, the method further includes: The first sample image is added to the historical reviewed sample set, and the first sample image contains at least the first attribute and the target sample review strategy information.

4. The automatic sample generation method according to claim 1, characterized in that, Before acquiring the first identified sample image, the process also includes: Unidentified sample images are acquired, their Regions of Interest (ROIs) are extracted, and object recognition algorithms are used to identify the ROIs to obtain the first identified sample image.

5. The automatic sample generation method according to claim 4, characterized in that, The extraction of the Region of Interest (ROI) of the unidentified sample image includes: Based on the object recognition algorithm identifier and the acquisition device identifier of the unrecognized sample image, the target ROI region determination rule is matched from the ROI rule base, and the ROI region of the unrecognized sample image is extracted according to the target ROI region determination rule.

6. The automatic sample generation method according to claim 5, characterized in that, The ROI rule base includes multiple ROI area determination rules. Each ROI area determination rule includes an identification attribute, which includes at least the object recognition algorithm identifier, the data acquisition device identifier, and the range of the ROI area to be identified. The method involves matching target ROI region determination rules from the ROI rule base based on partial information, including: matching the object recognition algorithm identifier and the acquisition device identifier of the unrecognized sample image with multiple ROI region determination rules in the ROI rule base to obtain target ROI region determination rules, and extracting the ROI region of the unrecognized sample image based on the range of the ROI region to be identified in the target ROI region determination rules.

7. The automatic sample generation method according to claim 1, characterized in that, While generating multiple sample review strategies, it also includes statistical analysis of historical reviewed sample sets, classifying historical reviewed sample sets according to the identification algorithm identifier and the acquisition device identifier to obtain multiple historical reviewed sample classes, and statistically analyzing the regions to be identified for each sample in each historical reviewed sample class. Thus, multiple ROI region determination rules are generated based on the identification algorithm identifier, acquisition device identifier, and regions to be identified for each historical reviewed sample class.

8. The automatic sample generation method according to claim 5, characterized in that, The step of extracting the ROI region of the unidentified sample image according to the target ROI region determination rule specifically includes: extracting the initial ROI region of the unidentified sample image using a first ROI region extraction algorithm, and correcting the ROI region of the unidentified sample image according to the target ROI region determination rule.

9. The automatic sample generation method according to claim 1, characterized in that, The first sample image is added to the training sample pool. The algorithm reads the evaluated sample images in the training sample pool and updates the algorithm's parameters.

10. An automatic sample generation device, the automatic sample generation device comprising: The strategy generation module is used to generate multiple sample review strategies, which are used to evaluate the first attribute of the sample. The image acquisition module is used to acquire the first identified sample image; An image evaluation module is used to match a target sample review strategy from the plurality of sample review strategies based on the first information of the first sample image, and evaluate the first attribute of the first sample image according to the target sample review strategy. The process involves acquiring the identified first sample image according to a first cycle, where the first attribute is used to indicate the positive or negative type of the sample, and the first information of the first sample image includes at least a first recognition algorithm identifier and a first acquisition device identifier; each of the multiple sample review strategies has a strategy attribute, which includes at least a second recognition algorithm identifier, a second acquisition device identifier, and a strategy confidence level; the first and second recognition algorithm identifiers are used to characterize the image recognition algorithm, and the first and second acquisition device identifiers are used to characterize the camera and camera angle used to acquire the image; Matching a target sample review strategy from the plurality of sample review strategies based on the first information of the first sample image includes: Match the first identification algorithm identifier and the first acquisition device identifier among multiple second identification algorithm identifiers and second acquisition device identifiers to determine the matched target review strategy. The second identification algorithm identifier and the second acquisition device identifier of the target review strategy are the same as the first identification algorithm identifier and the first acquisition device identifier of the first sample image. Evaluating the first attribute of the first sample image according to the target sample review strategy includes: evaluating the first attribute of the first sample image according to the strategy confidence of the target sample review strategy, wherein the first attribute includes at least correct, incorrect, minor, and inappropriate.

11. An algorithm self-evolution method, the algorithm self-evolution method comprising: The algorithm is triggered according to the first preset trigger condition to read the labeled sample image set, and the algorithm is trained and the algorithm parameters are updated based on the read labeled sample image set; wherein, the labeled sample image set includes multiple labeled sample images, and each labeled sample image is an evaluated first sample image obtained by the automatic sample generation method according to any one of claims 1-9.

12. An algorithm self-evolution device, the algorithm self-evolution device comprising the sample automatic generation device as described in claim 10, the algorithm self-evolution device further comprising: The algorithm self-evolution module is used to trigger the algorithm to read the labeled sample image set according to the first preset trigger condition, train the algorithm based on the read labeled sample image set, and update the algorithm parameters. The labeled sample image set includes multiple labeled sample images, each labeled sample image being an evaluated first sample image obtained by the automatic sample generation device.

13. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps in the algorithm self-evolution method as described in claim 11.

Images