Image data processing method, storage medium, processor and system

Abstract

The application discloses an image data processing method, a storage medium, a processor and a system. The method comprises the following steps: acquiring an image to be recognized, wherein the image to be recognized contains at least one region, and at least one object exists in each region; processing image data of the image to be recognized to obtain distribution data corresponding to the at least one region; and determining whether a region type of the at least one region is a first region type according to the distribution data corresponding to the at least one region, wherein the object in the region corresponding to the first region type is not stored in a preset storage region. The application solves the technical problem of high region labeling cost caused by the fact that an effective region cannot be effectively screened out from multiple regions in an image in the prior art.

Description

Technical Field

[0001] This application relates to the field of image recognition, and more specifically, to an image data processing method, storage medium, processor, and system. Background Technology

[0002] In the process of algorithm development, it is generally assumed that the test data and training data are independent and identically distributed. However, in practical applications (such as remote sensing image land cover classification), due to the large number of data categories in the image and the fact that new categories of data are usually generated according to different actual needs, the data processed by the model after deployment is often not completely under control. That is, the distribution of the data received by the model may be different from the training data in the training phase. Such new categories of data are called OOD (Out-of-Distribution) samples or anomalous samples.

[0003] To recognize images containing out-of-deformation (OOD) samples, existing techniques involve collecting specific images and labeling all regions within those images based on actual needs, before training the model. This method increases the cost of region labeling and consequently reduces image recognition efficiency because it cannot effectively filter out valid regions from multiple areas in an image.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides an image data processing method, storage medium, processor, and system to at least solve the technical problem of high region annotation costs caused by the inability to effectively filter out effective regions from multiple regions in an image in the prior art.

[0006] According to one aspect of the embodiments of this application, an image data processing method is provided, comprising: acquiring an image to be identified, wherein the image to be identified contains at least one region, and each region contains at least one object; processing image data of the image to be identified to obtain distribution data corresponding to at least one region; determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, wherein the objects in the region corresponding to the first region type are not stored in a preset storage area.

[0007] Furthermore, the image data processing method also includes: calculating the image data of the image to be identified based on a preset machine learning model to obtain the mean and variance values ​​corresponding to at least one region, wherein the distribution data includes at least the mean and variance values, the mean representing the feature information of at least one region, and the magnitude of the variance value representing the probability that at least one region is a first region type.

[0008] Furthermore, the image data processing method also includes: when there is a first target region with a variance value greater than a preset threshold in at least one region, identifying the region type of the first target region as a first region type; when there is a second target region with a variance value less than or equal to a preset threshold in at least one region, identifying the region type of the second target region as a second region type, wherein the objects in the region corresponding to the second region type are stored in a preset storage region.

[0009] Furthermore, the image data processing method also includes: after identifying the region type of the first target region as the first region type, obtaining the first mean value corresponding to the first target region and the second mean value of the region corresponding to the second region type; comparing the first mean value and the second mean value, determining the target region with the highest similarity to the first target region from the region data stored in the preset storage area; and determining the region type corresponding to the target region as the initial annotation type corresponding to the first target region.

[0010] Furthermore, the image data processing method also includes: after determining that the region type corresponding to the target region is the initial annotation type corresponding to the first target region, responding to the update instruction, annotating the initial annotation type of the first target region to obtain the annotation region; updating the data in the preset dataset based on the annotation region, wherein the preset dataset is used to optimize and train the machine learning model.

[0011] Furthermore, the image data processing method also includes: after identifying the region type of the first target region as a first region type, in response to a deletion instruction, deleting the region corresponding to the first region type from at least one region.

[0012] Furthermore, the image data processing method further includes: before determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, obtaining first distribution information of the region corresponding to the first region type and second distribution information of the region corresponding to the second region type from a preset historical record; and determining a preset threshold based on the first distribution information and the second distribution information.

[0013] Furthermore, the image data processing method also includes: calculating the mean and variance values ​​corresponding to at least one region based on the image data of the image to be recognized using a preset machine learning model; resampling the variance values ​​corresponding to at least one region to obtain a scaling factor; classifying and calculating the mean values ​​corresponding to at least one region to obtain a score value corresponding to at least one region; calculating the ratio between the scaling factor and the score value to obtain a target score value; calculating the loss function value corresponding to the machine learning model based on the target score value; and optimizing the network parameters of the machine learning model when the loss function value meets preset conditions.

[0014] According to another aspect of the embodiments of this application, an image data processing method is also provided, including: reading an image to be identified, wherein the image to be identified contains at least one region, and each region contains at least one object; responding to an image recognition instruction, displaying distribution data corresponding to at least one region and recognition results corresponding to at least one region, wherein the distribution data corresponding to at least one region is obtained by processing the image data of the image to be identified, and the recognition results are obtained by identifying at least one region based on the distribution data corresponding to at least one region, and the recognition results characterize whether the region type of at least one region is a first region type, and the object in the region corresponding to the first region type is not stored in a preset storage area.

[0015] According to another aspect of the embodiments of this application, an image data processing method is also provided, comprising: a cloud server receiving an image to be identified containing at least one region sent by a terminal device, wherein at least one object exists in each region; the cloud server processing the image data of the image to be identified to obtain distribution data corresponding to at least one region; the cloud server determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, and obtaining a recognition result, wherein the object in the region corresponding to the first region type is not stored in a preset storage area; and the cloud server returning the recognition result to the terminal device.

[0016] Furthermore, the image data processing method also includes: after the cloud server returns the recognition results to the terminal device, the cloud server obtains the annotation results of the terminal device on the first region type of region in at least one region, and updates the preset dataset based on the annotated first region type of region to optimize the training of the machine learning model.

[0017] According to another aspect of the embodiments of this application, an image data processing method is also provided, comprising: acquiring a remote sensing image to be identified, wherein the remote sensing image contains at least one remote sensing object; processing the image pixels of the remote sensing image to obtain distribution data corresponding to at least one remote sensing object; determining whether the object type of at least one remote sensing object is a first object type based on the distribution data corresponding to at least one remote sensing object, wherein the object corresponding to the first object type is not stored in a preset storage area.

[0018] Furthermore, at least one remote sensing object includes at least one of the following: remote sensing ground features, building objects, and remote sensing water bodies.

[0019] According to another aspect of the embodiments of this application, an image data processing method is also provided, including: acquiring an image to be identified, wherein the image to be identified contains at least one item; processing the image pixels of the image to be identified to obtain distribution data corresponding to at least one item; identifying the item type corresponding to at least one item based on the distribution data corresponding to at least one item; and determining whether at least one item is a non-compliant item based on the item type.

[0020] According to another aspect of the embodiments of this application, a storage medium is also provided, the storage medium including a stored program, wherein, when the program is running, it controls the device where the storage medium is located to execute the above-described image data processing method.

[0021] According to another aspect of the embodiments of this application, a processor is also provided, which is used to run a program, wherein the program executes the above-described image data processing method when it runs.

[0022] According to another aspect of the embodiments of this application, a system for recognizing images is also provided, including: a processor; and a memory connected to the processor, for providing the processor with instructions to process the following processing steps: acquiring an image to be recognized, wherein the image to be recognized contains at least one region, and each region contains at least one object; processing image data of the image to be recognized to obtain distribution data corresponding to at least one region; determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, wherein the objects in the region corresponding to the first region type are not stored in a preset storage area.

[0023] In this embodiment of the invention, a method is adopted to determine whether a region is a valid region based on the distribution data of each region in an image. This involves acquiring the image to be identified, processing the image data of the image to be identified, obtaining the distribution data corresponding to at least one region, and then determining whether the region type of at least one region is a first region type based on the distribution data corresponding to the at least one region. The image to be identified contains at least one region, and each region contains at least one object. The objects in the regions corresponding to the first region type are not stored in a preset storage area.

[0024] In the above process, the distribution data corresponding to each region is acquired, and the region type of each region is determined based on the distribution data. This achieves effective filtering of valid regions in the image, making it easier for users to identify the regions that need to be labeled within the valid regions. This avoids the labeling of invalid regions in the image to be identified by related systems, thereby reducing the region labeling cost. Furthermore, in this application, by determining the region type of a region based on the distribution data corresponding to each region, the statistical analysis of classification probabilities and the setting of thresholds by humans are avoided, reducing the randomness caused by the statistical properties of the data. It also avoids using an additional classifier to classify the features of the original model, reducing additional computational overhead. This improves the accuracy of region type judgment and reduces computational cost.

[0025] Therefore, the solution provided in this application achieves the goal of determining whether a region is a valid region based on the distribution data of each region in the image, thereby reducing the technical effect of region annotation cost and solving the technical problem of high region annotation cost caused by the inability to effectively filter out valid regions from multiple regions in the image in the prior art. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0027] Figure 1 This is a hardware structure block diagram of an optional computer terminal (or mobile device) according to an embodiment of this application;

[0028] Figure 2 This is a schematic diagram of an optional image data processing method according to an embodiment of this application;

[0029] Figure 3 This is a flowchart of an optional image data processing method according to an embodiment of this application;

[0030] Figure 4 This is a flowchart of an optional image data processing method according to an embodiment of this application;

[0031] Figure 5 This is a schematic diagram of an optional image data processing method according to an embodiment of this application;

[0032] Figure 6 This is a schematic diagram of an optional human-computer interaction operation according to an embodiment of this application;

[0033] Figure 7 This is a schematic diagram of an optional image data processing method according to an embodiment of this application;

[0034] Figure 8 This is a schematic diagram of an optional image data processing method according to an embodiment of this application;

[0035] Figure 9 This is a schematic diagram of an optional image data processing method according to an embodiment of this application;

[0036] Figure 10 This is a structural block diagram of an optional computer terminal according to an embodiment of this application. Detailed Implementation

[0037] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0038] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0039] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0040] Land feature classification: Classifying each pixel unit in remote sensing image according to land feature type. Typically, the classification types can include nine categories, such as cultivated land, forest land, grassland, buildings, roads, structures, artificial excavations, bare land, and water. It is a semantic segmentation task.

[0041] Uncertainty quantification: Uncertainty estimation is performed on remote sensing images. In testing and actual use, if the uncertainty of the input sample is low, it is determined to be an ID (In-Distribution) sample; if the uncertainty of the input sample is high, it is determined to be an OOD (Out-of-Distribution) sample.

[0042] Example 1

[0043] According to an embodiment of this application, an embodiment of an image data processing method is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0044] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing an image data processing method is shown. Figure 1 As shown, the computer terminal 10 (or mobile device 10) may include one or more processors (shown as 102a, 102b, ..., 102n in the figure) (the processor may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0045] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be wholly or partially embodied in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or wholly or partially integrated into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuit serves as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0046] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the image data processing method in this embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the above-mentioned image data processing method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0047] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0048] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0049] It should be noted here that, in some optional embodiments, the above... Figure 1 The computer device (or mobile device) shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 1 This is only one instance of a specific particular instance and is intended to illustrate the types of components that may exist in the aforementioned computer device (or mobile device).

[0050] Under the aforementioned operating environment, this application provides the following: Figure 2 The image data processing method shown. Figure 2 This is a schematic diagram of an optional image data processing method according to an embodiment of this application.

[0051] Step S202: Obtain the image to be identified, wherein the image to be identified contains at least one region, and at least one object exists in each region.

[0052] Optionally, this application can be applied to scenarios such as remote sensing image feature classification, e-commerce item classification, and traffic safety item classification. Feature classification is a fundamental remote sensing capability, a semantic segmentation task that classifies each pixel of a remote sensing image (i.e., the image to be identified) to obtain its basic category. Currently, there are nine basic categories: cultivated land, forest land, grassland, buildings, roads, structures, artificial excavations, bare land, and water. These basic categories are included in the training samples of the machine learning model. In practical applications, machine learning models often need to analyze data from new scene regions in images due to the need to classify specific categories, significant differences in satellite imaging, and poor image quality. These new scene regions differ from the training samples used in the machine learning model's training phase and can therefore be called OOD samples or anomalous samples. Specifically, new scenes are constantly emerging. If the classification results for new scenes are of low quality, it will negatively impact the user experience. For example, compared to the nine categories used in conventional land cover classification applications, the yurts scattered across vast grasslands in Inner Mongolia are unique. These samples are typically specific to certain needs and are rarely seen in general training sets. Directly applying machine learning models can easily lead to misclassification of yurts in remote sensing images, affecting the user experience. Therefore, during image recognition, regions with uncertain new scenes (i.e., OOD samples) can be selected from the image. This allows users to choose whether to accumulate new scene data, filter out desired regions from these uncertain new scene regions, and then label these desired regions as samples to optimize the model, improve the model's classification accuracy for new scenes, and enhance algorithm optimization efficiency.

[0053] Specifically, the image to be identified can be acquired through electronic devices, application systems, servers, etc. In this embodiment, the image to be identified is acquired through an image recognition system. The image to be identified can be a remote sensing image for land cover classification. Each region in the image can be a single pixel unit or composed of multiple pixel units. Each region contains one object or multiple objects belonging to the same category. For example, in this embodiment, the image to be identified can be a remote sensing image of Inner Mongolia. Each pixel in this image is defined as a region. The objects in the region can be any of the following: grass, trees, houses, yurts, water, etc. Each region can contain one tree (i.e., one object) or multiple trees (i.e., multiple objects belonging to the same category).

[0054] It should be noted that the image to be identified is obtained in order to subsequently determine the type of the first region.

[0055] Step S204: Process the image data of the image to be identified to obtain distribution data corresponding to at least one region.

[0056] In step S204, the image recognition system can process the image to be recognized based on a machine learning model with uncertainty quantization to obtain distribution data corresponding to each region. The machine learning model includes at least network classification models or other models with classification functions for different application scenarios, and may also include feature generation models, feature recognition models, etc. Uncertainty quantization represents the uncertainty estimation of the image, and the distribution data can be used to characterize the category and uncertainty of the region. The uncertainty is used to characterize the probability that at least one region is of the first region type (i.e., a new scene region (OOD sample)).

[0057] Optionally, the distribution data can be any of the following distribution types, depending on the application scenario: binomial distribution, geometric distribution, hypergeometric distribution, Poisson distribution, exponential distribution, Gaussian distribution, uniform distribution, and chi-square distribution. Preferably, the distribution data is Gaussian distributed data. The distribution data includes at least a mean and a variance, where the mean can represent the characteristic information of at least one region, and the magnitude of the variance can represent the probability that at least one region belongs to the first region type.

[0058] It should be noted that image recognition systems can also obtain data for determining the region type of at least one area based on probabilistic statistical detection or classifier-based detection methods. Probabilistic statistical detection involves statistically analyzing the probabilities output by the machine learning model, maximizing the difference between ID (In-Distribution) samples (i.e., non-abnormal samples) and OOD samples, to select a threshold in subsequent stages to determine whether a sample belongs to ID or OOD. Classifier detection, on the other hand, uses an additional classifier to classify the features of the original model. However, while probabilistic statistical detection is simple and direct, requiring no modification to the machine learning model and only statistical analysis of classification probabilities, it suffers from drawbacks. It requires classifying and statistically analyzing all samples, making the process complex. Furthermore, the selection of the threshold significantly impacts the results, making it highly susceptible to human factors and the statistical distribution characteristics of the data, resulting in low versatility. Classifier detection, on the other hand, requires dedicated OOD training samples and incurs substantial additional computational overhead. Therefore, by obtaining distribution data corresponding to at least one region, on the one hand, it is convenient to determine the region type of the region in the future. On the other hand, it avoids statistical analysis of classification probability and setting thresholds manually, reduces the randomness caused by the statistical properties of the data, and avoids using an additional classifier to classify the features of the original model, reducing additional computational overhead. This makes it easier to improve the accuracy of subsequent region type judgment and reduce computational costs.

[0059] Step S206: Determine whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, wherein the objects in the region corresponding to the first region type are not stored in a preset storage region.

[0060] In step S206, the image recognition system can compare the values ​​in the distribution data of each region with a preset threshold to determine whether the region type of at least one region is the first region type; alternatively, based on the distribution of values ​​in the distribution data of each region, the regions can be sorted from highest to lowest probability of being the first region type, and the regions in the top X% of the sorted results based on the historical OOD sample occurrence rate X% are determined as the first region type. The objects in the regions corresponding to the first region type are not stored in the preset storage area, i.e., they are not included in the training samples. When a region is determined to be the first region type, it represents that region as a new scene region in the image, i.e., an OOD sample.

[0061] Furthermore, after determining whether the region type of at least one area is the first region type, the image recognition system can output the corresponding results and the selected OOD samples to the user, so as to inform the user that there are OOD samples in the image to be recognized, and let the user decide whether to accumulate new scene data, further filter out the desired region, so that the image recognition system can label the desired region and use the labeled desired region as a sample to update the data in the preset dataset, so as to optimize and train the machine learning model based on the preset dataset, thereby accelerating the cycle of model transfer and optimization for specific scene applications.

[0062] It should be noted that OOD samples include at least regions that can improve the classification accuracy of relevant models for identifying new scenes (e.g., for remote sensing images of Inner Mongolia, regions with yurts are regions that can improve the classification accuracy of relevant models). They may also include regions with poor image quality or regions imaged by different satellites. Therefore, OOD samples can be considered valid regions. However, regions that are not of the first region type are typically classified as conventional land cover types and cannot improve the classification accuracy of relevant models for identifying new scenes (e.g., for remote sensing images of Inner Mongolia, regions with buildings cannot improve the classification accuracy of relevant models). Therefore, these regions can be considered invalid regions. In existing technologies, when identifying images of new scenes, the relevant system labels all regions in the image (including the aforementioned valid and invalid regions), thus increasing the region labeling cost. Therefore, by determining the region type of at least one region based on the distribution data corresponding to at least one region, effective filtering of valid samples in the image to be identified is achieved. This facilitates users in identifying the regions that need to be labeled within the valid regions, avoiding the labeling of invalid regions in the image to be identified, thereby reducing the region labeling cost.

[0063] Based on the scheme defined in steps S202 to S206 above, it can be understood that in this embodiment of the invention, the method of determining whether a region is a valid region based on the distribution data of each region in the image is adopted. This involves acquiring the image to be identified, processing the image data of the image to be identified, obtaining the distribution data corresponding to at least one region, and then determining whether the region type of at least one region is a first region type based on the distribution data corresponding to the at least one region. The image to be identified contains at least one region, and each region contains at least one object. The objects in the regions corresponding to the first region type are not stored in a preset storage area.

[0064] It is noteworthy that in the above process, acquiring the distribution data corresponding to each region and determining the region type based on this data enables effective filtering of valid regions in the image. This facilitates users in identifying the regions that need to be labeled within the valid regions, avoiding the labeling of invalid regions in the image to be recognized by related systems, thereby reducing the cost of region labeling. Furthermore, in this application, determining the region type based on the distribution data corresponding to each region avoids statistical analysis of classification probabilities and manually setting thresholds, reducing the randomness caused by the statistical properties of the data. It also avoids using additional classifiers to classify the features of the original model, reducing additional computational overhead, thereby improving the accuracy of region type judgment and reducing computational costs.

[0065] Therefore, the solution provided in this application achieves the goal of determining whether a region is a valid region based on the distribution data of each region in the image, thereby reducing the technical effect of region annotation cost and solving the technical problem of high region annotation cost caused by the inability to effectively filter out valid regions from multiple regions in the image in the prior art.

[0066] In one optional embodiment, the image data of the image to be recognized is processed to obtain distribution data corresponding to at least one region. The image recognition system calculates the mean and variance corresponding to at least one region based on the image data of the image to be recognized using a preset machine learning model. The distribution data includes at least the mean and variance. The mean represents the feature information of at least one region, and the magnitude of the variance represents the probability that at least one region is a first region type.

[0067] The image recognition system can calculate the image data of the image to be recognized based on a preset machine learning model to obtain distribution data of any type, such as binomial distribution, geometric distribution, hypergeometric distribution, Poisson distribution, exponential distribution, Gaussian distribution, uniform distribution, and chi-square distribution. Preferably, in this embodiment, the image recognition system can perform Gaussian calculation on the image data of the image to be recognized based on the preset machine learning model to obtain Gaussian distribution data. The mean of this distribution data is used by the image recognition system to determine the category to which the area belongs (e.g., cultivated land, forest land, grassland, buildings, roads, structures, artificial excavated land, bare land, and water area, etc.), and the variance is used by the image recognition system to determine whether the area is an OOD sample (i.e., the aforementioned first area type). The larger the variance value, the higher the uncertainty of the area, that is, the higher the probability that the area is an OOD sample.

[0068] It should be noted that by calculating the mean and variance of at least one region from the image data of the image to be recognized, the category to which the region belongs is determined, and the region type is determined.

[0069] In an optional embodiment, during the process of determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, the image recognition system may identify the region type of the first target region as the first region type when there is a first target region with a variance value greater than a preset threshold in at least one region, and identify the region type of the second target region as the second region type when there is a second target region with a variance value less than or equal to the preset threshold in at least one region, wherein the objects in the region corresponding to the second region type are stored in a preset storage area.

[0070] Optionally, the preset threshold can be 1, or the user can set other threshold values ​​according to specific application requirements and discrimination accuracy requirements. In this embodiment, if the preset threshold is 1, then when the variance value of the region is greater than 1, the image recognition system determines that the region is of the first region type; when the variance value of the region is less than 1, the image region system determines that the region is of the second region type. The second region type is stored in a preset storage area, that is, it is included in the training samples.

[0071] Taking a remote sensing image of Inner Mongolia as an example, the multiple regions in this image can contain any of the following objects: grassland, roads, bare land, water, and yurts. Since grassland, roads, bare land, and water belong to conventional land cover classification types, they are usually included in the training samples during the machine learning model training phase. Therefore, the variance of regions containing these objects is usually close to 1 and less than 1, indicating low uncertainty. The image recognition system classifies these regions as the second region type. However, yurts, which are objects created for specific needs, are usually not included in the training samples during the machine learning model training phase. Therefore, the variance of regions containing yurts is usually greater than 1, indicating relatively high uncertainty. The image recognition system can classify these regions as the first region type, thereby achieving effective filtering of valid regions in the image.

[0072] It should be noted that by setting a preset threshold and comparing the distribution data corresponding to the region with the preset threshold to determine the region type, the effective division of valid and invalid regions is achieved.

[0073] In one optional embodiment, after identifying the region type of the first target region as the first region type, the image recognition system can obtain the first mean value corresponding to the first target region and the second mean value of the region corresponding to the second region type, and then compare the first mean value with the second mean value to determine the target region with the highest similarity to the first target region from the region data stored in the preset storage area, thereby determining that the region type corresponding to the target region is the initial annotation type corresponding to the first target region.

[0074] Optionally, after the image recognition system determines the region type of each region in the image based on a machine learning model, the image recognition system obtains the first mean value corresponding to the first target region and the second mean value of the region corresponding to the second region type. The second mean value varies depending on the object within the region. For any first target region, the image recognition system can determine the second mean value closest to its mean value, and then, based on this second mean value, determine the target region with the highest similarity to the first target region from the region data stored in a preset storage area, and determine the region type corresponding to the target region as the initial annotation type corresponding to the first target region.

[0075] For example, for areas with poor image quality, an image recognition system might identify them as the first target area. If there are blurry houses in this area, the first mean value calculated by the image recognition system for this area is generally close to the second mean value of areas with structures in the same image. Therefore, the image recognition system can determine the target area with the highest similarity to the first target area from the area data stored in the storage area based on this second mean value. If the area type of the target area is a structure, then the initial label type of the aforementioned first target area is also determined to be a structure. Optionally, for areas with new scenes (such as yurts), the image recognition system can also determine its corresponding initial label type.

[0076] It should be noted that by determining the initial annotation type corresponding to the first target area, it is easier for the user to determine whether the first target area is the desired area.

[0077] In one optional embodiment, after determining that the region type corresponding to the target region is the initial annotation type corresponding to the first target region, the image recognition system can respond to the update command, annotate the initial annotation type of the first target region to obtain the annotated region, and then update the data in the preset dataset based on the annotated region, wherein the preset dataset is used to optimize and train the machine learning model.

[0078] Optionally, when a user obtains the initial annotation type corresponding to each first target region output by the image recognition system, the user can determine whether the initial annotation type is correct based on their own recognition of the first target region, and input an update instruction to the image recognition system to annotate the first target region based on the initial annotation type or a new type, thereby obtaining an annotated region. The annotated region is then saved as a new sample to a preset dataset, and the machine learning model is optimized and trained based on the updated preset dataset, so that the machine learning model can be better suited for classifying new scenes.

[0079] Specifically, when a user obtains two regions whose initial label type is both structures, if the object in one region is a blurry house and the object in the other region is a yurt, the user can input an update command into the image recognition system. The region with the blurry house object will be labeled with the same type as the initial label, and the region with the yurt object will be labeled as a yurt. At least the labeled region with the yurt object will be saved as a new sample to the preset dataset, so that the network model can better identify and classify remote sensing images of Inner Mongolia.

[0080] It should be noted that by annotating the first target region to obtain the labeled region, and updating the data in the preset dataset based on the labeled region, the machine learning model can be better adapted to classify new scenes, thereby improving the recognition efficiency and accuracy of the machine learning model.

[0081] In one alternative embodiment, after identifying the region type of the first target region as a first region type, the image recognition system can respond to a deletion instruction and delete the region corresponding to the first region type from at least one region.

[0082] Optionally, after the image recognition system identifies the region type of the first target region as the first region type, the image recognition system can output the classification result of that region, and can also output the uncertainty score, i.e., the variance value, of the first target region. It should be noted that when the image recognition system fails to identify the first target region, the image recognition system only outputs the classification result of that region.

[0083] Furthermore, the user can determine whether to discard the region corresponding to the first target region in at least one region of the image based on their needs. The discarded region can be all or only a portion of the region corresponding to the first target region. When the user determines that at least one region needs to be discarded, the user can send a deletion command to the image recognition system, which will respond to the deletion command by deleting the user-specified region from at least one region, or simultaneously deleting other regions with the same mean as the specified region.

[0084] Specifically, when a user determines that the area corresponding to the first target area is an undesirable area, such as the area where the object is a blurry house, using it as a sample cannot optimize the model's classification effect on the remote sensing image of Inner Mongolia, the user can send a deletion command to the image recognition system, which will then delete the area where the object is a blurry house in at least one region, or simultaneously delete other regions with the same mean as the area, in order to delete the invalid area.

[0085] It should be noted that by responding to the deletion command and deleting the region corresponding to the first region type from at least one region, the effective region can be filtered more effectively, thereby improving the efficiency of this application.

[0086] In an optional embodiment, before determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, the image recognition system can obtain the first distribution information of the region corresponding to the first region type and the second distribution information of the region corresponding to the second region type from a preset historical record, thereby determining a preset threshold based on the first distribution information and the second distribution information.

[0087] Optionally, the image recognition system can retrieve historical records from a preset storage area or other storage device, and obtain first distribution information and second distribution information from the historical records. The first distribution information may include the numerical distribution of variance values ​​for the region corresponding to the first region type, and the second distribution information may include the numerical distribution of variance values ​​for the region corresponding to the second region type. The image recognition system can determine a preset threshold such that the preset threshold can distinguish between the values ​​in the first distribution information and the values ​​in the second distribution information. For example, when the values ​​in the first distribution information are distributed in the interval (2, 4) and the values ​​in the second distribution information are distributed in the interval (5, 6), the image recognition system can select any value in the interval (4, 5) as the preset threshold.

[0088] It should be noted that by determining the preset threshold based on historical records, the determined preset threshold is made more reasonable, thereby improving the accuracy of the application in filtering the effective area.

[0089] In one optional embodiment, after calculating the mean and variance values ​​of at least one region based on the image data of the image to be recognized using a preset machine learning model, the image recognition system can resample the variance values ​​of at least one region to obtain a scaling factor, then classify and calculate the mean values ​​of at least one region to obtain a score value corresponding to at least one region, then calculate the ratio between the scaling factor and the score value to obtain a target score value, and calculate the loss function value corresponding to the machine learning model based on the target score value, thereby optimizing the network parameters of the machine learning model when the loss function value meets preset conditions.

[0090] Optionally, the image recognition system can obtain the mean and variance values ​​of each region output by the machine learning model, and then resample the variance values ​​corresponding to at least one region using a standard normal distribution ∈ ~N(0,1) to obtain a scaling factor (i.e., scaling coefficient):

[0091] t=v·∈

[0092] Where t represents the scaling factor, v represents the variance, and ∈ represents the standard normal distribution. Then, the image recognition system can classify and calculate the mean value corresponding to at least one region based on the classification function, obtaining the score value corresponding to at least one region, and obtain the target score value based on the following formula:

[0093] s′(k)=s(k) / t

[0094] Where s′(k) represents the target score, s(k) represents the score corresponding to at least one region, and t represents the scaling factor. The target score is used to calculate the loss function value (Loss) of the machine learning model, and when the loss function value meets preset conditions, the network parameters of the machine learning model are optimized.

[0095] It should be noted that by optimizing the network parameters of the machine learning model based on the mean and variance values ​​corresponding to at least one region, the machine learning model was trained, thereby improving the classification accuracy of the machine learning model.

[0096] In one optional embodiment, an application scenario of this application is described. For example... Figure 3 As shown, the image recognition system first inputs remote sensing image samples of land cover classification into a machine learning model to train the model. Specifically, the machine learning model calculates the mean and variance of each region in the remote sensing image of land cover classification, where the mean represents the feature information of at least one region (i.e., Figure 3 The variance of the feature vectors in the model represents the probability that at least one region belongs to the first region type (i.e., the feature vectors in the model). Figure 3 (Uncertainty in the process). Then, the image recognition system calculates a scaling factor based on the variance, calculates the score of each region based on the mean, and calculates the loss function value corresponding to the machine learning model based on the scaling factor and the score value. In this way, the network model parameters are optimized based on the loss function value to train the machine learning model.

[0097] Furthermore, after obtaining the trained machine learning model, the image recognition system can input the remote sensing image of the land cover classification to be identified into the trained machine learning model, so that the trained machine learning model can calculate the mean and variance values ​​of each region in the image. Then, the image recognition system can determine whether at least one region belongs to the first region type based on the variance value, that is, whether the region is an OOD sample. For regions that are not OOD samples, the image recognition system will not perform further processing on those regions. For regions that are OOD samples, the image recognition system will output them to the user for the user to discard or further label. Regions that are further labeled are used as new samples to optimize the dataset for specific application scenarios.

[0098] Furthermore, such as Figure 4 As shown, after the image recognition system obtains the mean (i.e., the land cover classification result) and variance (uncertainty judgment result) of each region of the image to be recognized from the machine learning model output, if the variance (i.e., the uncertainty score) is less than a preset threshold, the image recognition system determines the region as a deterministic result (i.e., the region type is the second region type, and the region is an ID sample) and outputs the land cover classification result for that region. If the variance (i.e., the uncertainty score) is greater than the preset threshold, the image recognition system determines the region as an uncertain result (i.e., the region type is the first region type, and the region is an OOD sample), and outputs the land cover classification result and its variance value for that region to inform the user that the result is uncertain. Afterwards, the user can decide whether to collect the OOD sample based on their needs. If the user decides not to collect the OOD sample, they can input a delete command to have the image recognition system delete the region (i.e., discard the result). If the user decides to collect the OOD sample, they can input an update command to have the image recognition system label the region and accumulate the OOD sample to update the data in the preset dataset, which is then used to optimize and train the machine learning model.

[0099] It should be noted that this application uses a machine learning model to calculate the mean (i.e., feature information) and variance (i.e., uncertainty) of each region in the image to be identified. By quantifying the magnitude of uncertainty, OOD samples are identified, avoiding the need to pre-train the model using OOD training samples, thus achieving effective and rapid filtering of valid regions. Furthermore, this application informs users, allowing them to decide whether to accumulate OOD data, which can directly and quickly select the desired region, thus avoiding the need to label all regions in the image, greatly reducing labeling costs, and thereby accelerating the collection, labeling, and model training of samples in specific scenarios, shortening the algorithm optimization cycle.

[0100] Therefore, the solution provided in this application achieves the goal of determining whether a region is a valid region based on the distribution data of each region in the image, thereby reducing the technical effect of region annotation cost and solving the technical problem of high region annotation cost caused by the inability to effectively filter out valid regions from multiple regions in the image in the prior art.

[0101] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0102] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0103] Example 2

[0104] Embodiments of this application also provide an image data processing method. Optionally, Figure 5 This is a schematic diagram of an optional image data processing method according to an embodiment of this application, such as... Figure 5 As shown, the method includes:

[0105] Step S302: Read the image to be identified, wherein the image to be identified contains at least one region, and each region contains at least one object;

[0106] Step S304: In response to the image recognition instruction, display the distribution data corresponding to at least one region and the recognition result corresponding to at least one region. The distribution data corresponding to at least one region is obtained by processing the image data of the image to be recognized. The recognition result is obtained by recognizing at least one region based on the distribution data corresponding to at least one region. The recognition result indicates whether the region type of at least one region is a first region type. The object in the region corresponding to the first region type is not stored in the preset storage area.

[0107] Specifically, in practical applications, such as Figure 6 As shown, a user can input an image to be recognized into the image recognition system, which then reads the image. Next, the user inputs an image recognition command to initiate the process. Based on this command, the system feeds the image into a pre-defined machine learning model. The model then calculates the image data to obtain distribution data for at least one region. This distribution data includes at least the mean and variance. The mean represents the feature information of at least one region, and the variance represents the probability that the at least one region belongs to a first region type.

[0108] Furthermore, the image recognition system compares the variance value in the distribution data with a preset threshold to determine whether at least one area belongs to the first area type, i.e., whether at least one area is an ODD sample, thereby determining the corresponding recognition result. The recognition result includes at least the variance value (i.e., the uncertainty score). After the image recognition system determines the distribution data and recognition result, it can display them on a corresponding display device so that the user can view the land cover classification results for each area in the image and whether each area is an ODD sample. When the image recognition system does not identify an ODD sample, the image recognition module only outputs the distribution data.

[0109] Afterwards, users can decide whether to collect the OOD sample based on their needs. When a user decides not to collect the OOD sample, they can input a deletion command into the image recognition system to delete the region (i.e., discard the result). When a user decides to collect the OOD sample, they can input an update command into the image recognition system to label the region and accumulate the data in the preset dataset to update the data of the OOD sample, which can then be used to optimize and train the machine learning model.

[0110] It should be noted that this application informs the user by reading the image to be identified and displaying the distribution data corresponding to at least one region in the image and the recognition result corresponding to at least one region. This makes it easier for the user to decide whether to accumulate OOD data and directly and quickly select the desired region, thereby avoiding the need to label all regions in the image, greatly reducing the labeling cost, and thus accelerating the collection, labeling and model training of samples in specific scenarios, and shortening the algorithm optimization cycle.

[0111] Example 3

[0112] Embodiments of this application also provide an image data processing method, optionally, Figure 7This is a schematic diagram of an optional image data processing method according to an embodiment of this application, such as... Figure 7 As shown, the method includes:

[0113] Step S402: The cloud server receives an image to be identified from the terminal device, which contains at least one region, wherein at least one object exists in each region;

[0114] Step S404: The cloud server processes the image data of the image to be recognized to obtain distribution data corresponding to at least one region. The size of the distribution data represents the probability that at least one region is a first region type. The objects in the first region type are objects that are not stored in the preset storage area.

[0115] Step S406: The cloud server determines whether the region type of at least one region is the first region type based on the distribution data corresponding to at least one region, and obtains the identification result, wherein the objects in the region corresponding to the first region type are not stored in the preset storage region.

[0116] In step S408, the cloud server returns the recognition result to the terminal device.

[0117] Specifically, in practical applications, the aforementioned terminal device can be a mobile phone, tablet, computer, portable wearable device, or other device. When image recognition is required, the user can send the image to be recognized stored in the terminal device's memory, or an image obtained from the internet, to the cloud server. The terminal device can transmit data with the cloud server via WiFi, data interface, or other methods. After receiving the image, the cloud server can process the image data using a machine learning model to obtain distribution data for at least one region in the image. This distribution data includes at least a variance value, which represents the probability that at least one region is of a first region type. It may also include a mean value, which represents the feature information of at least one region. The cloud server then compares the variance value in the distribution data with a preset threshold to determine whether the region type of at least one region is the first region type, i.e., whether at least one region is an ODD sample, thereby determining the corresponding recognition result.

[0118] Furthermore, once the cloud server has determined the recognition result, it can return the result to the terminal device for the user to access in real time.

[0119] In one optional embodiment, after the cloud server returns the recognition results to the terminal device, the cloud server obtains the annotation results of the terminal device on the first region type of region in at least one region, and updates the preset dataset based on the annotated first region type of region to optimize the training of the machine learning model.

[0120] Specifically, after a user receives the recognition results output by the cloud server via their terminal device, the user can decide whether to collect the OOD sample based on their needs. If the user decides not to collect the OOD sample, they can send a deletion command to the cloud server via their terminal device to delete the region (i.e., discard the result). If the user decides to collect the OOD sample, they can input an update command to the cloud server via their terminal device to obtain the annotation results of the terminal device on at least one region of the first region type, and then annotate the region based on the annotation results. The cloud server then updates the data in the preset dataset based on the annotated regions of the first region type, which are then used to optimize and train the machine learning model.

[0121] It should be noted that this application receives the image to be recognized from the terminal device via a cloud server and returns the recognition result corresponding to at least one region to the terminal device. This informs the user, allowing them to decide whether to accumulate OOD data and quickly select the desired region. This avoids labeling all regions in the image, significantly reducing labeling costs and accelerating sample collection, labeling, and model training in specific scenarios, thus shortening the algorithm optimization cycle. Furthermore, this application determines the recognition result of the image to be recognized via a cloud server and sends the result to the terminal device, enabling users to obtain the information in real time and improving the user experience.

[0122] Example 3

[0123] Embodiments of this application also provide an image data processing method, optionally, Figure 8 This is a schematic diagram of an optional image data processing method according to an embodiment of this application, such as... Figure 8 As shown, the method includes:

[0124] Step S502: Obtain the remote sensing image to be identified, wherein the remote sensing image contains at least one remote sensing object.

[0125] In step S502, the image to be identified can be a remote sensing image of various regions, such as a remote sensing image of a plateau region, a remote sensing image of a basin region, a remote sensing image of a desert region, or a remote sensing image of Inner Mongolia or Hainan. The remote sensing objects in the remote sensing image can be people, buildings, plants, animals, water flow, etc.

[0126] Step S504: Process the image pixels of the remote sensing image to obtain distribution data corresponding to at least one remote sensing object.

[0127] In step S504, the image recognition system can process each pixel in the remote sensing image individually based on a machine learning model, or it can process each set of pixels in the remote sensing image individually. Each set of pixels consists of multiple pixels, and each pixel or set of pixels contains one object or multiple objects belonging to the same category, thereby obtaining distribution data corresponding to at least one remote sensing object. For example, in this embodiment, the image to be identified can be a remote sensing image of a desert region. The object present in each pixel or set of pixels in this remote sensing image can be any of the following: grass, trees, houses, cacti, camels, water, etc. Each pixel or set of pixels can contain one cactus (i.e., one object) or multiple cacti (i.e., multiple objects belonging to the same category).

[0128] It should be noted that by obtaining the distribution data corresponding to at least one remote sensing object, it is possible to accurately determine the object type of each remote sensing object in the future.

[0129] Step S506: Determine whether the object type of at least one remote sensing object is a first object type based on the distribution data corresponding to at least one remote sensing object, wherein the object corresponding to the first object type is not stored in a preset storage area.

[0130] In step S506, the image recognition system can compare the values ​​in the distribution data of each remote sensing object with a preset threshold to determine whether the object type of at least one remote sensing object is a first object type. When the object is determined to be of the first type, it indicates that the object is a new object compared to objects in historical training samples or known data. For example, when judging the object type of each object in a remote sensing image of a desert region, since cacti and camels are unique to desert regions and are rarely seen in general training sets, the variance values ​​in the distribution data corresponding to cacti and camels will be higher than the variance values ​​in the distribution data corresponding to basic objects (such as grass, trees, and houses). Therefore, based on an appropriate preset threshold, cacti and camels can be effectively filtered out from at least one remote sensing object.

[0131] Furthermore, after identifying the remote sensing objects of the first object type (such as the aforementioned cactus and camel), users can label the desired remote sensing objects and add them as new samples to the preset dataset to optimize and train the machine learning model. This will enable the machine learning model to effectively classify the various objects in the remote sensing area when the corresponding remote sensing image (such as the aforementioned desert image) is input into the machine learning model next time, thereby improving the classification effect.

[0132] It should be noted that by determining whether the object type of at least one remote sensing object is the first object type based on the distribution data corresponding to at least one remote sensing object, effective filtering of valid remote sensing objects in the remote sensing image to be identified is achieved. This makes it easier for users to identify the remote sensing objects that need to be labeled among the valid remote sensing objects, avoids labeling invalid remote sensing objects in the remote sensing image to be identified, and thus reduces the cost of regional labeling.

[0133] Based on the scheme defined in steps S502 to S506 above, it can be understood that in this embodiment, the method of determining whether a remote sensing object is a valid remote sensing object based on the distribution data of each remote sensing object in the remote sensing image is adopted. By acquiring the remote sensing image to be identified, and then processing the image pixels of the remote sensing image, the distribution data corresponding to at least one remote sensing object is obtained. Thus, the object type of at least one remote sensing object is determined to be a first object type based on the distribution data corresponding to at least one remote sensing object. The remote sensing image contains at least one remote sensing object, and the object corresponding to the first object type is not stored in the preset storage area.

[0134] It is noteworthy that in the above process, acquiring the distribution data corresponding to each remote sensing object and determining the object type of the remote sensing object based on the distribution data enables effective filtering of valid remote sensing objects in the remote sensing image. This facilitates users in identifying the remote sensing objects that need to be labeled from among the valid remote sensing objects, avoiding the labeling of invalid remote sensing objects in the remote sensing image to be identified by related systems, thereby reducing the cost of remote sensing object labeling. Furthermore, in this application, determining the object type of the remote sensing object based on the distribution data corresponding to each remote sensing object avoids statistical analysis of classification probabilities and manually setting thresholds, reducing the randomness caused by the statistical nature of the data. It also avoids using an additional classifier to classify the features of the original model, reducing additional computational overhead, thereby improving the accuracy of judging the object type of the remote sensing object and reducing computational costs.

[0135] Therefore, the solution provided in this application achieves the goal of determining whether a remote sensing object is a valid remote sensing object based on the distribution data of each remote sensing object in the image, thereby reducing the technical effect of remote sensing object annotation cost and solving the technical problem of high remote sensing object annotation cost caused by the inability to effectively filter out valid remote sensing objects from multiple remote sensing objects in the image in the prior art.

[0136] In one alternative embodiment, at least one remote sensing object includes at least one of the following: a remote sensing ground object, a building object, or a remote sensing water body object.

[0137] Optionally, remote sensing ground objects can include objects such as cultivated land, forest land, grassland, buildings, roads, structures, artificial excavations, bare land, and water bodies, as well as objects such as animals. Building objects can be further subdivided into various building objects, such as commercial buildings, residential buildings, and amusement park buildings. Remote sensing water objects can include objects such as lakes, seas, and rivers.

[0138] Example 4

[0139] Embodiments of this application also provide an image data processing method, optionally, Figure 9 This is a schematic diagram of an optional image data processing method according to an embodiment of this application, such as... Figure 9 As shown, the method includes:

[0140] Step 602: Obtain the image to be identified, wherein the image to be identified contains at least one item.

[0141] In step S602, the image to be identified can be an image acquired in an e-commerce scenario or a traffic safety scenario, and the image to be identified includes at least one item. Specifically, in an e-commerce scenario, the image to be identified can be an image of the item to be sold; in a traffic safety scenario, the image to be identified can be an image of the item the user wants to transport, and this image can be generated based on taking a picture or scanning.

[0142] Step S604: Process the image pixels of the image to be identified to obtain distribution data corresponding to at least one item.

[0143] In step S604, the image recognition system can process each pixel of the image to be recognized separately based on a machine learning model, or it can process each set of pixels in the image to be recognized separately. Each pixel or set of pixels contains one object or multiple objects belonging to the same category, thus obtaining distribution data corresponding to at least one item. Specifically, whether in e-commerce or traffic safety scenarios, the object in each pixel or set of pixels in the image to be recognized can be any of the following: electrical appliances, household goods, clothing, electronic devices, prohibited items, etc. Each pixel or set of pixels can contain one piece of clothing (i.e., one object) or multiple pieces of clothing (i.e., multiple objects belonging to the same category).

[0144] It should be noted that by obtaining the distribution data corresponding to at least one item, it is possible to accurately determine the item type of each item in the future.

[0145] Step S606: Identify the item type corresponding to at least one item based on the distribution data corresponding to at least one item.

[0146] In step S606, the distribution data includes at least the mean and variance. The mean represents the feature information of at least one item, and the variance represents the probability that at least one item is an anomalous item. In practical applications, the item type (first item type or second item type) can be determined based on the variance of the item in the image to be identified. Items corresponding to the first item type are not stored in a preset storage area, while items corresponding to the second item type are stored in the preset storage area. This achieves effective determination of item type.

[0147] Step S608: Determine whether at least one item is a non-compliant item based on the item type.

[0148] Optionally, for known items (including both non-violation and non-violation items), they can be added to a preset storage area. If the item type is a second item type, the image recognition system determines the specific item type (e.g., clothing, jewelry, violation items, etc.) based on the average value of items in the image to be recognized. For unknown or incomplete violation items, since they are not in the preset storage area, if the item type is determined to be a first item type, it indicates a high probability that it is a violation item, and it can be further pushed to regulatory personnel for confirmation. This allows the image recognition system to effectively identify both traditional and novel violation items, thereby improving the recognition efficiency of this application.

[0149] Furthermore, after identifying non-compliant items among items of the first object type, regulators can label the non-compliant items and add them as new samples to the preset dataset to optimize and train the machine learning model. This will enable the machine learning model to effectively classify each item in the image when the image to be identified is input into it next time, thereby improving the classification effect and enabling the relevant system to effectively identify non-compliant items.

[0150] Based on the scheme defined in steps S602 to S608 above, it can be understood that in this embodiment, the method of identifying the item type corresponding to at least one item based on the distribution data of each item, thereby determining whether each item is a non-compliant item, involves acquiring an image to be identified, processing the image pixels of the image to be identified to obtain the distribution data corresponding to at least one item, and then identifying the item type corresponding to at least one item based on the distribution data of at least one item, thereby determining whether at least one item is a non-compliant item based on the item type. The image to be identified contains at least one item.

[0151] It is noteworthy that in the above process, the distribution data corresponding to each item is acquired, the item type is determined based on the distribution data, and whether the item is a violation item is determined based on the item type. This achieves effective filtering of valid items in the image to be identified. This makes it easier for users to identify the items that need to be labeled from among the valid items, avoiding the labeling of invalid items in the image by the relevant system, thereby reducing the cost of item labeling.

[0152] Therefore, the solution provided in this application achieves the goal of identifying the item type corresponding to at least one item based on the distribution data of each item, thereby determining whether each item is a non-compliant item. This achieves the technical effect of reducing the cost of item labeling and solves the technical problem of high item labeling cost caused by the inability to effectively filter out valid items from multiple items in an image in the prior art.

[0153] Example 5

[0154] Embodiments of this application may provide a computer terminal, which may be any computer terminal device in a group of computer terminals. Optionally, in this embodiment, the aforementioned computer terminal may also be replaced by a mobile terminal or other terminal device.

[0155] Optionally, in this embodiment, the computer terminal may be located in at least one of a plurality of network devices in a computer network.

[0156] In this embodiment, the computer terminal described above can execute the program code of the following steps in the image data processing method: acquiring an image to be identified, wherein the image to be identified contains at least one region, and each region contains at least one object; processing the image data of the image to be identified to obtain distribution data corresponding to at least one region; determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, wherein the objects in the region corresponding to the first region type are not stored in a preset storage area.

[0157] Optionally, Figure 10 This is a structural block diagram of an optional computer terminal according to an embodiment of this application. For example... Figure 10 As shown, the computer terminal 10 may include one or more (only one is shown in the figure) processors 802, memory 804, and memory controller.

[0158] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the image data processing method and apparatus in this embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned image data processing method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0159] The processor can invoke information and application programs stored in the memory through the transmission device to perform the following steps: acquiring an image to be identified, wherein the image to be identified contains at least one region, and each region contains at least one object; processing the image data of the image to be identified to obtain distribution data corresponding to at least one region; determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, wherein the objects in the region corresponding to the first region type are not stored in a preset storage area.

[0160] Optionally, the processor may also execute program code for the following steps: calculating the image data of the image to be recognized based on a preset machine learning model to obtain the mean and variance values ​​corresponding to at least one region, wherein the distribution data includes at least the mean and variance values, the mean representing the feature information of at least one region, and the magnitude of the variance value representing the probability that at least one region is a first region type.

[0161] Optionally, the processor may also execute program code that performs the following steps: when there is a first target region with a variance value greater than a preset threshold in at least one region, the region type of the first target region is identified as a first region type; when there is a second target region with a variance value less than or equal to a preset threshold in at least one region, the region type of the second target region is identified as a second region type, wherein the objects in the region corresponding to the second region type are stored in a preset storage region.

[0162] Optionally, the processor may also execute program code that performs the following steps: after identifying the region type of the first target region as the first region type, obtain the first mean value corresponding to the first target region and the second mean value of the region corresponding to the second region type; compare the first mean value and the second mean value, and determine the target region with the highest similarity to the first target region from the region data stored in the preset storage area; determine the region type corresponding to the target region as the initial annotation type corresponding to the first target region.

[0163] Optionally, the processor may also execute program code that performs the following steps: after determining that the region type corresponding to the target region is the initial label type corresponding to the first target region, it responds to the update instruction, labels the initial label type of the first target region, and obtains the labeled region; and updates the data in the preset dataset based on the labeled region, wherein the preset dataset is used to optimize and train the machine learning model.

[0164] Optionally, the processor may also execute program code that performs the following steps: after identifying the region type of the first target region as the first region type, in response to a deletion instruction, deletes the region corresponding to the first region type from at least one region.

[0165] Optionally, the processor may also execute program code that performs the following steps: before determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, obtaining the first distribution information of the region corresponding to the first region type and the second distribution information of the region corresponding to the second region type from a preset historical record; and determining a preset threshold based on the first distribution information and the second distribution information.

[0166] Optionally, the processor may also execute program code for the following steps: after calculating the mean and variance values ​​of at least one region based on the image data of the image to be recognized using a preset machine learning model, resampling the variance values ​​of at least one region to obtain a scaling factor; classifying and calculating the mean values ​​of at least one region to obtain a score value for at least one region; calculating the ratio between the scaling factor and the score value to obtain a target score value; calculating the loss function value corresponding to the machine learning model based on the target score value; and optimizing the network parameters of the machine learning model when the loss function value meets preset conditions.

[0167] Optionally, the processor may also execute program code for the following steps: reading an image to be recognized, wherein the image to be recognized contains at least one region, and each region contains at least one object; responding to an image recognition instruction, displaying distribution data corresponding to at least one region and recognition results corresponding to at least one region, wherein the distribution data corresponding to at least one region is obtained by processing the image data of the image to be recognized, and the recognition results are obtained by recognizing at least one region based on the distribution data corresponding to at least one region, and the recognition results indicate whether the region type of at least one region is a first region type, and the objects in the region corresponding to the first region type are not stored in a preset storage area.

[0168] Optionally, the processor may also execute program code for the following steps: the cloud server receives an image to be identified from a terminal device, containing at least one region, wherein at least one object exists in each region; the cloud server processes the image data of the image to be identified to obtain distribution data corresponding to at least one region; the cloud server determines whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, and obtains a recognition result, wherein the object in the region corresponding to the first region type is not stored in a preset storage area; the cloud server returns the recognition result to the terminal device.

[0169] Optionally, the processor may also execute program code that performs the following steps: after the cloud server returns the recognition results to the terminal device, the cloud server obtains the annotation results of the terminal device for the first region type of region in at least one region, and updates the preset dataset based on the annotated first region type of region to optimize the training of the machine learning model.

[0170] Optionally, the processor may also execute program code for the following steps: acquiring a remote sensing image to be identified, wherein the remote sensing image contains at least one remote sensing object; processing the image pixels of the remote sensing image to obtain distribution data corresponding to at least one remote sensing object; determining whether the object type of at least one remote sensing object is a first object type based on the distribution data corresponding to at least one remote sensing object, wherein the object corresponding to the first object type is not stored in a preset storage area.

[0171] Optionally, the processor may also execute program code that includes at least one of the following steps: at least one remote sensing object includes at least one of the following: a remote sensing ground object, a building object, or a remote sensing water object.

[0172] Optionally, the processor may also execute program code for the following steps: acquiring an image to be identified, wherein the image to be identified contains at least one item; processing the image pixels of the image to be identified to obtain distribution data corresponding to at least one item; identifying the item type corresponding to at least one item based on the distribution data corresponding to at least one item; and determining whether at least one item is a non-compliant item based on the item type.

[0173] Those skilled in the art will understand that Figure 10 The structure shown is for illustrative purposes only. The computer terminal can also be a smartphone (such as an Android phone, an iOS phone, etc.), a tablet computer, a mobile internet device (MID), a PAD, and other terminal devices. Figure 10 This does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include components that are more advanced than those described above. Figure 10 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 10 The different configurations shown.

[0174] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0175] Example 6

[0176] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the image data processing method provided in Embodiment 1.

[0177] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0178] Optionally, in this embodiment, the storage medium is configured to store program code for performing the following steps: acquiring an image to be identified, wherein the image to be identified contains at least one region, and each region contains at least one object; processing the image data of the image to be identified to obtain distribution data corresponding to at least one region; determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, wherein the objects in the region corresponding to the first region type are not stored in a preset storage area.

[0179] Optionally, the storage medium may also store program code for performing the following steps: calculating the image data of the image to be recognized based on a preset machine learning model to obtain the mean and variance values ​​corresponding to at least one region, wherein the distribution data includes at least the mean and variance values, the mean representing the feature information of at least one region, and the magnitude of the variance representing the probability that at least one region is a first region type.

[0180] Optionally, the storage medium may also store program code for performing the following steps: when there is a first target region with a variance value greater than a preset threshold in at least one region, the region type of the first target region is identified as a first region type; when there is a second target region with a variance value less than or equal to a preset threshold in at least one region, the region type of the second target region is identified as a second region type, wherein objects in the region corresponding to the second region type are stored in a preset storage region.

[0181] Optionally, the storage medium may also store program code for performing the following steps: after identifying the region type of the first target region as the first region type, obtaining the first mean value corresponding to the first target region and the second mean value of the region corresponding to the second region type; comparing the first mean value and the second mean value, determining the target region with the highest similarity to the first target region from the region data stored in the preset storage area; determining the region type corresponding to the target region as the initial annotation type corresponding to the first target region.

[0182] Optionally, the storage medium may also store program code for performing the following steps: after determining that the region type corresponding to the target region is the initial label type corresponding to the first target region, responding to the update instruction, labeling the initial label type of the first target region to obtain the labeled region; updating the data in the preset dataset based on the labeled region, wherein the preset dataset is used to optimize and train the machine learning model.

[0183] Optionally, the storage medium may also store program code for performing the following steps: after identifying the region type of the first target region as a first region type, in response to a deletion instruction, deleting the region corresponding to the first region type from at least one region.

[0184] Optionally, the storage medium may also store program code for performing the following steps: before determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, obtaining first distribution information of the region corresponding to the first region type and second distribution information of the region corresponding to the second region type from a preset historical record; and determining a preset threshold based on the first distribution information and the second distribution information.

[0185] Optionally, the storage medium may also store program code for performing the following steps: after calculating the image data of the image to be recognized based on a preset machine learning model to obtain the mean and variance values ​​corresponding to at least one region, resampling the variance values ​​corresponding to at least one region to obtain a scaling factor; classifying and calculating the mean values ​​corresponding to at least one region to obtain a score value corresponding to at least one region; calculating the ratio between the scaling factor and the score value to obtain a target score value; calculating the loss function value corresponding to the machine learning model based on the target score value; and optimizing the network parameters of the machine learning model when the loss function value meets preset conditions.

[0186] Optionally, the storage medium may also store program code for performing the following steps: reading an image to be recognized, wherein the image to be recognized contains at least one region, and each region contains at least one object; responding to an image recognition instruction, displaying distribution data corresponding to at least one region and recognition results corresponding to at least one region, wherein the distribution data corresponding to at least one region is obtained by processing the image data of the image to be recognized, and the recognition results are obtained by recognizing at least one region based on the distribution data corresponding to at least one region, and the recognition results characterize whether the region type of at least one region is a first region type, and the objects in the region corresponding to the first region type are not stored in a preset storage area.

[0187] Optionally, the storage medium may also store program code for performing the following steps: the cloud server receives an image to be identified from a terminal device, containing at least one region, wherein at least one object exists in each region; the cloud server processes the image data of the image to be identified to obtain distribution data corresponding to at least one region; the cloud server determines whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, and obtains a recognition result, wherein the object in the region corresponding to the first region type is not stored in a preset storage area; the cloud server returns the recognition result to the terminal device.

[0188] Optionally, the storage medium may also store program code for performing the following steps: after the cloud server returns the recognition results to the terminal device, the cloud server obtains the annotation results of the terminal device for the first region type of region in at least one region, and updates the preset dataset based on the annotated first region type of region to optimize the training of the machine learning model.

[0189] Optionally, the storage medium may also store program code for performing the following steps: acquiring a remote sensing image to be identified, wherein the remote sensing image contains at least one remote sensing object; processing the image pixels of the remote sensing image to obtain distribution data corresponding to at least one remote sensing object; determining whether the object type of at least one remote sensing object is a first object type based on the distribution data corresponding to at least one remote sensing object, wherein the object corresponding to the first object type is not stored in a preset storage area.

[0190] Optionally, the storage medium may also store program code for performing the following steps: at least one remote sensing object includes at least one of the following: remote sensing ground object, building object, remote sensing water body object.

[0191] Optionally, the storage medium may also store program code for performing the following steps: acquiring an image to be identified, wherein the image to be identified contains at least one item; processing the image pixels of the image to be identified to obtain distribution data corresponding to at least one item; identifying the item type corresponding to at least one item based on the distribution data corresponding to at least one item; and determining whether at least one item is a non-compliant item based on the item type.

[0192] Example 7

[0193] Embodiments of this application also provide a system for recognizing images. Optionally, the system includes: a processor; and a memory connected to the processor, for providing the processor with instructions to process the following processing steps: acquiring an image to be recognized, wherein the image to be recognized contains at least one region, and each region contains at least one object; processing image data of the image to be recognized to obtain distribution data corresponding to at least one region; determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region, wherein the objects in the region corresponding to the first region type are not stored in a preset storage area.

[0194] As can be seen from the above, in this application, the method of determining whether a region is a valid region based on the distribution data of each region in the image is adopted. This involves acquiring the image to be recognized, processing the image data of the image to be recognized, obtaining the distribution data corresponding to at least one region, and then determining whether the region type of at least one region is a first region type based on the distribution data corresponding to at least one region. Specifically, the image to be recognized contains at least one region, and each region contains at least one object. The objects in the regions corresponding to the first region type are not stored in a preset storage area.

[0195] It is noteworthy that in the above process, acquiring the distribution data corresponding to each region and determining the region type based on this data enables effective filtering of valid regions in the image. This facilitates users in identifying the regions that need to be labeled within the valid regions, avoiding the labeling of invalid regions in the image to be recognized by related systems, thereby reducing the cost of region labeling. Furthermore, in this application, determining the region type based on the distribution data corresponding to each region avoids statistical analysis of classification probabilities and manually setting thresholds, reducing the randomness caused by the statistical properties of the data. It also avoids using additional classifiers to classify the features of the original model, reducing additional computational overhead, thereby improving the accuracy of region type judgment and reducing computational costs.

[0196] Therefore, the solution provided in this application achieves the goal of determining whether a region is a valid region based on the distribution data of each region in the image, thereby reducing the technical effect of region annotation cost and solving the technical problem of high region annotation cost caused by the inability to effectively filter out valid regions from multiple regions in the image in the prior art.

[0197] It should be noted that the processor in this embodiment can execute the image data processing method provided in Embodiment 1. The relevant content has been described in Embodiment 1 and will not be repeated here.

[0198] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0199] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0200] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0201] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0202] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0203] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0204] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An image data processing method, characterized by, The method comprises: acquiring a to-be-identified image, wherein the to-be-identified image contains at least one region, and each region contains at least one object; processing image data of the to-be-identified image to obtain distribution data corresponding to the at least one region; determining whether the region type of the at least one region is a first region type according to the distribution data corresponding to the at least one region, wherein the object in the region corresponding to the first region type is not stored in a preset storage region; wherein the method further comprises: in response to the presence of at least one first target region of the first region type in the at least one region, labeling the first target region to obtain a labeled region; and training a machine learning model using the labeled region, wherein the machine learning model is used to process image data of the to-be-identified image.

2. The method of claim 1, wherein, The method further comprises: calculating the image data of the to-be-identified image based on a preset machine learning model to obtain a mean value and a variance value corresponding to the at least one region, wherein the distribution data at least includes the mean value and the variance value, the mean value represents the feature information of the at least one region, and the size of the variance value represents the probability that the at least one region is of the first region type.

3. The method of claim 2, wherein, The method further comprises: when the first target region with a variance value greater than a preset threshold exists in the at least one region, identifying the region type of the first target region as the first region type; when a second target region with a variance value less than or equal to the preset threshold exists in the at least one region, identifying the region type of the second target region as a second region type, wherein the object in the region corresponding to the second region type is stored in the preset storage region.

4. The method of claim 3, wherein, The method further comprises: acquiring a first mean value corresponding to the first target region and a second mean value corresponding to the region of the second region type; comparing the first mean value and the second mean value to determine a target region with the highest similarity to the first target region from the region data stored in the preset storage region; determining that the region type corresponding to the target region is an initial labeling type corresponding to the first target region; in response to an update instruction, labeling the initial labeling type of the first target region to obtain the labeled region.

5. The method of claim 4, wherein, The method further comprises: updating data in a preset data set based on the labeled region, wherein the preset data set is used to optimize the training of the machine learning model.

6. The method of claim 3, wherein, After identifying the region type of the first target region as the first region type, the method further comprises: in response to a deletion instruction, deleting the region corresponding to the first region type from the at least one region.

7. The method of claim 3, wherein, Before determining whether the region type of the at least one region is a first region type according to the distribution data corresponding to the at least one region, the method further comprises: obtaining first distribution information of a region corresponding to the first region type and second distribution information of a region corresponding to the second region type from preset historical records; determining the preset threshold based on the first distribution information and the second distribution information.

8. An image data processing method characterized by, Comprise: reading a to-be-recognized image, wherein the to-be-recognized image contains at least one region, and each region contains at least one object; in response to an image recognition instruction, displaying distribution data corresponding to the at least one region and a recognition result corresponding to the at least one region, wherein the distribution data corresponding to the at least one region is obtained by processing image data of the to-be-recognized image, the recognition result is obtained by recognizing the at least one region according to the distribution data corresponding to the at least one region, and the recognition result indicates whether the region type of the at least one region is a first region type, and objects in the region corresponding to the first region type are not stored in a preset storage region; wherein the method further comprises: in response to at least one first target region of the first region type existing in the at least one region, labeling the first target region to obtain a labeled region; and training a machine learning model using the labeled region, wherein the machine learning model is used to process image data of the to-be-recognized image.

9. An image data processing method characterized by, Comprise: A cloud server receives a to-be-recognized image containing at least one region sent by a terminal device, wherein each region contains at least one object; The cloud server processes image data of the to-be-recognized image to obtain distribution data corresponding to the at least one region; The cloud server determines whether the region type of the at least one region is a first region type according to the distribution data corresponding to the at least one region to obtain a recognition result, wherein objects in the region corresponding to the first region type are not stored in a preset storage region; The cloud server returns the recognition result to the terminal device; wherein the cloud server obtains a labeling result of the terminal device labeling a region of the first region type in the at least one region; and the cloud server trains a machine learning model based on the labeled region of the first region type, wherein the machine learning model is used to process image data of the to-be-recognized image.

10. The method of claim 9, wherein, The cloud server trains a machine learning model based on the labeled region of the first region type, comprising: The cloud server updates a preset data set based on the labeled region of the first region type to optimize the training of the machine learning model.

11. An image data processing method, characterized by, Comprise: obtaining a to-be-recognized remote sensing image, wherein the remote sensing image contains at least one remote sensing object; processing image pixels of the remote sensing image to obtain distribution data corresponding to the at least one remote sensing object; determining whether the object type of the at least one remote sensing object is a first object type according to the distribution data corresponding to the at least one remote sensing object, wherein objects corresponding to the first object type are not stored in a preset storage area; wherein the method further comprises: in response to at least one first target object of the first object type existing in the at least one remote sensing object, labeling the first target object to obtain a labeled object; and training a machine learning model using the labeled object, wherein the machine learning model is used to process the image pixels of the remote sensing image.

12. A storage medium, characterized by The storage medium includes a stored program, wherein the program controls a device in which the storage medium is located to perform the image data processing method of any one of claims 1 to 11 when the program is running.

13. A processor, comprising: The processor is configured to run a program, wherein the program performs the image data processing method of any one of claims 1 to 11 when the program is running.

14. A system for identifying an image, characterized by comprising: a processor; and a memory connected to the processor and configured to provide the processor with instructions for processing the following processing steps: obtaining an image to be identified, wherein the image to be identified includes at least one region, and each region includes at least one object; processing image data of the image to be identified to obtain distribution data corresponding to the at least one region; determining whether the region type of the at least one region is a first region type according to the distribution data corresponding to the at least one region, wherein objects in the region corresponding to the first region type are not stored in a preset storage area; wherein the memory is further configured to provide the processor with instructions for processing the following processing steps: in response to at least one first target region of the first region type existing in the at least one region, labeling the first target region to obtain a labeled region; and training a machine learning model using the labeled region, wherein the machine learning model is used to process the image data of the image to be identified.

Images