Three-dimensional target object identification method and device, electronic equipment and medium

Abstract

The invention provides a three-dimensional target object identification method, which can be used in the technical field of security check and comprises the following steps: acquiring three-dimensional target object data acquired from three-dimensional data acquisition or reconstruction equipment; processing the obtained target object data based on a processing model to obtain three-dimensional identification information of the target object; the recognition result of the target object is obtained based on the three-dimensional recognition information, the processing model is a model obtained by conducting fine tuning processing on a multi-task processing model trained based on source domain data, and the fine tuning processing comprises the steps that the source domain data and the target domain data are processed based on the multi-task processing model, and candidate region information is obtained; obtaining a plurality of domain migration sub-modules based on a multi-task structure in the multi-task processing model; obtaining updated candidate region information based on the plurality of domain migration sub-modules; and the multi-task processing model is finely adjusted based on the updated candidate region information. The invention further provides a three-dimensional target object recognition device, equipment, a medium and a program product.

Description

[0001] This application is a divisional application of Chinese invention patent application No. 202510577930.1 (application date: May 7, 2025; invention title: three-dimensional target object recognition method, device, electronic device, medium). Technical Field

[0002] This invention relates to the fields of artificial intelligence technology and security inspection technology, and more specifically to a method, apparatus, equipment, medium and program product for three-dimensional target object recognition. Background Technology

[0003] In 3D target recognition tasks, methods such as CT (Computed Tomography) imaging, LiDAR scanning, or volume reconstruction are used to acquire the 3D structural data of the target object, which is then trained using deep neural networks to achieve automatic recognition of 3D targets. Currently, most mainstream technologies are based on supervised learning frameworks, relying on large amounts of labeled 3D data for model training.

[0004] However, in practical applications, due to differences in equipment parameters, imaging methods, and environmental conditions, 3D data acquired under different acquisition conditions suffers from inconsistent quality, insufficient sample size, and significant differences in data distribution. Furthermore, the inherent complexity of 3D recognition tasks makes model fine-tuning and optimization even more difficult. Current common practice involves acquiring large amounts of labeled data for new scenes and training from scratch or making large-scale adjustments to existing models. However, in practical applications, it is often difficult to obtain the required amount of data and labeling quality in a timely manner. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method, apparatus, device, medium and program product for three-dimensional target object recognition.

[0006] A first aspect of the present invention provides a method for identifying a three-dimensional target object, comprising: acquiring target object data to be identified, wherein the target object data represents three-dimensional target object data acquired from a three-dimensional data acquisition or reconstruction device; inputting the target object data into a target multi-task processing model, processing the target object data based on the target multi-task processing model, and acquiring three-dimensional recognition information of the target object; and acquiring the recognition result of the target object based on the three-dimensional recognition information, wherein the target multi-task processing model is a model obtained by fine-tuning a multi-task processing model trained on source domain data, the fine-tuning process comprising: processing the source domain data and target domain data based on the multi-task processing model to acquire candidate region information; acquiring multiple domain transfer sub-modules based on the multi-task structure in the multi-task processing model; training multiple domain transfer sub-modules based at least on the source domain data, the target domain data, and the candidate region information to acquire updated candidate region information; and fine-tuning the multi-task processing model based at least on the updated candidate region information.

[0007] According to an embodiment of the present invention, obtaining multiple domain migration sub-modules based on the multi-task structure in the multi-task processing model specifically includes: decoupling the multi-task structure in the multi-task processing model to obtain multiple task sub-paths; constructing multiple sub-network structures based on the multiple task sub-paths; and embedding domain discriminators in the multiple sub-network structures to obtain multiple domain migration sub-modules.

[0008] According to an embodiment of the present invention, the plurality of said domain transfer submodules include a gradient inversion layer connected to the domain discriminator, the gradient inversion layer being used to invert the gradient direction of the domain discriminator during training.

[0009] According to an embodiment of the present invention, the target multi-task processing model includes a 3D feature extraction module, which extracts three-dimensional feature representations with geometric and semantic information from the source domain data and the target domain data based on a three-dimensional convolutional network or a sparse volume coding structure.

[0010] According to an embodiment of the present invention, the target object data includes three-dimensional image data of the target object obtained through security inspection equipment, and the recognition result is used to identify the category information and location information of the target object.

[0011] According to an embodiment of the present invention, the target multi-task processing model is at least used for target detection of the target object.

[0012] According to an embodiment of the present invention, the step of processing the source domain data and target domain data based on the multi-task processing model to obtain candidate region information specifically includes: inputting the source domain data and the target domain data into the multi-task processing model respectively to obtain corresponding preliminary detection results; and extracting at least the category prediction score and the three-dimensional bounding box parameters from the preliminary detection results as the candidate region information.

[0013] According to an embodiment of the present invention, the plurality of domain migration submodules include at least a classification migration submodule and a regression migration submodule, wherein the classification migration submodule is used to predict the category information of the target object, and the regression migration submodule is used to predict the location information of the target object.

[0014] According to an embodiment of the present invention, the input of the regression transfer submodule includes at least the category prediction result output by the classification transfer submodule.

[0015] According to an embodiment of the present invention, the candidate region information includes source domain positive sample candidate region information, source domain negative sample candidate region information, target domain positive sample candidate region information, and target domain negative sample candidate region information.

[0016] According to an embodiment of the present invention, the step of training multiple domain transfer submodules based at least on the source domain data, the target domain data, and the candidate region information specifically includes: constructing paired training samples composed of positive and negative samples corresponding to the source domain data and the target domain data, wherein the paired training samples are associated with the candidate region information; inputting the paired training samples into the corresponding domain transfer submodules to obtain target prediction results; and updating the candidate region information based on the target prediction results to generate the updated candidate region information.

[0017] A second aspect of the present invention provides a three-dimensional target object recognition device, comprising: a data acquisition module, configured to: acquire target object data to be recognized, wherein the target object data represents three-dimensional target object data acquired from a three-dimensional data acquisition or reconstruction device; a processing module, configured to: input the target object data into a target multi-task processing model, process the target object data based on the target multi-task processing model, and acquire three-dimensional recognition information of the target object, wherein the target multi-task processing model is a model obtained by fine-tuning a multi-task processing model trained on source domain data, the fine-tuning process comprising: processing the source domain data and target domain data based on the multi-task processing model to acquire candidate region information; acquiring multiple domain transfer sub-modules based on the multi-task structure in the multi-task processing model, training multiple domain transfer sub-modules based at least on the source domain data, the target domain data, and the candidate region information to acquire updated candidate region information; fine-tuning the multi-task processing model based at least on the updated candidate region information; and a recognition result acquisition module, configured to: acquire the recognition result of the target object based on the three-dimensional recognition information.

[0018] A third aspect of the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0019] A fourth aspect of the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0020] A fifth aspect of the present invention also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0021] According to embodiments of the present invention, through multi-task processing, multi-path training, and domain transfer mechanisms, the transfer submodule optimizes task branches, enabling the model to automatically adapt to different data distributions. This improves the applicability and robustness of the 3D recognition model under new environments, new devices, or new data sources, significantly reducing the burden of manual retraining. Simultaneously, by utilizing source domain data and candidate region information, 3D recognition capabilities can be transferred under low-label or unlabeled conditions, significantly reducing model deployment and transfer costs. Thus, while maintaining model structural consistency, more refined cross-task transfer optimization is achieved, improving the recognition accuracy of classification and regression tasks. Attached Figure Description

[0022] The above-described features, other objects, and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0023] Figure 1 The illustrations depict application scenarios of the three-dimensional target object recognition method, apparatus, device, medium, and program products according to embodiments of the present invention.

[0024] Figure 2 A flowchart illustrating a three-dimensional target object recognition method according to an embodiment of the present invention is shown schematically;

[0025] Figure 3 This diagram schematically illustrates a model architecture of a three-dimensional target object recognition system according to an embodiment of the present invention.

[0026] Figure 4A This diagram schematically illustrates the domain migration submodule structure of a classification task according to an embodiment of the present invention.

[0027] Figure 4B This diagram schematically illustrates the domain migration submodule structure of a regression task according to an embodiment of the present invention.

[0028] Figure 5 A schematic diagram illustrating the structure of a three-dimensional target object recognition device according to an embodiment of the present invention is shown; and

[0029] Figure 6 A block diagram of an electronic device suitable for implementing a three-dimensional target object recognition method according to an embodiment of the present invention is shown schematically. Detailed Implementation

[0030] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the invention. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0031] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0032] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0033] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0034] First, the technical terms used in this article are explained and clarified as follows.

[0035] The source domain refers to the data distribution environment corresponding to existing, fully labeled training data, which is used to train the base model or initialize model parameters. Source domain data typically comes from prior scenarios, standard equipment, or controlled experimental conditions, and has large-scale, high-quality labeled information, which can serve as the basis for model learning.

[0036] The target domain refers to the actual data scenario where the model needs to be applied, and its data distribution has different data characteristics than that of the source domain.

[0037] Adversarial networks are a type of deep learning method that involves a game-like structure between two subnetworks, typically consisting of a generator and a discriminator.

[0038] In current scenarios of 3D object analysis and processing, computed tomography (CT) equipment is widely used for the automated inspection and analysis of various objects to improve efficiency and intelligence. CT equipment, through X-ray scanning and image reconstruction technology, can generate tomographic images containing 3D physical information such as density and material composition, providing fundamental data support for various tasks. To achieve accurate analysis and processing of complex 3D structures, existing technologies typically employ deep learning methods, training multi-task processing models with large amounts of labeled data to extract features from CT images and complete tasks such as object localization, classification, segmentation, tracking, and quality assessment.

[0039] However, the applicant's research revealed that the image acquisition environment for the target object varies significantly in practical applications. Different models of equipment, variations in scanning parameters, differences in imaging algorithms, and the specific types of objects being detected in the application scenarios all lead to significant differences in data distribution, density characteristics, and noise levels in the acquired CT images. These differences in data distribution severely affect the adaptability of the original model to new equipment or new scenarios.

[0040] Typically, to meet new recognition requirements, it is necessary to collect a large amount of target data that conforms to the new distribution and perform fine annotation. However, in some fields, obtaining training samples faces multiple difficulties. On the one hand, some specific targets appear infrequently, resulting in a scarcity of relevant samples; on the other hand, some items have privacy or control restrictions, leading to high annotation difficulty, long processing time, and high cost, making it difficult to support the construction of large-scale annotated datasets. Therefore, target data not only has a small sample size but also suffers from class imbalance and an imbalanced ratio of positive to negative samples, making it difficult to support effective training based on deep neural networks.

[0041] The applicant also noted that existing multi-task processing models are often designed with complex task structures, simultaneously performing multiple tasks (such as object classification and location prediction, 3D segmentation, object tracking, quality detection, etc.). Compared to traditional single-task models, multi-task models have mechanisms for feature coupling and parameter sharing during training, which makes their adaptation requirements for different task branches in cross-domain applications higher, and the transfer process more complex. Most existing transfer learning methods are geared towards single tasks and are difficult to fully adapt to the coordinated adjustment of model parameters and feature spaces under multi-task structures, resulting in unstable training effects and low transfer efficiency.

[0042] Specifically, when a model trained on source domain data is directly applied to the target domain, it often faces the problem of "domain shift". That is, a model that performs well in the source domain may have a significantly lower recognition accuracy in the target domain. This not only affects the detection accuracy of the target object, but also restricts the widespread application of the model in different fields and tasks.

[0043] Based on this, embodiments of the present invention provide a three-dimensional target object recognition method, comprising: acquiring target object data to be recognized; inputting the target object data into a target multi-task processing model; processing the target object data based on the target multi-task processing model to obtain three-dimensional recognition information of the target object; and obtaining the recognition result of the target object based on the three-dimensional recognition information. The target multi-task processing model is a model obtained by fine-tuning a multi-task processing model trained on source domain data. The fine-tuning process includes: processing source domain data and target domain data based on the multi-task processing model to obtain candidate region information; obtaining multiple domain transfer sub-modules based on the multi-task structure in the multi-task processing model; training multiple domain transfer sub-modules based at least on source domain data, target domain data, and candidate region information to obtain updated candidate region information; and fine-tuning the multi-task processing model based at least on the updated candidate region information. The three-dimensional target object recognition method provided by the embodiments of the present invention, through multi-task processing, multi-path training, and domain transfer mechanisms, optimizes task branches in the transfer sub-modules, enabling the model to automatically adapt to different data distributions. This improves the applicability and robustness of the three-dimensional recognition model under new environments, new devices, or new data sources, and significantly reduces the burden of manual retraining. Meanwhile, by utilizing source domain data and candidate region information, it is possible to transfer 3D recognition capabilities under low-label or unlabeled conditions, significantly reducing model deployment and migration costs. This allows for more refined cross-task transfer optimization while maintaining model structure consistency, thereby improving the recognition accuracy of classification and regression tasks.

[0044] Figure 1 The illustration schematically depicts application scenarios of a three-dimensional target object recognition method, apparatus, device, medium, and program product according to embodiments of the present invention.

[0045] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0046] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc.

[0047] The first terminal device 101 and the second terminal device 102 can be devices deployed in different application scenarios, such as industrial inspection equipment, medical imaging equipment, and environmental monitoring equipment. For example, the first terminal device 101 and the second terminal device 102 may include baggage CT scanners, millimeter-wave body scanners, X-ray inspection machines, industrial CT scanners, medical imaging scanners, and environmental detection instruments. The first terminal device 101 and the second terminal device 102 can collect various data of the target being inspected, such as metadata including image information, material property information, three-dimensional structure information, inspection time, and device number. Through the network 104, the collected data can be uploaded to the server 105 in real time or periodically for subsequent data processing.

[0048] The third terminal device 103 can retrieve and display data related to detection records, identification results, or device status from the server 105 according to the user's query requirements. The third terminal device 103 can be various electronic terminal devices with display functions and support network browsing, including but not limited to central control room terminals, duty management terminals, desktop computers, etc. in centralized monitoring systems.

[0049] Server 105 may include a model processing module, a data storage module, and a task scheduling module for target object recognition. It is used to receive detection data uploaded from multiple terminal devices and process and schedule it uniformly. For example, server 105 can perform recognition model inference operations based on the received 3D image data, extract recognition information such as the target object's category, bounding box, and positional relationship, and return the processing results to the corresponding terminal or upload them to the monitoring system.

[0050] It should be noted that the three-dimensional target object recognition method provided in the embodiments of the present invention can generally be executed by server 105. Correspondingly, the three-dimensional target object recognition device provided in the embodiments of the present invention can generally be located in server 105. The three-dimensional target object recognition method provided in the embodiments of the present invention can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the three-dimensional target object recognition device provided in the embodiments of the present invention can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0051] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0052] The following will be based on Figure 1The described scene, through Figures 2 to 4B The three-dimensional target object recognition method of the present invention embodiments is described in detail.

[0053] Figure 2 A flowchart illustrating a three-dimensional target object recognition method according to an embodiment of the present invention is shown.

[0054] like Figure 2 As shown, the three-dimensional target object recognition method of this embodiment includes operations S210 to S230, and the three-dimensional target object recognition method can be executed by server 105.

[0055] In operation S210, data on the target object to be identified is acquired.

[0056] In embodiments of the present invention, the target object data can originate from various types of 3D data acquisition or reconstruction devices, such as 3D LiDAR, structured light scanners, depth cameras, medical imaging devices (such as magnetic resonance imaging (MRI) devices, CT devices, etc.), industrial non-destructive testing devices, computational photography devices, or virtual modeling systems. Depending on the specific application scenario, the representation of the target object data can also be diverse, such as point cloud data, regular or irregular voxel mesh data, 3D image data (such as slice sequences), volume density map representation, or sparse 3D feature representation extracted from the original data, etc.

[0057] In operation S220, target object data is input into the target multi-task processing model. The model processes the target object data to obtain its 3D recognition information. This model possesses multi-task capabilities, such as simultaneously performing target object classification and 3D position estimation. During processing, the model first extracts features from the target object data, for example, by combining classification and regression branches to output the target object's category prediction results and 3D bounding box parameters, forming an intermediate recognition layer representation containing target information.

[0058] In operation S230, the recognition result of the target object is obtained based on the three-dimensional recognition information. For example, the three-dimensional recognition information may specifically include, but is not limited to, the target object's category information, location information, the three-dimensional position and scale information of the target object's bounding box, and the target object's confidence score. Based on the three-dimensional recognition information, embodiments of the present invention can output the corresponding recognition result for further business analysis, decision control, automated operation, or human-computer interaction presentation.

[0059] In an embodiment of the present invention, the target multi-task processing model can be a cross-domain adaptation model obtained through fine-tuning, and its initial structure can be based on a multi-task processing model trained on source domain data.

[0060] In embodiments of the present invention, the multi-task processing model may have a multi-task structure, such as a classification branch for object classification and a regression branch for 3D localization, and has the ability to perform a comprehensive detection task on the input 3D data.

[0061] The target object data originates from real objects in the scene to be detected. These objects may exhibit diversity, occlusion, deformation, and even varying noise levels and spatial resolutions, resulting in significant semantic and geometric differences between the data and the source domain data used during model training. Therefore, directly using the original trained model for recognition on this target object data often leads to "domain shift" problems such as decreased accuracy and insufficient generalization performance.

[0062] According to embodiments of the present invention, in order to enable the model to adapt to the distribution characteristics of the target object data and improve the recognition accuracy in the target domain, the multi-task processing model can be fine-tuned. Fine-tuning may include multiple consecutive training phases to gradually guide the model from the source domain to the target domain.

[0063] Specifically, the original multi-task processing model is first used to perform forward inference processing on the source domain data and the target domain data to obtain candidate region information corresponding to the two data domains, which is used to guide feature alignment and structure adaptation in the subsequent training process.

[0064] For example, candidate region information may include the output results of each task branch, such as the predicted category and score of the classification branch, the 3D bounding box information of the regression branch, the region feature vector, the location information of the candidate region, etc., which will be used in subsequent transfer training.

[0065] In some embodiments, candidate region information may include source domain positive sample candidate region information, source domain negative sample candidate region information, target domain positive sample candidate region information, and target domain negative sample candidate region information. Specifically, the source domain positive sample candidate region information and source domain negative sample candidate region information can be obtained through manual annotation or automatic screening by existing models, and have high label accuracy; the target domain positive sample candidate region information and target domain negative sample candidate region information can be generated based on existing weakly labeled or unlabeled data in the target domain through preliminary inference or pseudo-label mechanisms. This candidate region information will serve as an important component of the supervision signal during subsequent training to guide the model's transfer adaptation between different domains.

[0066] After obtaining candidate region information, the overall network can be decoupled into multiple task sub-paths based on the multi-task structure of the multi-task processing model. Each task sub-path corresponds to a domain transfer sub-module. The domain transfer sub-module structure can include components such as a feature extraction sub-network, a domain discriminator, and an adversarial training mechanism, supporting decoupled optimization of classification domain transfer and regression domain transfer. By feeding the candidate region information in the source and target domains for each task into the corresponding domain transfer sub-modules and combining it with the design of an adversarial loss function, feature space alignment between the source and target domains at the task level can be achieved.

[0067] In some embodiments, the domain transfer submodule may also include a gradient inversion layer, which is used to invert the gradient direction of the domain discriminator during training to achieve effective alignment of the feature spaces of the source and target domains while maintaining the discriminative ability of each task.

[0068] In embodiments of the present invention, data samples from the source and target domains can be constructed as paired training samples, and combined with their respective candidate region information, input into each domain transfer submodule for training. For example, multiple domain transfer submodules can be jointly trained based on source domain data, target domain data, and candidate region information. During training, the paired training samples are composed of candidate region information for positive samples from the source domain, candidate region information for negative samples from the source domain, candidate region information for positive samples from the target domain, and candidate region information for negative samples from the target domain, respectively. Each pair of paired training samples is associated with corresponding candidate region information, used to indicate the key spatial regions or target parts that the model focuses on.

[0069] During the training phase, the constructed paired training samples are input into the corresponding domain transfer submodules. Each domain transfer submodule is responsible for extracting features at different levels and completing the target prediction task, while capturing the semantic differences between the source and target domains. As the model receives paired training samples, it utilizes candidate region information to guide attention mechanisms or region feature fusion strategies, thereby improving the model's generalization ability and robustness in the target domain.

[0070] After training the domain transfer submodules, the initial candidate region information can be updated based on the output of each submodule to obtain updated candidate region information. Updating the candidate region information more accurately reflects the structural characteristics of the target domain data, thus providing effective support for subsequent model fine-tuning.

[0071] In embodiments of the present invention, the original multi-task processing model can be fine-tuned and optimized by utilizing updated candidate region information and combining source domain data and target domain data. This fine-tuning process can employ a joint optimization strategy to achieve a balance between maintaining source domain recognition performance and improving target domain recognition performance. The target multi-task processing model obtained through the above fine-tuning process possesses stronger cross-domain adaptability and can achieve stable and reliable recognition results in complex 3D recognition tasks with biased feature distributions.

[0072] The 3D target recognition method provided by the embodiments of the present invention can be used not only for object recognition and target detection, but also extended to tasks such as 3D scene segmentation, pose estimation, point cloud instance segmentation, and multi-view reconstruction. It is also adaptable to various industry scenarios, such as industrial defect detection, medical image recognition, autonomous driving perception systems, virtual reality modeling, warehouse robot operation, and structural scanning and recognition in complex spaces.

[0073] It should be noted that the application fields of the three-dimensional target object recognition method provided in the embodiments of the present invention are not limited to the above scenarios, and can be flexibly expanded and adapted according to actual needs to meet the needs of technological development and market demand.

[0074] Figure 3 The schematic diagram illustrates a model architecture diagram of a three-dimensional target object recognition system according to an embodiment of the present invention.

[0075] like Figure 3 As shown, the three-dimensional target object recognition system according to an embodiment of the present invention may include multiple functional modules for effectively transferring the source domain model to the target domain. For example, it may include: a 3D feature extraction module, multiple task domain transfer sub-modules, a candidate region information module, and a 3D multi-task prediction component.

[0076] It should be noted that the various functional modules in the 3D target recognition system constitute a specific architectural expression of the "target multi-task processing model" in the implementation of this invention. The target multi-task processing model is obtained by pre-training based on source domain data and fine-tuning and optimizing it by combining target domain data and candidate region information.

[0077] Reference Figure 3 Image data from both the source and target domains are simultaneously input into the 3D feature extraction module. This module extracts 3D feature representations with geometric and semantic information based on a 3D convolutional network or a sparse volumetric coding structure. This module performs shared feature encoding on both types of data to ensure the comparability of features across different domains.

[0078] After feature extraction, the model can generate candidate region information for subsequent processing. For example, candidate region information generated from source domain data can be represented as Prop. sThe candidate region information generated from the target domain data can be represented as Prop t Task1, Task2...Taskn can represent Task 1, Task 2...Taskn respectively. Each candidate region can be further subdivided into structural information components specific to different tasks, for example, Prop... s_task1 Prop represents candidate region features in the source domain that are relevant to Task 1; t_task2 This represents the candidate region features in the target domain that are relevant to Task 2.

[0079] Reference Figure 3 The 3D object recognition system, based on the multi-task structure of the multi-task processing model, decouples the overall model into multiple task sub-paths (such as classification tasks, regression tasks, semantic segmentation tasks, etc.), and constructs a corresponding domain transfer sub-module for each sub-path (Task1 domain transfer sub-module, Task2 domain transfer sub-module...Taskn domain transfer sub-module). The design of each domain transfer sub-module may include: a sub-network for feature processing of the current task; a domain discriminator for discriminating the data domain (source domain / target domain); and an optional gradient inversion layer to achieve gradient direction control in adversarial training.

[0080] It should be noted that, Figure 3 The target domain candidate region information Prop is schematically drawn. t The input is fed into each task domain transfer submodule. In fact, in some other embodiments, each domain transfer submodule may also simultaneously receive paired training samples from the source and target domains to achieve task-level feature alignment.

[0081] Reference Figure 3 Each domain transfer submodule gradually reduces the distributional differences in corresponding task features between the source and target domains through adversarial training, thereby improving the model's expressive ability in target domain task execution. After training, the optimization results of each task path will be used to update the target domain candidate region information Prop. t This results in a more structured input that is more adaptable to task distribution.

[0082] The updated target domain candidate region information will be input into the 3D multi-task prediction component, which will fuse the recognition results of multiple task paths and output a comprehensive recognition result of the target object. The output result may include task-related information such as target category, 3D bounding box position and size, orientation angle, and confidence score.

[0083] The preferred embodiments of the present invention will be described in detail below using the three-dimensional detection task of security inspection CT data targets as an example.

[0084] In this embodiment, the target object data includes three-dimensional tomographic image data acquired by security CT equipment. The images contain feature information such as the spatial shape and material density of the object. By processing these three-dimensional images, the system can automatically identify the object's category and location, assisting in tasks such as dangerous goods identification, baggage sorting, and security control.

[0085] In embodiments of the present invention, the data in the source domain and the target domain can be processed separately based on a trained multi-task processing model to obtain preliminary detection results. The source domain data can be historically acquired and labeled security inspection CT image data. For example, the source domain data can be image data acquired using a slip-ring CT scanner. This type of scanner has a stable rotating scanning structure and a mature accumulation of labeled samples, and has been used for training conventional 3D target object recognition models for a long time, exhibiting characteristics such as stable distribution, high image quality, and a sufficient number of training samples. The target domain data can be new data from different security inspection sites, different equipment, or different setting conditions, and may be unlabeled or have only a few labels.

[0086] In this embodiment, the candidate region information can be derived by extracting the corresponding category prediction score, prediction label, and 3D bounding box parameters (such as the target center point position, size, and orientation angle) for each candidate target region. This candidate region information will be used in subsequent transfer submodule training and model optimization stages. Multiple domain transfer submodules may include a classification transfer submodule and a regression transfer submodule. The classification transfer submodule is used to perform transfer learning on the category information of the candidate regions to reduce the difference in category distribution between the source and target domains; the regression transfer submodule performs transfer training on the position parameters to improve the 3D localization accuracy in the target domain.

[0087] Each domain transfer submodule can include a feature extraction branch, a task discriminator, and a domain discriminator module in its network structure, and can optionally be equipped with gradient inversion layers to achieve adversarial optimization training objectives. The training of each domain transfer submodule can be executed in parallel or alternately, and the overall synergistic effect can be improved through joint optimization strategies.

[0088] It should be noted that the specific network structure of the domain transfer submodule can be flexibly configured according to different task types or target domain data characteristics. For example, for some task subpaths that are not sensitive to domain differences or already have good cross-domain generalization capabilities, only feature extraction branches and task discriminators can be set to simplify the network structure and reduce model complexity and computational resource overhead. In addition, domain discriminators can be dynamically enabled or disabled for some submodules according to the importance of different tasks or training progress. For example, in the early stages of multi-task training, task discriminators can be trained first to quickly converge basic task capabilities; then, domain discriminators can be gradually introduced to improve cross-domain generalization performance through adversarial mechanisms, thereby achieving a flexible balance between phased optimization and resource scheduling.

[0089] In some embodiments, the regression transfer submodule can further incorporate the output of the classification submodule as input to enhance its prediction robustness. Specifically, the category prediction results output by the classification task, such as category labels and confidence scores, can be injected into the regression path as high semantic information to guide the bounding box prediction process to focus more on the real object target in the candidate region, thereby reducing the error propagation caused by background noise or false targets.

[0090] For example, when the confidence level of the candidate region is high and the predicted category is clear, the regression module can assign a higher weight to the region to improve the accuracy of the bounding box localization; when the uncertainty of the category prediction is high, the regression module can appropriately suppress the output amplitude of the location regression to reduce the risk of false detection.

[0091] In some embodiments, positive and negative sample regions corresponding to the source and target domain data can be divided based on the candidate region information corresponding to the source and target domain data, and paired training samples can be constructed. The construction of paired training samples not only helps the domain discriminator learn in adversarial training, but also makes full use of the source domain structural information under the condition of scarce target domain samples, thereby improving the sample efficiency of model training.

[0092] Paired training samples can be input into each domain transfer submodule for training. During training, the domain discriminator can be used to perform adversarial training on samples from the source and target domains to optimize the module's discriminative and feature alignment capabilities, ultimately obtaining more suitable feature representations for the target domain under each task sub-path.

[0093] In some preferred embodiments, the stability and convergence efficiency of adversarial training can be further enhanced by introducing domain-mixed discriminative loss, multi-scale attention mechanisms, or inter-layer consistency constraints.

[0094] After the domain transfer submodule is trained, the original candidate region information can be updated based on the classification prediction results and bounding box prediction results. For example, for some low-confidence candidate regions in the target domain, their class labels or bounding box positions can be re-evaluated by combining the output of the transfer submodule, thereby improving their representation accuracy under the target domain conditions.

[0095] According to an embodiment of the present invention, the update mechanism is iterative and can further refine the candidate regions through a phased training and feedback mechanism, providing higher quality training signals for fine-tuning the backbone model.

[0096] The updated candidate region information can be used to fine-tune the multi-task processing model, enhancing its adaptability to target domain data while maintaining source domain recognition accuracy. The optimized target multi-task processing model can be directly deployed on different security CT systems to achieve automated 3D recognition of luggage or items in complex environments.

[0097] Figure 4A The diagram illustrates the structure of the domain migration submodule for a classification task according to an embodiment of the present invention.

[0098] like Figure 4A As shown, in this structure, the input includes positive sample candidate region information P from the source domain. s,prop and negative sample candidate region information N s,prop And the positive sample candidate region information P of the target domain t,prop and negative sample candidate region information N t,prop For example, these samples can be constructed based on the predicted scores of candidate regions in classification tasks. Positive samples are typically high-confidence target candidate regions, while negative samples are background regions or low-confidence regions.

[0099] For example, in the actual construction process, the division of positive and negative samples can be combined with heuristic threshold settings. For example, samples with a classification confidence greater than a certain threshold are considered positive samples, and samples with a confidence lower than another threshold are considered negative samples. Samples in the middle region can be selectively ignored or re-evaluated in subsequent stages.

[0100] Reference Figure 4A Candidate region samples from both the source and target domains can be input into the classification domain transfer feature extractor to extract intermediate feature representations F(s) and F(t), corresponding to the feature outputs of the source and target domains, respectively. F(s) and F(t) are then input into the domain discriminator. Through an adversarial training mechanism, the feature extractor learns domain-independent discriminative features, thereby achieving feature alignment between the source and target domains in the classification task. Based on this, F(s) can be further input into the classifier module, combined with the true label information from the source domain, to train the predictive ability of the target category and generate the classification prediction result P. t,prop,cls .

[0101] In other embodiments, candidate region samples of the target domain can also be input to a target domain candidate region information classification and prediction module. In this module, pre-classification modeling or pseudo-label prediction of candidate regions is performed, and the generated region classification prediction information is used as a guiding signal to further input to the classifier to generate a classification prediction result P. t,prop,cls .

[0102] In this embodiment, the domain discriminator can be based on a binary classification network structure, the goal of which is to distinguish whether the input features come from the source domain or the target domain; while the classification domain transfer feature extractor can maximize the error of the domain discriminator through a gradient inversion layer, so that its output features are insensitive to domain information.

[0103] Figure 4B The diagram illustrates the domain migration submodule structure of a regression task according to an embodiment of the present invention.

[0104] like Figure 4B As shown, in this structure, the input is the positive sample candidate region information P of the source domain. s,prop Information P of positive candidate regions of the target domain t,prop Unlike classification tasks, regression tasks can be trained and aligned only on positive samples, thus negative sample information is not processed in this path. This is because the core objective of regression tasks is to accurately predict continuous variables such as the position and size of the target object's 3D bounding box. This type of supervision is only effective for the real target region; the background or invalid regions do not have effective regression targets.

[0105] Reference Figure 4B Positive sample candidate region information P in the source domain s,prop Information P of positive candidate regions of the target domain t,prop The inputs are fed into the regression domain transfer feature extractor, which generates intermediate features F(s) and F(t) for the source and target domains, respectively. During training, F(s) and F(t) are also fed into the regression discriminator for adversarial training to align the regression feature space between the source and target domains, thereby improving the accuracy of the target domain bounding box prediction. Based on this, F(s) can be further input into the classifier module, which combines the true label information from the source domain to output the regression prediction result P. t,prop,reg .

[0106] In some embodiments, the target domain candidate region information P t, prop The data can also be input into the target domain candidate region information regression prediction module to model or predict the geometric features of candidate regions, thereby enhancing the representation capability of region localization. The output of this module is then input into the regressor module, which, together with the intermediate features F(s) from the feature extractor, trains the regression capability of the 3D bounding box and finally outputs the regression prediction result P. t,prop,reg , representing the regression prediction result of the candidate region of the target domain.

[0107] The structure of the regression discriminator is similar to that of the discriminator in the classification path, but its input features may focus more on spatial continuity and structural expressiveness, such as introducing local geometric encoding, voxel statistical features, or scale-sensitive representations. The purpose of adversarial training is to guide the regression features to have good cross-domain consistency, thereby improving the localization robustness of the target domain when precise annotations are lacking.

[0108] Furthermore, in some preferred embodiments, the feature extractor in the regression task can further integrate the output information of the classification submodule, such as the category prediction results or category probability distribution, to assist in guiding the selection of the scale, shape, or target perception range of the bounding box, forming a weakly coupled structure between tasks and further improving the overall detection performance.

[0109] Based on the above-described three-dimensional target object recognition method, embodiments of the present invention also provide a three-dimensional target object recognition device. The following will be combined with... Figure 5 The device is described in detail.

[0110] Figure 5 A schematic diagram of a three-dimensional target object recognition device according to an embodiment of the present invention is shown.

[0111] like Figure 5 As shown, the three-dimensional target object recognition device 500 of this embodiment includes a data acquisition module 510, a processing module 520, and a recognition result acquisition module 530.

[0112] The data acquisition module 510 can be used to acquire data of the target object to be identified. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0113] The processing module 520 can be used to input target object data into a target multi-task processing model, process the target object data based on the target multi-task processing model, and obtain the three-dimensional recognition information of the target object. The target multi-task processing model includes a model obtained by fine-tuning a multi-task processing model trained on source domain data. The fine-tuning process includes: processing source domain data and target domain data based on the multi-task processing model to obtain candidate region information; obtaining multiple domain transfer sub-modules based on the multi-task structure in the multi-task processing model; training multiple domain transfer sub-modules based at least on source domain data, target domain data, and candidate region information to obtain updated candidate region information; and fine-tuning the multi-task processing model based at least on the updated candidate region information. In one embodiment, the processing module 520 can be used to execute the operation S220 described above, which will not be repeated here.

[0114] The recognition result acquisition module 530 can be used to acquire the recognition result of the target object based on the three-dimensional recognition information. In one embodiment, the recognition result acquisition module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0115] According to an embodiment of the present invention, the processing module 520 can also be used to decouple the multi-task structure in the multi-task processing model to obtain multiple task sub-paths; construct multiple sub-network structures based on the multiple task sub-paths; and embed domain discriminators in the multiple sub-network structures to obtain multiple domain migration sub-modules.

[0116] According to an embodiment of the present invention, a plurality of domain transfer submodules include a gradient inversion layer connected to a domain discriminator, the gradient inversion layer being used to invert the gradient direction of the domain discriminator during training.

[0117] According to an embodiment of the present invention, the processing module 520 can also be used to input source domain data and target domain data into a multi-task processing model respectively to obtain corresponding preliminary detection results; and to extract at least the category prediction score and three-dimensional bounding box parameters from the preliminary detection results as candidate region information.

[0118] According to an embodiment of the present invention, the multiple domain transfer submodules include at least a classification transfer submodule and a regression transfer submodule. The classification transfer submodule is used to predict the category information of the candidate region, and the regression transfer submodule is used to predict the location information of the candidate region. The input of the regression transfer submodule includes at least the category prediction result output by the classification transfer submodule.

[0119] According to an embodiment of the present invention, the candidate region information includes source domain positive sample candidate region information, source domain negative sample candidate region information, target domain positive sample candidate region information, and target domain negative sample candidate region information.

[0120] According to an embodiment of the present invention, the processing module 520 can also be used to construct paired training samples consisting of positive and negative samples corresponding to source domain data and target domain data, and the paired training samples are associated with candidate region information; input the paired training samples into the corresponding domain transfer submodule to obtain category prediction results and bounding box prediction results; and update the candidate region information based on the category prediction results and bounding box prediction results to generate updated candidate region information.

[0121] According to embodiments of the present invention, any plurality of modules among the data acquisition module 510, processing module 520, and recognition result acquisition module 530 may be combined into one module, or any one of these modules may be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules may be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of the present invention, at least one of the data acquisition module 510, processing module 520, and recognition result acquisition module 530 may be at least partially implemented as a hardware circuit, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuit, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the data acquisition module 510, processing module 520, and recognition result acquisition module 530 may be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0122] Figure 6 A block diagram of an electronic device suitable for implementing a three-dimensional target object recognition method according to an embodiment of the present invention is shown schematically.

[0123] like Figure 6 As shown, an electronic device 600 according to an embodiment of the present invention includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM 602) or a program loaded from a storage portion 608 into a random access memory (RAM 603). The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0124] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 602 and / or RAM 603. It should be noted that programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in one or more memories.

[0125] According to an embodiment of the present invention, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0126] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0127] According to embodiments of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM 603), read-only memory (ROM 602), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0128] Embodiments of the present invention also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the user interaction method provided in the embodiments of the present invention.

[0129] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this invention. According to embodiments of the invention, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0130] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0132] Those skilled in the art will understand that the features described in the various embodiments of the present invention can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments of the present invention can be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All such combinations and / or combinations fall within the scope of the present invention.

Claims

1. A method for recognizing three-dimensional targets, characterized in that, include: Acquire target object data to be identified, wherein the target object data is three-dimensional target object data acquired from a three-dimensional data acquisition or reconstruction device; The target object data is input into the target multi-task processing model, and the target object data is processed based on the target multi-task processing model to obtain the three-dimensional recognition information of the target object; as well as The recognition result of the target object is obtained based on the three-dimensional recognition information. The target multi-task processing model is a model obtained by fine-tuning a multi-task processing model trained on source domain data. The fine-tuning process includes: processing the source domain data and target domain data based on the multi-task processing model to obtain candidate region information; obtaining multiple domain migration sub-modules based on the multi-task structure in the multi-task processing model; training multiple domain migration sub-modules based on at least the source domain data, the target domain data, and the candidate region information to obtain updated candidate region information; and fine-tuning the multi-task processing model based on at least the updated candidate region information.

2. The method according to claim 1, characterized in that, The process of obtaining multiple domain migration sub-modules based on the multi-task structure in the multi-task processing model specifically includes: The multi-task structure in the multi-task processing model is decoupled to obtain multiple task sub-paths. Construct multiple sub-network structures based on multiple task sub-paths; and Domain discriminators are embedded in multiple sub-network structures to obtain multiple domain migration sub-modules.

3. The method according to claim 2, characterized in that, The plurality of said domain transfer submodules include a gradient inversion layer connected to the domain discriminator, the gradient inversion layer being used to invert the gradient direction of the domain discriminator during training.

4. The method according to claim 1, characterized in that, The target multi-task processing model includes a 3D feature extraction module, which extracts three-dimensional feature representations with geometric and semantic information from the source domain data and target domain data based on a three-dimensional convolutional network or a sparse volume coding structure.

5. The method according to any one of claims 1 to 4, characterized in that, The target object data includes three-dimensional image data of the target object obtained through security inspection equipment, and the recognition result is used to identify the category information and location information of the target object.

6. The method according to claim 5, characterized in that, The target multi-task processing model is used at least for target detection of the target object.

7. The method according to claim 6, characterized in that, The process of processing the source domain data and target domain data based on the multi-task processing model to obtain candidate region information specifically includes: The source domain data and the target domain data are respectively input into the multi-task processing model to obtain the corresponding preliminary detection results; and At least the category prediction score and the three-dimensional bounding box parameters are extracted from the preliminary detection results as the candidate region information.

8. The method according to claim 6, characterized in that, The plurality of domain migration submodules include at least a classification migration submodule and a regression migration submodule, wherein the classification migration submodule is used to predict the category information of the target object, and the regression migration submodule is used to predict the location information of the target object.

9. The method according to claim 8, characterized in that, The input to the regression transfer submodule includes at least the category prediction results output by the classification transfer submodule.

10. The method according to any one of claims 1 to 4, 6 to 9, characterized in that, The candidate region information includes source domain positive sample candidate region information, source domain negative sample candidate region information, target domain positive sample candidate region information, and target domain negative sample candidate region information.

11. The method according to claim 10, characterized in that, The training of multiple domain transfer submodules based at least on the source domain data, the target domain data, and the candidate region information specifically includes: Construct paired training samples consisting of positive and negative samples corresponding to the source domain data and the target domain data, wherein the paired training samples are associated with the candidate region information; The paired training samples are input into the corresponding domain transfer submodule to obtain the target prediction result; and Based on the target prediction results, the candidate region information is updated to generate the updated candidate region information.

12. A three-dimensional target object recognition device, characterized in that, The device includes: The data acquisition module is used to: acquire target object data to be identified, wherein the target object data is three-dimensional target object data acquired from a three-dimensional data acquisition or reconstruction device; The processing module is configured to: input the target object data into a target multi-task processing model; process the target object data based on the target multi-task processing model to obtain the three-dimensional recognition information of the target object; wherein the target multi-task processing model is a model obtained by fine-tuning a multi-task processing model trained on source domain data; the fine-tuning process includes: processing the source domain data and target domain data based on the multi-task processing model to obtain candidate region information; obtaining multiple domain transfer sub-modules based on the multi-task structure in the multi-task processing model; training multiple domain transfer sub-modules based at least on the source domain data, the target domain data, and the candidate region information to obtain updated candidate region information; and fine-tuning the multi-task processing model based at least on the updated candidate region information; and The recognition result acquisition module is used to: acquire the recognition result of the target object based on the three-dimensional recognition information.

13. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 11.

15. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 11.

Images