Age prediction method and device based on tooth image, equipment and storage medium

Abstract

The application provides a tooth image-based age prediction method, device and equipment and a storage medium, and relates to the technical field of image processing. The method comprises the following steps: acquiring a tooth image of a to-be-measured object, and performing positioning detection on the tooth image to obtain position information of a target tooth in the tooth image; performing cutting and tensor normalization on the tooth image according to the position information of the target tooth in the tooth image to obtain an input tensor corresponding to the target tooth; inputting the input tensor corresponding to the target tooth into a tooth age regression network model, capturing a high-dimensional feature tensor of the target tooth by the tooth age regression network model, performing dimension reduction on the high-dimensional feature tensor to obtain an age-related apparent feature vector of the target tooth, and obtaining an age prediction result output by the tooth age regression network model according to the age-related apparent feature vector of the target tooth, thereby realizing low-cost, non-radiation and automatic tooth age inference, and solving the problem that an existing AI model cannot perform age inference based on a tooth image.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and more specifically, to a method, apparatus, device, and storage medium for age prediction based on dental images. Background Technology

[0002] The morphology and structure of teeth change with age. Therefore, the age of the subject can be inferred based on age-related characteristics such as tooth wear, secondary dentin formation, and changes in tooth surface morphology and appearance.

[0003] Among existing methods for estimating tooth age, one type relies on examiners observing characteristics such as tooth wear, cracks, and color changes, and then determining the age range based on corresponding grading standards or regression relationships. This type of method depends to some extent on the examiner's experience, and it is quite difficult to objectively and stably quantify surface features such as localized wear and color / texture changes.

[0004] With the development of medical image analysis and deep learning technologies, some solutions have attempted to automatically estimate tooth age based on data such as panoramic oral radiographs and cone-beam computed tomography (CBCT) images. Among them, solutions based on panoramic oral radiographs mainly utilize the developmental status of teeth and the morphological features of two-dimensional medical images; while solutions based on CBCT images mostly utilize pulp chamber volume, tooth volume ratio, and three-dimensional structural parameters of the tooth root or crown.

[0005] Existing methods for estimating tooth age are still insufficient to objectively and stably quantify and comprehensively utilize age-related surface information such as the three-dimensional morphology, color, texture, cracks, and local wear boundaries of the tooth surface, and to directly obtain age estimation results based on this information.

[0006] Intraoral scanning devices can acquire three-dimensional morphological information of the tooth surface, and some devices can also simultaneously acquire color texture information of the tooth surface. Existing intraoral scanning solutions are mainly used for recording tooth surface morphology, detecting tooth wear, and monitoring changes; however, a stable technical approach for directly inferring age based on intraoral scanning data is still lacking. How to extract age-related surface morphology and appearance features from intraoral scanning data and establish the correspondence between these features and age remains to be solved.

[0007] Therefore, it is necessary to provide an age estimation method based on tooth surface scanning data to improve the objectivity, stability, and processing efficiency of tooth age estimation. Summary of the Invention

[0008] The purpose of this application is to address the shortcomings of the prior art by providing a method, apparatus, device, and storage medium for age prediction based on dental images, so as to solve the technical problems existing in the prior art.

[0009] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, embodiments of this application provide an age prediction method based on dental images, the method comprising: Acquire a tooth image of the subject to be tested, and perform localization detection on the tooth image to obtain the position information of the target tooth in the tooth image; Based on the position information of the target tooth in the tooth image, the tooth image is cropped and tensor normalized to obtain the input tensor corresponding to the target tooth; The input tensor corresponding to the target tooth is input into a pre-trained tooth age regression network model. The tooth age regression network model captures the high-dimensional feature tensor of the target tooth and reduces the dimensionality of the high-dimensional feature tensor to obtain the age-related apparent feature vector of the target tooth. Based on the age-related apparent feature vector of the target tooth, the age prediction result output by the tooth age regression network model is obtained.

[0010] Optionally, the dental age regression network model includes: a bone feature extraction network, a feature projection module, and an age regression module; The input tensor is input into the backbone feature extraction network, which then captures the high-dimensional feature tensor of the target tooth. The high-dimensional feature tensor is input into the feature projection module, which then compresses the high-dimensional feature tensor into the age-related apparent feature vector. The age-related epigenetic feature vector is input into the age regression module, which then maps the age-related epigenetic feature vector to obtain the age prediction result.

[0011] Optionally, the backbone feature extraction network includes a multi-layer convolutional network and an average pooling layer connected in sequence; The high-dimensional feature tensor of the target tooth captured by the bone feature extraction network includes: The input tensor is encoded in multiple layers by the multi-layer convolutional network to obtain deep visual features. Each layer of encoding is used to encode one type of tooth information in the input tensor. The tooth information includes at least one of the following: local texture, cracks, color changes, dentin exposure boundaries and morphological information of the target tooth. The average pooling layer compresses the deep visual features into a feature tensor of a preset dimension and outputs the high-dimensional feature tensor.

[0012] Optionally, the feature projection module includes: an image feature flattening submodule, a first fully connected layer, a nonlinear activation layer, and a random deactivation layer; The step of compressing the high-dimensional feature tensor into the age-related apparent feature vector by the feature projection module includes: The image feature flattening submodule performs a flattening operation on the high-dimensional feature tensor and outputs a one-dimensional feature vector; The first fully connected layer performs dimension mapping on the one-dimensional feature vector and outputs a feature vector with a preset dimension. The nonlinear activation layer reorganizes the feature vector of the preset dimension to obtain the reorganized features; The age-related epigenetic feature vector is obtained by randomly discarding some of the recombined features by the random deactivation layer.

[0013] Optionally, the age regression module includes: a gender feature fusion submodule and an age regression submodule; The process of mapping the age-related epigenetic feature vector by the age regression module to obtain the age prediction result of the test subject includes: The gender feature fusion submodule fuses the pre-constructed gender feature vector with the age-related apparent feature vector to obtain a joint feature vector; The age prediction result of the test object is obtained by performing regression mapping on the joint feature vector by the age regression submodule.

[0014] Optionally, the training process of the dental age regression network model includes: Acquire dental sample images and real age labels of multiple historical subjects, and preprocess the dental sample images of each historical subject to obtain the input tensor corresponding to the dental sample images of each historical subject. A training sample set is constructed based on the input tensor and real age label corresponding to the tooth sample images of each historical subject. The input tensor and real age label corresponding to all tooth images of the same historical subject belong to the same data subset. Based on the training sample set, the pre-constructed initial dental age regression network model is iteratively trained to obtain the dental age regression network model.

[0015] Optionally, the method further includes: Obtain the age-related epigenetic feature vectors and age prediction results corresponding to multiple dental images of the subject under test; Clustering is performed on the age-related epigenetic feature vectors corresponding to the multiple dental images to obtain multiple clusters; Based on the age prediction results corresponding to each tooth image in each cluster, the interpretability level of the age prediction results is obtained. The interpretability level is used to characterize the relative apparent aging stage of the test object.

[0016] Secondly, embodiments of this application also provide an age prediction device based on dental images, the device comprising: The acquisition module is used to acquire images of the teeth of the object to be tested. The processing module is used to perform localization detection on the tooth image to obtain the position information of the target tooth in the tooth image; based on the position information of the target tooth in the tooth image, the tooth image is cropped and tensor normalized to obtain the input tensor corresponding to the target tooth; The prediction module is used to input the input tensor corresponding to the target tooth into a pre-trained tooth age regression network model. The tooth age regression network model captures the high-dimensional feature tensor of the target tooth and reduces the dimensionality of the high-dimensional feature tensor to obtain the age-related apparent feature vector of the target tooth. Based on the age-related apparent feature vector of the target tooth, the age prediction result output by the tooth age regression network model is obtained.

[0017] Optionally, the dental age regression network model includes: a bone diaphragm feature extraction network, a feature projection module, and an age regression module; the prediction module is specifically used for: The input tensor is input into the backbone feature extraction network, which then captures the high-dimensional feature tensor of the target tooth. The high-dimensional feature tensor is input into the feature projection module, which then compresses the high-dimensional feature tensor into the age-related apparent feature vector. The age-related epigenetic feature vector is input into the age regression module, which then maps the age-related epigenetic feature vector to obtain the age prediction result.

[0018] Optionally, the backbone feature extraction network includes a multi-layer convolutional network and an average pooling layer connected in sequence; The prediction module is specifically used for: The input tensor is encoded in multiple layers by the multi-layer convolutional network to obtain deep visual features. Each layer of encoding is used to encode one type of tooth information in the input tensor. The tooth information includes at least one of the following: local texture, cracks, color changes, dentin exposure boundaries and morphological information of the target tooth. The average pooling layer compresses the deep visual features into a feature tensor of a preset dimension and outputs the high-dimensional feature tensor.

[0019] Optionally, the feature projection module includes: an image feature flattening submodule, a first fully connected layer, a nonlinear activation layer, and a random deactivation layer; The prediction module is specifically used for: The high-dimensional feature tensor is compressed into the age-related apparent feature vector, including: The image feature flattening submodule performs a flattening operation on the high-dimensional feature tensor and outputs a one-dimensional feature vector; The first fully connected layer performs dimension mapping on the one-dimensional feature vector and outputs a feature vector with a preset dimension. The nonlinear activation layer reorganizes the feature vector of the preset dimension to obtain the reorganized features; The age-related epigenetic feature vector is obtained by randomly discarding some of the recombined features by the random deactivation layer.

[0020] Optionally, the age regression module includes: a gender feature fusion submodule and an age regression submodule; The prediction module is specifically used for: The gender feature fusion submodule fuses the pre-constructed gender feature vector with the age-related apparent feature vector to obtain a joint feature vector; The age prediction result of the test object is obtained by performing regression mapping on the joint feature vector by the age regression submodule.

[0021] Optionally, the device further includes: a training module, used for: Acquire dental sample images and real age labels of multiple historical subjects, and preprocess the dental sample images of each historical subject to obtain the input tensor corresponding to the dental sample images of each historical subject. A training sample set is constructed based on the input tensor and real age label corresponding to the tooth sample images of each historical subject. The input tensor and real age label corresponding to all tooth images of the same historical subject belong to the same data subset. Based on the training sample set, the pre-constructed initial dental age regression network model is iteratively trained to obtain the dental age regression network model.

[0022] Optionally, the acquisition module is used to acquire age-related phenotypic feature vectors and age prediction results corresponding to multiple dental images of the subject under test; The clustering module is used to cluster the age-related epigenetic feature vectors corresponding to the multiple dental images to obtain multiple clusters. The processing module is further configured to obtain the interpretability level of the age prediction result based on the age prediction result corresponding to each tooth image in each cluster, wherein the interpretability level is used to characterize the relative apparent aging stage of the test object.

[0023] Thirdly, embodiments of this application provide an electronic device, which includes a memory for storing one or more programs and a processor. When the one or more programs are executed by the processor, the above-described age prediction method based on dental images is implemented.

[0024] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described age prediction method based on dental images.

[0025] The beneficial effects of this application are: This application provides a method, apparatus, device, and storage medium for age prediction based on dental images. In this application, a dental image of the subject is acquired and preprocessed, such as through localization detection, segmentation, and background zeroing / cropping, to obtain an input tensor corresponding to the target tooth. This input tensor is then fed into a dental age regression network model, which captures the high-dimensional feature tensor of the target tooth. The high-dimensional feature tensor is then reduced in dimensionality to obtain an age-related apparent feature vector of the target tooth. Based on this vector, the age prediction result of the subject is inferred. This achieves low-cost, non-radioactive, and automated dental age inference, improving the objectivity, stability, and processing efficiency of dental age inference, and solving the problem that existing AI models cannot perform age inference based on dental images.

[0026] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart illustrating an age prediction method based on dental images provided in this application embodiment; Figure 2 This is a schematic diagram of the structure of a dental age regression network model provided in an embodiment of this application; Figure 3 A flowchart illustrating another age prediction method based on dental images provided in this application embodiment; Figure 4 This is a schematic diagram of another dental age regression network model provided in an embodiment of this application; Figure 5 A flowchart illustrating another age prediction method based on dental images provided in this application embodiment; Figure 6 A flowchart illustrating another age prediction method based on dental images provided in this application embodiment; Figure 7 A flowchart illustrating another age prediction method based on dental images provided in this application embodiment; Figure 8 A schematic diagram of the structure of an age prediction device based on dental images provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0030] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0031] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0032] To address the problems existing in the prior art, this application proposes an age prediction method based on dental images. In this application, after acquiring the dental images of the subject, the dental images are preprocessed to obtain the input tensor corresponding to the target tooth in the dental image. The input tensor corresponding to the target tooth is then input into a pre-trained dental age regression network model to obtain the age prediction result of the subject. This method achieves low-cost, non-radioactive, and automated dental age inference without relying on X-rays, improving the objectivity, stability, and processing efficiency of dental age inference, and solving the problem that existing AI models cannot perform age inference based on dental images.

[0033] Optionally, refer to Figure 1 The diagram shown is a flowchart illustrating the age prediction method based on dental images provided in this application. Figure 1As shown, the subject of this method is an electronic device with data processing capabilities, such as an edge server or an embedded AI terminal, on which a pre-trained dental age regression model is deployed.

[0034] It should be understood that in other embodiments, the order of some steps in the pulp condition recognition model training method can be interchanged according to actual needs, or some steps can be omitted or deleted. For example... Figure 1 As shown, the method includes: S101. Obtain the tooth image of the object to be tested, and perform localization detection on the tooth image to obtain the position information of the target tooth in the tooth image.

[0035] For example, the tooth image can be a three-channel color image with dimensions denoted as H×W×3, where H is the image height and W is the image width. H×W can vary depending on the acquisition device or field of view and is not fixed in advance. Therefore, this application improves the repeatability of acquiring age-related appearance information of the tooth surface by acquiring standardized two-dimensional true-color images of the subject.

[0036] In one feasible approach, operators can use a digital intraoral scanner (such as Tero) to perform routine scans of the oral cavity or extracted teeth of the subject. While performing 3D modeling, the system background simultaneously captures a set of high-definition 2D raw images with standard white light or polarized light illumination, thereby obtaining standardized tooth surface images with a unified color space and no obvious shadow distortion, reducing generalization errors caused by differences in ambient light.

[0037] During the localization and detection of tooth images, the tooth image can be input into a pre-trained target detection network (such as YOLO) for full-image scanning to obtain the location information of the target tooth in the tooth image. The location information includes: bounding box coordinates and standard tooth position number. The target tooth can be the first molar, or it can be extended to other molars or multiple tooth positions.

[0038] Optionally, in this embodiment, by target detection and segmentation, and preferably by selecting the first molar as the target tooth, the adverse effects of complex oral background on age inference are reduced; S102. Based on the position information of the target tooth in the tooth image, the tooth image is cropped and the tensor is normalized to obtain the input tensor corresponding to the target tooth.

[0039] Optionally, the bounding box of the target tooth is used as the explicit spatial cue input image segmentation network (SAM) to generate a binary pixel mask that closely follows the anatomical edge of the target tooth. Based on the mask, a fixed number of pixels are extended outward from the bounding box to generate an extended bounding box, which is then cropped. The background pixels outside the mask are set to zero to reduce gingival and saliva reflections and interference from adjacent teeth. Then, the single-tooth image is scaled to a uniform input size, and tensor normalization is performed according to preset parameters to obtain the input tensor corresponding to the target tooth.

[0040] S103. Input the input tensor corresponding to the target tooth into the pre-trained tooth age regression network model. The tooth age regression network model captures the high-dimensional feature tensor of the target tooth and reduces the dimensionality of the high-dimensional feature tensor to obtain the age-related apparent feature vector of the target tooth. Based on the age-related apparent feature vector of the target tooth, the age prediction result output by the tooth age regression network model is obtained.

[0041] Among them, the dental age regression network model is pre-trained based on a large number of historical dental sample images and real age values ​​of the subjects, which can realize the age inference of the subjects.

[0042] For example, the input to the tooth age regression network model is a single tooth 224×224×3 true color tensor (i.e., not the original oral scan image, not the OPG grayscale image) that has been masked, zeroed, adaptively cropped, and normalized. This can eliminate interference from the gums, saliva, or adjacent teeth, ensuring that the model only learns the apparent features of the target tooth's occlusal surface itself.

[0043] In this embodiment, the input tensor corresponding to the target tooth is fed into a tooth age regression network model, which captures a high-dimensional feature tensor, such as a 1664-dimensional deep semantic feature map. However, considering that high-dimensional features contain a large amount of noise (such as individual differences and acquisition artifacts), direct regression is prone to overfitting. Therefore, it is proposed that a pre-defined Projector module can be used to perform supervised dimensionality reduction on the high-dimensional feature tensor to obtain the age-related phenotypic feature vector of the target tooth, compressing the 1664-dimensional vector into a 512-dimensional vector. This allows the dimensionality-reduced vector to retain age-related features such as wear, staining, and cracks. Based on the age-related phenotypic feature vector of the target tooth, the predicted age of the subject is inferred, such as 42.5. This achieves low-cost, non-radioactive, and automated tooth age inference, solving the problem that existing AI models cannot perform age inference based on tooth images.

[0044] Optionally, in this application, a digital intraoral scanner is used to perform routine scanning of the oral cavity or extracted teeth of the test subject, and the synchronously generated two-dimensional tooth images are input into the system. After the system completes target tooth detection, segmentation, background zeroing and cropping, and forward inference, it outputs the age prediction result corresponding to the target tooth. For example, when there are multiple target tooth images that meet the conditions for the same test subject, the age prediction value of each target tooth can be obtained separately, and the final age estimate at the individual level is output through age fusion at the decision layer based on weighted average. This effectively overcomes the specific errors that are very easy to occur in single-tooth assessment (such as unilateral chewing, local lesions, or poor quality of a single image), thereby achieving a higher fault tolerance and stronger robustness of individual age inference, improving the robustness of individual-level age estimation, and realizing end-to-end fully automatic output. At the same time, the overall solution does not rely on X-rays, realizing low-cost, non-radiation, and automated tooth age inference.

[0045] In summary, this application provides an age prediction method based on dental images. In this application, a dental image of the subject is acquired and preprocessed, such as through localization detection, segmentation, and background zeroing / cropping, to obtain an input tensor corresponding to the target tooth. This input tensor is then fed into a dental age regression network model, which captures the high-dimensional feature tensor of the target tooth. The high-dimensional feature tensor is then reduced in dimensionality to obtain the age-related apparent feature vector of the target tooth. Based on this vector, the age prediction result of the subject is inferred. This method achieves low-cost, non-radiative, and automated dental age inference, improving the objectivity, stability, and processing efficiency of dental age inference, and solving the problem that existing AI models cannot perform age inference based on dental images.

[0046] Meanwhile, compared with imaging pathways such as OPG and CBCT, this application does not rely on radiation imaging, making it easier to deploy in low-threshold, rapid screening scenarios; at the same time, it utilizes the apparent information such as the color, texture, cracks and morphology of the real tooth surface, forming different data sources and application advantages compared with existing X-ray pathways.

[0047] Optionally, refer to Figure 2 As shown, the dental age regression network model includes: a bone feature extraction network, a feature projection module, and an age regression module; (Reference) Figure 3 As shown, step S103 above includes: S201. Input the input tensor into the backbone feature extraction network, which then captures the high-dimensional feature tensor of the target tooth.

[0048] For example, the backbone feature extraction network can be DenseNet169, ResNet, EfficientNet, VGG, or ConvNeXt, or other convolutional neural networks capable of extracting deep image representations.

[0049] S202. Input the high-dimensional feature tensor into the feature projection module, which then compresses the high-dimensional feature tensor into an age-related apparent feature vector.

[0050] S203. Input the age-related apparent feature vector into the age regression module, and the age regression module maps the age-related apparent feature vector to obtain the age prediction result of the subject.

[0051] In this embodiment, the input tensor corresponding to the target tooth obtained through preprocessing is input to the bone feature extraction network. The input tensor corresponding to the target tooth can be a three-channel two-dimensional color image tensor of uniform size (224×224×3). The bone feature extraction network captures the 1664-dimensional high-dimensional feature tensor of the target tooth and inputs the high-dimensional feature tensor to the feature projection module. The feature projection module compresses the high-dimensional feature tensor into a 512-dimensional age-related appearance feature vector, which is a low-dimensional age-related appearance feature vector that is more suitable for continuous age regression and subsequent optional interpretation analysis. Finally, the age-related appearance feature vector is input to the age regression module, which performs regression mapping on the 512-dimensional age-related appearance feature vector to obtain the age prediction result of the subject.

[0052] Therefore, this application uses deep learning to automatically extract fine-grained appearance features such as color, texture, cracks, staining, and dentin exposure boundaries, thereby reducing the subjectivity of manual scoring.

[0053] Optionally, refer to Figure 4 As shown, the backbone feature extraction network consists of a series of interconnected multi-layer convolutional networks and an average pooling layer. like Figure 4 As shown, the backbone feature extraction network includes a multi-layer convolutional network, such as a first-layer convolutional network, a second-layer convolutional network, ..., an Nth-layer convolutional network. Specifically, the first-layer convolutional network encodes the shallow edges of the input tensor, and the second-layer convolutional network encodes the local texture of the input tensor.

[0054] refer to Figure 5 As shown, step S201 above includes: S301. The input tensor is encoded by a multi-layer convolutional network to obtain deep visual features. Each layer of encoding is used to encode one type of tooth information in the input tensor. The tooth information includes at least one of the following: local texture, cracks, color changes, dentin exposure boundaries and morphological information of the target tooth.

[0055] In this embodiment, the input tensor is sequentially subjected to multi-layer convolution operations, non-linear activation, and inter-layer feature transfer through a multi-layer convolutional network. This process encodes the local texture, cracks, color changes, dentin exposure boundaries, and morphological information of the target tooth occlusal surface layer by layer, allowing shallow visual information to be gradually combined into deep visual features with stronger semantic expressive power.

[0056] S302. The average pooling layer compresses the deep visual features into a feature tensor of a preset dimension and outputs a high-dimensional feature tensor.

[0057] Continue to refer to Figure 4 As shown, the fully connected layers at the ends of the backbone feature network used for image classification are removed, retaining only the feature extraction part, and then an adaptive global average pooling layer is connected in series. The purpose of removing the fully connected layers at the ends is to prevent the network from continuing to output classification probabilities for the class discrimination task, and instead use it as a visual feature extractor to extract image features suitable for the age regression task.

[0058] The purpose of cascading adaptive global average pooling layers is to compress the feature map with spatial dimensions into a feature tensor with fixed dimensions, thereby reducing the sensitivity of the subsequent fully connected mapping module to changes in the input size, and enabling the subsequent projection and regression modules to stably receive features of a uniform dimension.

[0059] In this embodiment, the deep visual features are compressed into a 1664-dimensional feature tensor by an adaptive global average pooling layer, reducing the sensitivity of the subsequent fully connected mapping module to changes in the input size.

[0060] Optionally, continue to refer to Figure 4 As shown, the feature projection module includes: an image feature flattening submodule, a first fully connected layer, a nonlinear activation layer, and a random deactivation layer.

[0061] Step S202 above includes: The image feature flattening submodule performs a flattening operation on the high-dimensional feature tensor and outputs a one-dimensional feature vector; the first fully connected layer performs dimensional mapping on the one-dimensional feature vector and outputs a feature vector of a preset dimension; the nonlinear activation layer performs feature recombination on the feature vector of the preset dimension to obtain the recombined features; and the random deactivation layer randomly discards some of the recombined features to obtain an age-related appearance feature vector.

[0062] In this embodiment, the feature projection module and the backbone feature extraction network are connected sequentially. First, the image feature flattening submodule performs a flattening operation on the high-dimensional feature tensor, converting the high-dimensional feature tensor into a one-dimensional feature vector. Then, the one-dimensional feature vector is input into the first fully connected layer, which performs dimensional mapping on the one-dimensional feature vector to obtain a preset dimension feature vector, such as mapping to 1024 dimensions. Then, the preset dimension feature vector is reorganized by a nonlinear activation layer. After that, the output of some neurons is masked by a random deactivation layer to reduce the risk of overfitting, and a 512-dimensional low-dimensional age-related apparent feature vector is output.

[0063] Therefore, through the feature projection module, the high-dimensional comprehensive representation output by the backbone feature network is gradually compressed into a 512-dimensional low-dimensional age-related appearance feature vector. This reduces redundant features and lowers the parameter scale and modeling difficulty of subsequent prediction modules while retaining key age-related information. This low-dimensional age-related appearance feature vector can be used as input to the age regression module, or in the preferred embodiment, it can be further used for subsequent cluster analysis, interpretability ranking generation, or other auxiliary interpretation tasks.

[0064] Optionally, continue to refer to Figure 4 As shown, the age regression module includes: a gender feature fusion submodule and an age regression submodule; The above step S203 includes: The gender feature fusion submodule fuses the pre-constructed gender feature vector with the age-related apparent feature vector to obtain a joint feature vector; the age regression submodule performs regression mapping on the joint feature vector to obtain the age prediction result of the subject.

[0065] like Figure 4 As shown, the gender feature fusion submodule includes a first fully connected layer, a non-linear activation layer, a second fully connected layer, and a feature concatenation unit connected in sequence. The first fully connected layer maps the one-dimensional gender label to a 16-dimensional intermediate feature. The second fully connected layer further maps the 16-dimensional intermediate feature to a 32-dimensional gender feature vector. Then, the 32-dimensional gender feature vector is concatenated with the 512-dimensional low-dimensional age-related phenotypic feature vector output by the feature projection module to form a 544-dimensional joint feature vector, which is then input into the subsequent age regression submodule. If the image branch output has other dimensions, the dimensions of the joint feature vector can be adjusted accordingly. Therefore, in this embodiment, by introducing the gender feature fusion submodule, the phenotypic information of the target tooth image and the possible differences in wear between individuals of different genders can be considered simultaneously during age prediction, thereby improving the model's adaptability to samples of different genders.

[0066] Optionally, when the gender information of the subject to be tested is known, the gender information can be input as an additional feature into the gender feature fusion submodule. The gender information is preferably input in the form of a one-dimensional numerical label, such as male, female or unknown state represented by a preset numerical code. After the gender information is encoded, it is fused with the image feature vector by the gender feature fusion module. This allows the age regression model to consider gender differences in the prediction process, thereby reducing the potential prediction bias caused by the difference in the wear and tear process of individuals of different genders and improving the pertinence and accuracy of the age inference results.

[0067] Continue to refer to Figure 4 As shown, the feature projection module, gender feature fusion submodule, and age regression submodule are sequentially connected. The age regression submodule includes a hidden fully connected layer, a non-linear activation layer, and a single-node linear layer. First, the 544-dimensional joint feature vector is input into the hidden fully connected layer, which maps the 544-dimensional vector to 128-dimensional vectors. Then, the non-linear activation layer enhances the non-linear expressive power. Finally, the activated 128-dimensional feature vector is input into the single-node linear output to obtain the age prediction result.

[0068] Optionally, if gender feature fusion is not enabled, the age regression module may also receive only the 512-dimensional low-dimensional age-related apparent feature vector output by the feature projection module for continuous age prediction.

[0069] Optionally, continue to refer to Figure 4 As shown, in the actual process of dental age estimation, multiple target tooth images of the same individual are usually acquired. To eliminate age prediction errors caused by chewing habits, local occlusal abnormalities, or fluctuations in the image acquisition quality of individual teeth, this application also introduces a multi-source information fusion mechanism at the decision layer. Specifically, the system inputs N qualified target tooth images into a pre-trained dental age regression network, performs forward inference independently, and obtains N independent sets of continuous age prediction values. ,in, The continuous age prediction value is 1. For continuous age prediction value 2, and Let N be the continuous age prediction value. Subsequently, the system uses a weighted average algorithm to fuse these prediction values ​​at the decision-making level.

[0070] Optionally, when a dental image involves multiple tooth positions, a separate modeling approach can be adopted for each tooth position. Alternatively, the tooth position number can be one-hot encoded or embedded encoded and then concatenated with the image features, and input into the age regression model. If the gender information of the subject is known, the gender information can be numerically encoded, one-hot encoded, or embedded encoded and then fused with the image features or tooth position features through the gender feature fusion module, and input into the regression model so that the model can simultaneously consider the influence of tooth position differences and gender differences on age inference.

[0071] Optionally, refer to Figure 6 As shown, the training process of the dental age regression network model includes: S401. Obtain dental sample images and real age labels of multiple historical subjects, and preprocess the dental sample images of each historical subject to obtain the input tensor corresponding to the dental sample images of each historical subject.

[0072] Among them, the tooth sample image can be a two-dimensional color image of the occlusal surface generated synchronously using an intraoral scanning device.

[0073] The preprocessing includes: identifying the location of the first molar in the tooth sample image through an object detection network and outputting a bounding box; then obtaining the target tooth mask through an instance segmentation model; subsequently, zeroing out the background outside the mask, cropping the single tooth region, and uniformly scaling it to a 224×224 pixel input format to obtain the input tensor corresponding to the tooth sample image of each historical subject.

[0074] S402. Construct a training sample set based on the input tensor corresponding to the tooth sample images of each historical examinee and the real age label.

[0075] In this study, the input tensors and real age labels corresponding to all dental images of the same historical subject belong to the same subset of data.

[0076] Optionally, to reduce the risk of data leakage due to model memorization of individual specificities, it is proposed that when constructing the training sample set, the identifier ID of each historical examinee is used as the smallest partitioning unit to ensure that the input tensors and real age labels corresponding to all dental images of the same historical examinee belong to the same data subset. For example, when a historical examinee provides multiple target tooth images, all multiple target tooth images of that historical examinee are either used in training or all are excluded, without cross-patient pooling.

[0077] S403. Based on the training sample set, iteratively train the pre-constructed initial dental age regression network model to obtain the dental age regression network model.

[0078] During training, the Adam optimization algorithm is used to calculate gradients through backpropagation, and the model parameters of the initial dental age regression network are updated synchronously and end-to-end.

[0079] Optionally, during model training, considering that extreme dietary habits, restorations, abnormal wear, or special pathologies may lead to outliers in the "age-apparent change relationship," mean absolute error (MAE / L1Loss) or other robust regression loss functions can be used during training to improve the model's tolerance to outliers. Furthermore, the Adam optimization algorithm is used during training, and gradients are calculated through backpropagation. The parameters of the initial tooth age regression network model, including the sex feature fusion submodule, age regression submodule, feature projection module, and bone feature extraction network, are updated synchronously and end-to-end, enabling the model to gradually learn the nonlinear mapping relationship between the visual features of the target tooth image, sex features, and age values.

[0080] Optionally, refer to Figure 7 As shown, the method also includes: S501. Obtain the age-related epigenetic feature vectors and age prediction results corresponding to multiple dental images of the subject.

[0081] S502. Cluster the age-related phenotypic feature vectors corresponding to multiple dental images to obtain multiple clusters.

[0082] For example, the clustering algorithm can be: K-means, hierarchical clustering, spectral clustering, or multi-algorithm consensus clustering.

[0083] S503. Based on the age prediction results corresponding to multiple tooth images in each cluster, obtain the interpretability level of the age prediction results.

[0084] Among them, the interpretability rating is used to characterize the relative performance aging stage of the test object.

[0085] In one feasible approach, to enhance the interpretability of the age prediction results, this application proposes that an interpretable level output can be further generated based on a low-dimensional age-related apparent feature vector. This interpretable level output is not a necessary prerequisite for the success of this approach, but rather an auxiliary interpretation module based on the main output of the age prediction results.

[0086] Specifically, adaptive K-value inference: First, obtain the maximum and minimum values ​​of the age prediction results corresponding to multiple teeth of the subject, and then dynamically determine the number of clusters by combining them with a preset age span constant. This allows the number of explanatory levels to be adjusted as the age range of the sample changes. .

[0087] in, This represents the maximum value of the age prediction result. This represents the minimum value of the age prediction result. This is a preset age span constant.

[0088] Optionally, to balance the application habits of forensic medicine with the stability of unsupervised classification, the number of clusters K can be constrained within a preset range, such as 2 to 6 levels.

[0089] To improve the stability of unsupervised classification, cluster analysis can be performed on the age-related phenotypic feature vectors corresponding to multiple dental images to obtain multiple clusters. The grade labels are then rearranged according to the size of the age prediction results corresponding to multiple dental images in each cluster to obtain the interpretability grade output from Grade 1 to Grade K. For example, the final age prediction result is expressed as "42.6 years old, interpretability grade is Grade 4". The Grade result is used to help explain the relative aging stage of the sample.

[0090] The process of determining the rating label includes: Low-dimensional age-related phenotypic feature vectors corresponding to multiple historical subjects are obtained. These vectors are then input in parallel into various heterogeneous clustering algorithms to obtain multiple sets of different initial clustering labels. Finally, the co-occurrence frequency of any pair of samples being grouped into the same cluster in the results of different clustering algorithms is calculated. ,in, Let the co-occurrence frequency of tooth images i and j be assigned to the same cluster. For the tooth image in the m-th clustering result The assigned cluster label, For the tooth image in the m-th clustering result The assigned cluster label, This represents the total number of times the clustering algorithm was run.

[0091] Based on the co-occurrence frequencies obtained above, a co-occurrence matrix is ​​constructed, and the results of each clustering algorithm are fused according to the co-occurrence matrix to finally obtain a globally consistent initial consensus clustering label, which is used for subsequent age semantic ranking.

[0092] Since the initial cluster labels only represent the aggregation relationship in the feature space and lack direct age semantics, we further calculate the statistics of the age prediction results corresponding to each historical examinee generated by the age regression submodule within each cluster.

[0093] The preferred value is the mean, for example, , This represents the average of all samples within each cluster. For the first The total number of samples contained in each cluster. This provides the predicted age for each historical subject. In the presence of outliers, the median can be used as an alternative ranking criterion. Finally, based on the ascending order of statistical age within each cluster, the original unordered clusters are reordered to obtain an interpretative rank output consistent with the age change trend.

[0094] Based on the statistical age of each cluster from smallest to largest, the original unordered clusters are reordered to obtain an explanatory grade output consistent with the age change trend. This grade can be represented as Grade1 to GradeK, which is used to help explain the relative apparent aging stage of the sample.

[0095] Therefore, compared with existing oral scan studies that focus on wear detection and OPG or CBCT studies that focus on X-ray image age estimation, this application uses standardized two-dimensional color images of the occlusal surface of teeth generated simultaneously by oral scans as the core input object. It constructs an automatic age inference scheme for forensic medicine around the direct mapping from the target tooth image to continuous age values, so as to achieve low-cost, non-radioactive, and automated tooth age inference.

[0096] Optionally, refer to Figure 8 The diagram shown is a structural schematic of an age prediction device based on dental images provided in this application. The device includes: Acquisition module 801 is used to acquire images of the teeth of the object to be tested; Processing module 802 is used to perform localization detection on the tooth image to obtain the position information of the target tooth in the tooth image; and to perform cropping and tensor normalization on the tooth image based on the position information of the target tooth in the tooth image to obtain the input tensor corresponding to the target tooth. The prediction module 803 is used to input the input tensor corresponding to the target tooth into a pre-trained tooth age regression network model, which captures the high-dimensional feature tensor of the target tooth and reduces the dimensionality of the high-dimensional feature tensor to obtain the age-related apparent feature vector of the target tooth. Based on the age-related apparent feature vector of the target tooth, the age prediction result output by the tooth age regression network model is obtained.

[0097] Optionally, the dental age regression network model includes: a bone feature extraction network, a feature projection module, and an age regression module; the prediction module 803 is specifically used for: The input tensor is input into the backbone feature extraction network, which then captures the high-dimensional feature tensor of the target tooth. The high-dimensional feature tensor is input into the feature projection module, which then compresses the high-dimensional feature tensor into the age-related apparent feature vector. The age-related epigenetic feature vector is input into the age regression module, which then maps the age-related epigenetic feature vector to obtain the age prediction result.

[0098] Optionally, the backbone feature extraction network includes a multi-layer convolutional network and an average pooling layer connected in sequence; The prediction module 803 is specifically used for: The input tensor is encoded in multiple layers by the multi-layer convolutional network to obtain deep visual features. Each layer of encoding is used to encode one type of tooth information in the input tensor. The tooth information includes at least one of the following: local texture, cracks, color changes, dentin exposure boundaries and morphological information of the target tooth. The average pooling layer compresses the deep visual features into a feature tensor of a preset dimension and outputs the high-dimensional feature tensor.

[0099] Optionally, the feature projection module includes: an image feature flattening submodule, a first fully connected layer, a nonlinear activation layer, and a random deactivation layer; The prediction module 803 is specifically used for: The high-dimensional feature tensor is compressed into the age-related apparent feature vector, including: The image feature flattening submodule performs a flattening operation on the high-dimensional feature tensor and outputs a one-dimensional feature vector; The first fully connected layer performs dimension mapping on the one-dimensional feature vector and outputs a feature vector with a preset dimension. The nonlinear activation layer reorganizes the feature vector of the preset dimension to obtain the reorganized features; The age-related epigenetic feature vector is obtained by randomly discarding some of the recombined features by the random deactivation layer.

[0100] Optionally, the age regression module includes: a gender feature fusion submodule and an age regression submodule; The prediction module 803 is specifically used for: The gender feature fusion submodule fuses the pre-constructed gender feature vector with the age-related apparent feature vector to obtain a joint feature vector; The age prediction result of the test object is obtained by performing regression mapping on the joint feature vector by the age regression submodule.

[0101] Optionally, the device further includes: a training module, used for: Acquire dental sample images and real age labels of multiple historical subjects, and preprocess the dental sample images of each historical subject to obtain the input tensor corresponding to the dental sample images of each historical subject. A training sample set is constructed based on the input tensor and real age label corresponding to the tooth sample images of each historical subject. The input tensor and real age label corresponding to all tooth images of the same historical subject belong to the same data subset. Based on the training sample set, the pre-constructed initial dental age regression network model is iteratively trained to obtain the dental age regression network model. During the training process, the Adam optimization algorithm is used to calculate the gradient through the backpropagation algorithm, and the model parameters of the initial dental age regression network are updated synchronously and end-to-end.

[0102] Optionally, the acquisition module is used to acquire age-related phenotypic feature vectors and age prediction results corresponding to multiple dental images of the subject under test; The clustering module is used to cluster the age-related epigenetic feature vectors corresponding to the multiple dental images to obtain multiple clusters. The processing module is further configured to obtain the interpretability level of the age prediction result based on the age prediction result corresponding to each tooth image in each cluster, wherein the interpretability level is used to characterize the relative apparent aging stage of the test object.

[0103] The above-described device is used to execute the method provided in the foregoing embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

[0104] These modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more digital signal processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). Alternatively, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a system-on-a-chip (SOC).

[0105] Figure 9This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device can be integrated into a terminal device or a chip of a terminal device. The electronic device can be a computing device with data processing function, used to execute the method steps provided in the above embodiments to realize the automated inference of dental age.

[0106] The electronic device includes: processor 901 and memory 902.

[0107] The memory 902 is used to store programs, and the processor 901 calls the programs stored in the memory 902 to execute the above method embodiments. The specific implementation and technical effects are similar, and will not be described in detail here.

[0108] Optionally, this application also provides a program product, such as a computer-readable storage medium, including a program that, when executed by a processor, performs the above-described method embodiments.

[0109] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. 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 marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive 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 and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0110] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0111] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they 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 a portion 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 USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0112] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0113] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An age prediction method based on dental images, characterized in that, The method includes: A tooth image of the subject to be tested is acquired, and the tooth image is used for localization detection to obtain the position information of the target tooth in the tooth image; the tooth image is a color image; Based on the position information of the target tooth in the tooth image, the tooth image is cropped and tensor normalized to obtain the input tensor corresponding to the target tooth; The input tensor corresponding to the target tooth is input into a pre-trained dental age regression network model. The dental age regression network model captures the high-dimensional feature tensor of the target tooth and performs dimensionality reduction on the high-dimensional feature tensor to obtain the age-related apparent feature vector of the target tooth. Based on the age-related apparent feature vector of the target tooth, the age prediction result output by the dental age regression network model is obtained. The dental age regression network model includes: a bone feature extraction network, a feature projection module, and an age regression module. The input tensor is input into the backbone feature extraction network, which then captures the high-dimensional feature tensor of the target tooth. The high-dimensional feature tensor is input into the feature projection module, which then compresses the high-dimensional feature tensor into the age-related apparent feature vector. The age-related epigenetic feature vector is input into the age regression module, which then maps the age-related epigenetic feature vector to obtain the age prediction result. The backbone feature extraction network includes a multi-layer convolutional network and an average pooling layer connected in sequence. The high-dimensional feature tensor of the target tooth captured by the bone feature extraction network includes: The input tensor is encoded in multiple layers by the multi-layer convolutional network to obtain deep visual features. Each layer of encoding is used to encode one type of tooth information in the input tensor. The tooth information includes at least one of the following: local texture, cracks, color changes, dentin exposure boundaries and morphological information of the target tooth. The average pooling layer compresses the deep visual features into a feature tensor of a preset dimension and outputs the high-dimensional feature tensor.

2. The method according to claim 1, characterized in that, The feature projection module includes: an image feature flattening submodule, a first fully connected layer, a nonlinear activation layer, and a random deactivation layer; The step of compressing the high-dimensional feature tensor into the age-related apparent feature vector by the feature projection module includes: The image feature flattening submodule performs a flattening operation on the high-dimensional feature tensor and outputs a one-dimensional feature vector; The first fully connected layer performs dimension mapping on the one-dimensional feature vector and outputs a feature vector with a preset dimension. The nonlinear activation layer reorganizes the feature vector of the preset dimension to obtain the reorganized features; The age-related epigenetic feature vector is obtained by randomly discarding some of the recombined features by the random deactivation layer.

3. The method according to claim 1, characterized in that, The age regression module includes: a gender feature fusion submodule and an age regression submodule; The process of mapping the age-related epigenetic feature vector by the age regression module to obtain the age prediction result of the test subject includes: The gender feature fusion submodule fuses the pre-constructed gender feature vector with the age-related apparent feature vector to obtain a joint feature vector; The age prediction result of the test object is obtained by performing regression mapping on the joint feature vector by the age regression submodule.

4. The method according to any one of claims 1-3, characterized in that, The training process of the dental age regression network model includes: Acquire dental sample images and real age labels of multiple historical subjects, and preprocess the dental sample images of each historical subject to obtain the input tensor corresponding to the dental sample images of each historical subject. A training sample set is constructed based on the input tensor and real age label corresponding to the tooth sample images of each historical subject. The input tensor and real age label corresponding to all tooth images of the same historical subject belong to the same data subset. Based on the training sample set, the pre-constructed initial dental age regression network model is iteratively trained to obtain the dental age regression network model.

5. The method according to claim 1, characterized in that, The method further includes: Obtain the age-related epigenetic feature vectors and age prediction results corresponding to multiple dental images of the subject under test; Clustering is performed on the age-related epigenetic feature vectors corresponding to the multiple dental images to obtain multiple clusters; Based on the age prediction results corresponding to each tooth image in each cluster, the interpretability level of the age prediction results is obtained. The interpretability level is used to characterize the relative apparent aging stage of the test object.

6. An age prediction device based on dental images, characterized in that, The device includes: The acquisition module is used to acquire images of the teeth of the object to be tested; the teeth images are color images. The processing module is used to perform localization detection on the tooth image to obtain the position information of the target tooth in the tooth image; based on the position information of the target tooth in the tooth image, the tooth image is cropped and tensor normalized to obtain the input tensor corresponding to the target tooth; The prediction module is used to input the input tensor corresponding to the target tooth into a pre-trained tooth age regression network model. The tooth age regression network model captures the high-dimensional feature tensor of the target tooth and reduces the dimensionality of the high-dimensional feature tensor to obtain the age-related apparent feature vector of the target tooth. Based on the age-related apparent feature vector of the target tooth, the age prediction result output by the tooth age regression network model is obtained. The dental age regression network model includes: a bone diaphragm feature extraction network, a feature projection module, and an age regression module; the prediction module is specifically used for: The input tensor is input into the backbone feature extraction network, which then captures the high-dimensional feature tensor of the target tooth. The high-dimensional feature tensor is input into the feature projection module, which then compresses the high-dimensional feature tensor into the age-related apparent feature vector. The age-related epigenetic feature vector is input into the age regression module, which then maps the age-related epigenetic feature vector to obtain the age prediction result. The backbone feature extraction network includes a series of interconnected multi-layer convolutional networks and average pooling layers; the prediction module is specifically used for: The high-dimensional feature tensor of the target tooth captured by the bone feature extraction network includes: The input tensor is encoded in multiple layers by the multi-layer convolutional network to obtain deep visual features. Each layer of encoding is used to encode one type of tooth information in the input tensor. The tooth information includes at least one of the following: local texture, cracks, color changes, dentin exposure boundaries and morphological information of the target tooth. The average pooling layer compresses the deep visual features into a feature tensor of a preset dimension and outputs the high-dimensional feature tensor.

7. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is in operation, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the method as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which is executed by a processor to perform the method as described in any one of claims 1-5.

Images