Method for identifying mixed dental sites of local oral image, storage medium and device

Abstract

The present application relates to the technical field of oral image recognition, and discloses a mixed tooth site recognition method for a local oral image, a storage medium and a device. The method comprises the following steps: collecting a clinical local oral image, manually marking tooth site numbers and category information, and forming an original data set; replacing the tooth site numbers of the original data set with corresponding permanent tooth numbers at the same position, deleting the category information, and forming a tooth site marked data set; deleting the tooth site numbers in the original data set, and numerically representing the category information, to form a classification marked data set; training a tooth site recognition model through the tooth site marked data set; training a classification recognition model through the classification marked data set; obtaining an oral image to be recognized, recognizing the tooth site numbers of the teeth in the oral image through the tooth site recognition model, and recognizing the category of the teeth in the oral image through the classification recognition model. The method can improve the stability and accuracy of recognizing the positions of the teeth from a local oral image.

Description

Technical Field

[0001] This invention relates to the field of oral image recognition technology, specifically to a method, storage medium, and device for mixed tooth position recognition of local oral images. Background Technology

[0002] With the rapid development of deep learning technology in medical image processing, it is now widely used in the diagnosis of oral problems. Existing tooth position recognition methods mostly require oral scanning to obtain full-tooth images or to reconstruct a 3D model of the oral cavity, utilizing stable panoramic information to accurately identify individual tooth positions. However, in actual clinical applications, oral observation instruments have gradually become an important auxiliary tool for dental diagnosis. But these instruments can only acquire high-definition images of local teeth, resulting in a lack of complete dentition structure in a single frame. Furthermore, factors such as tooth posture, occlusion, and viewing angle changes during imaging make it difficult for existing deep learning models to stably and accurately predict tooth positions in local oral images acquired by oral observation instruments. Moreover, in children and adolescents, permanent and deciduous teeth may coexist in the oral cavity, resulting in a "mixed tooth" situation. Existing tooth recognition solutions typically require each tooth to be classified individually. However, human teeth fall into 48 categories: 28 categories of permanent teeth and 20 categories of deciduous teeth. Learning to classify these 48 categories significantly increases the difficulty of model training and can also lead to feature confusion between categories, affecting accuracy and stability. This further complicates the task of reliably identifying tooth positions from incomplete dentition structures. Therefore, a novel tooth position recognition scheme suitable for oral imaging is needed to address these technical challenges. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a hybrid tooth position recognition method for local oral images, which can improve the stability and accuracy of tooth position recognition in local oral images obtained by oral observation instrument.

[0004] To solve the above problems, the technical solution adopted by the present invention is as follows: a method for hybrid tooth position recognition of local oral images, comprising the following steps:

[0005] A large number of clinical oral cavity images were collected, and the tooth position number and category information of each tooth in the oral cavity images were manually labeled.

[0006] The tooth position numbers of deciduous teeth in the annotated local oral images are replaced with the corresponding permanent tooth numbers according to the positional mapping relationship between deciduous teeth and permanent teeth, and the category information of each tooth is deleted to form a tooth position annotation dataset.

[0007] The tooth position numbers in the manually annotated local oral images are deleted, and the category information of each tooth is quantified to form a classification and annotation dataset.

[0008] By training the model using a tooth position annotation dataset, a tooth position recognition model that focuses on the relative position information of teeth can be obtained.

[0009] By training the model on a labeled dataset, a classification and recognition model that focuses on the category information of teeth can be obtained.

[0010] The system acquires an oral cavity image to be identified, identifies the tooth position number of the teeth in the oral cavity image using a tooth position recognition model, and identifies the category of the teeth in the oral cavity image using a classification recognition model.

[0011] Compared to existing technologies, the advantages of this invention are as follows: It breaks down the tooth position recognition process into two steps: tooth position number recognition and tooth type recognition. Separate recognition models are trained for each step. When training the tooth position number recognition model, the manually labeled deciduous tooth position numbers are replaced with the corresponding permanent tooth position numbers. This simplifies the recognition of a total of 48 categories of deciduous and permanent teeth to 28 categories of permanent teeth. This significantly reduces the number of tooth position recognition categories, lowers the model training difficulty, avoids the probability of feature confusion between categories, and improves the accuracy and stability of tooth position recognition in local oral images lacking complete dentition structures. Furthermore, this method uses manually labeled local oral images to train the tooth position recognition model, allowing the model to learn the relative positional relationships of teeth in local oral images under different poses, viewpoints, and occlusion conditions, thereby further improving the accuracy of tooth position recognition in local oral images acquired by an oral observation instrument.

[0012] The above-mentioned method for mixed tooth position recognition based on local oral images further includes the following steps:

[0013] Prior knowledge of tooth positions is used to correct the identification of tooth position numbers that are not continuous in spatial location.

[0014] A hybrid tooth position recognition method for local oral images, wherein the steps of training a model using a tooth position annotation dataset to obtain a tooth position recognition model focusing on the relative position information of teeth include:

[0015] Randomly occlude images in the tooth position annotation dataset to obtain a relative position dataset;

[0016] A tooth position recognition model that focuses on the relative position information of teeth is obtained by training a model using a relative position dataset.

[0017] The above-mentioned method for hybrid tooth position recognition based on local oral images includes the step of correcting the identified tooth position numbers that are spatially discontinuous using prior knowledge of tooth positions, which includes:

[0018] Establish a tooth position matrix corresponding to the actual spatial location of each tooth position number;

[0019] The number of each tooth position number identified from the oral cavity image is filled into the corresponding position in the tooth position matrix;

[0020] If a tooth position number in the tooth position matrix is ​​identified repeatedly, the number of identified teeth on both sides of the repeated tooth position number is compared. The segment with fewer teeth is shifted in the direction with fewer teeth to the direction with fewer teeth, and the gaps created by the shift are filled.

[0021] If there are gaps in the tooth position segments identified in the tooth position matrix, the number of identified tooth positions on both sides of the gap is compared, and the segment with fewer tooth positions is shifted to the segment with more tooth positions to fill the gap.

[0022] A storage medium storing a computer program, which, when executed by a processor, implements the above-described method for mixed tooth position recognition of local oral images.

[0023] A mixed tooth position recognition device for local oral images, characterized in that it includes a processor and a memory, the processor and the memory being electrically connected, the processor implementing the above-mentioned mixed tooth position recognition method for local oral images by calling and executing a computer program in the memory.

[0024] A hybrid tooth position recognition device for local oral images includes: an acquisition module for acquiring local oral images; a tooth position recognition model for identifying the tooth position numbers in the acquired local oral images, wherein the tooth position recognition model is trained on a tooth position annotation dataset formed by manually annotating the tooth position numbers of each tooth in the local oral images and replacing the tooth position numbers of deciduous teeth in the local oral images with the corresponding permanent tooth numbers according to the position mapping relationship between deciduous teeth and permanent teeth; and a classification recognition model for identifying the category of teeth in the acquired local oral images, wherein the classification recognition model is trained on a classification annotation dataset formed by manually annotating the category information of each tooth and quantifying the category information of each tooth.

[0025] The aforementioned hybrid tooth position recognition device for local oral images further includes a correction module, which is used to correct the identified tooth position numbers that are not continuous in spatial position by utilizing prior knowledge of tooth position.

[0026] In the above-mentioned hybrid tooth position recognition device for local oral images, during the training process of the tooth position recognition model, the images in the tooth position annotation dataset are first randomly occluded to obtain a relative position dataset, and then the tooth position recognition model is trained using the relative position dataset.

[0027] The aforementioned hybrid tooth position recognition device for local oral images includes a correction module comprising a tooth position matrix corresponding to the actual spatial position of each tooth position number, a tooth position filling module, and a tooth position moving module. The tooth position filling module is used to fill the number of each tooth position number identified by the tooth position recognition model into the corresponding position of the tooth position matrix. If a tooth position number in the tooth position matrix is ​​repeatedly identified, the tooth position moving module is used to compare the number of identified teeth on both sides of the repeatedly identified tooth position number, and shift the segment with fewer teeth positions in the direction with fewer teeth positions to the direction with fewer teeth positions, and fill the gaps created by the shift. If there are gaps in the identified tooth position segments in the tooth position matrix, the tooth position moving module is used to compare the number of identified teeth on both sides of the gap, and shift the segment with fewer teeth positions in the direction with more teeth positions to fill the gaps. Attached Figure Description

[0028] Figure 1 This is a flowchart of a method for identifying mixed tooth positions from local oral images according to a first embodiment of the present invention.

[0029] Figure 2 This is a flowchart of a method for identifying mixed tooth positions from local oral images according to a second embodiment of the present invention.

[0030] Figure 3 This is a schematic diagram of the hybrid tooth position recognition device for local oral images according to the first embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram of the hybrid tooth position recognition device for local oral images according to the second embodiment of the present invention. Detailed Implementation

[0032] The embodiments of the present invention are described in detail below, with reference to... Figure 1 The present invention provides a method for hybrid tooth position recognition based on local oral cavity images, comprising the following steps:

[0033] First, a large number of clinical oral cavity images were collected. Dental experts then manually labeled the tooth position numbers and tooth categories of each tooth in the collected oral cavity images from different angles, regions, and occlusion conditions to form the original dataset.

[0034] Then, the tooth position numbers of deciduous teeth in the annotated local oral images are replaced with the corresponding permanent tooth numbers according to the position mapping relationship between deciduous teeth and permanent teeth. For example, the tooth position numbers 51 and 52 of deciduous teeth are replaced with the tooth position numbers 11 and 12 of permanent teeth in the same position. The category information of each tooth is deleted to form the tooth position annotation dataset H_dataset, which contains annotations of 28 categories of permanent teeth.

[0035] Meanwhile, the tooth position numbers in the annotated local oral images are deleted, and the category information of each tooth is numericalized to form a classification annotation dataset HR_dataset. In this embodiment, the main identification object is the oral cavity of children or adolescents, so the classification annotation dataset HR_dataset mainly focuses on two categories: permanent teeth and deciduous teeth. The annotation of permanent teeth can be numericalized as "1", and the annotation of deciduous teeth can be numericalized as "2".

[0036] Subsequently, a tooth position recognition model HM_model focusing on the relative position information of teeth was obtained by training the model using the tooth position annotation dataset H_dataset; and a classification recognition model HR_model focusing on the category information of teeth was obtained by training the model using the classification annotation dataset HR_dataset.

[0037] Finally, the oral cavity image to be identified is obtained. The tooth position recognition model HM_model is used to identify the position information of each tooth in the acquired oral cavity image, i.e., the tooth position number of each tooth. The classification recognition model HR_model is used to identify the category information of each tooth in the acquired oral cavity image, i.e., the permanent protuberance information. After the identification is completed, the tooth position number and tooth category information of each identified tooth are integrated and the result is output, thus completing the tooth position identification.

[0038] This method breaks down the tooth position recognition process into two independent subtasks: tooth position recognition and tooth category recognition. It replaces the deciduous tooth position numbers in the tooth position annotation dataset H_dataset used to train the tooth position recognition model HM_model with the corresponding permanent tooth position numbers. This simplifies the traditional model's learning of 48 tooth categories to learning 28 categories of permanent tooth position numbers, and learning a total of 30 categories (permanent and deciduous teeth). This significantly reduces the number of categories the model needs to learn, thus lowering the probability of misidentification due to feature confusion between categories. By splitting tooth position categories into position categories and permanent / deciduous categories, and having two separate models learn their respective features, interference from other features is eliminated. This allows the tooth position recognition model HM_model to focus more on learning the relative positional relationships of each tooth position number, and the category recognition model HR_dataset to focus more on learning the morphological features of each tooth, thereby improving the recognition accuracy and stability of both models. By using manually annotated local images of the oral cavity to construct a dataset for model training, the model can learn the relative positional relationships and pose features of teeth in local images of the oral cavity under different poses, viewpoints, and occlusions. This allows the trained model to be more adapted to local images of the oral cavity and to more accurately and stably identify the position and category of each tooth in the image even when there is a lack of complete dental arch structure information, thereby further improving the stability and accuracy of recognition.

[0039] Understandably, in some possible implementations, the missing tooth annotations in the classification annotation dataset HR_dataset can also be numerically converted to "0".

[0040] Reference Figure 2 In some possible implementations, because the number of occlusion scenarios in clinically collected oral cavity images is limited, in order for the tooth position recognition model HM_model to focus more on the relative positional relationships between tooth position numbers and to enable the model to infer tooth position numbers based on surrounding information, such as inferring the position number of a side-obstructed tooth based on the positional relationship between adjacent tooth positions and reference features inside the oral cavity, such as the tongue, it is necessary to randomly occlude the images in the tooth position annotation dataset H_dataset to form a relative position dataset HM_dataset. Then, the tooth position recognition model HM_model is trained using this randomly occluded relative position dataset HM_dataset. In this embodiment, half of the teeth are randomly occluded each time. For example, if there are three teeth in an image, the pixels of one of the teeth are randomly set to zero, but the labeled tooth position number must be retained.

[0041] Reference Figure 2In some possible implementations, to further improve the accuracy of the identified tooth position numbers, after identifying the tooth position numbers of each tooth in the acquired oral cavity image using the tooth position recognition model HM_model, prior indications of the tooth positions can be used to correct tooth position numbers with discontinuous spatial positions. Specifically, a tooth position matrix is ​​established based on the actual spatial position relationships of the tooth position numbers of each permanent tooth, as shown in the table below:

[0042] Table 1 Original tooth position matrix

[0043]

[0044] The cells above and below each tooth position number are used to fill in the number of corresponding tooth position numbers in the oral cavity image identified by the tooth position recognition model HM_model. In practice, the tooth position matrix can be set in the storage medium as a two-dimensional array. The number of each tooth position number identified by the tooth position recognition model HM_model is stored in the element with the corresponding index in the two-dimensional array according to the mapping relationship between the tooth position number and the row and column numbers of the tooth position matrix. Assume that the tooth position matrix is ​​set in the storage medium as an array HM[a][b], where a represents the row of the tooth position matrix, with a value of 0 or 1, and b represents the column of the tooth position matrix, with a value of a positive number from 0 to 13. If the tooth position recognition model HM_model identifies a tooth with a tooth position number of 11 in the oral cavity image, the number "1" is written into HM[0][6].

[0045] In practice, the tooth position recognition model HM_model may identify the tooth position numbers of two adjacent teeth in an oral cavity image as the same tooth position number, resulting in duplicate recognition of unique tooth position numbers, as shown in Table 2; or it may identify the tooth position numbers incorrectly, resulting in spatial discontinuity in the filled tooth position matrix, as shown in Table 3.

[0046] Table 2 Tooth position matrix for repeated identification

[0047]

[0048] Table 3. Matrix of vacant teeth

[0049]

[0050] When the above situation occurs, the identified results need to be rewritten. Rewriting methods include "squeezing" the duplicated tooth position numbers to one side to reduce the number of digits; or shifting one segment of the two missing tooth position numbers to the other end to fill the gap. In this embodiment, when duplicate identification occurs, the number of teeth on both sides of the duplicated tooth position number is compared; the side with fewer teeth is "squeezed" to that side. Taking the situation shown in Table 2 as an example, two teeth are both identified as 42, meaning that tooth number 42 is identified twice. Since only one tooth is identified on the left side, the tooth positions on the left are shifted to the left to the position of the number of times the tooth was repeatedly identified. That is, the "1" of tooth number 43 is moved to the cell of tooth number 44, and the resulting gap is filled by filling the cell of tooth number 43, which has now become "0", with "1". After this, the number of times tooth number "42" was repeatedly identified is changed to 1. The corrected tooth position matrix is ​​shown in Table 4. When a misidentification causes a gap, the number of teeth identified on both sides of the gap is compared, and the segment with fewer teeth is shifted to the segment with more teeth to fill the gap. Taking the situation shown in Table 3 as an example, due to the incorrect identification of tooth position numbers, there is a gap at tooth position number 41. Two teeth with position numbers 42 and 43 are identified on the left side of the gap, and three teeth with position numbers 31, 32 and 33 are identified on the right side of the gap. Since the probability of all three teeth being identified incorrectly is lower than the probability of two teeth being identified incorrectly at the same time, the left side with fewer teeth is moved one position to the right. That is, the number 1 filled in the cell corresponding to tooth position numbers 43 and 42 in the tooth position matrix is ​​moved one position to the right to the cell corresponding to tooth position numbers 42 and 42, as shown in Table 5.

[0051] Table 4 Tooth position matrix after correction of Table 2

[0052]

[0053] Table 5. Tooth position matrix after correction of Table 3

[0054]

[0055] Understandably, after the tooth position recognition model HM_model and the category recognition model HR_model identify the tooth position number and permanent / deciduous tooth category in the oral cavity images captured by the oral observation instrument, the recognition results need to be integrated and labeled on the oral cavity images before outputting the results. If a tooth in the same position is identified as a permanent tooth, the tooth position number identified by the tooth position recognition model HM_model can be directly output; if a tooth in the same position is identified as a deciduous tooth, the tooth position number identified by the tooth position recognition model HM_model needs to be restored to the corresponding deciduous tooth position number before being labeled on the corresponding position in the oral cavity image. Understandably, both the tooth position recognition model HM_model and the category recognition model HR_model can use neural network models such as Mask R-CNN, Faster R-CNN, or U-Net.

[0056] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the above-described method for mixed tooth position recognition of local oral images.

[0057] In some possible implementations, various aspects of the mixed tooth position recognition method for local oral images provided by the present invention can also be implemented in the form of a program product, which includes program code that, when the program product is run on a device, causes the control device to perform the steps in the mixed tooth position recognition method for local oral images according to various exemplary embodiments of the present application described above.

[0058] Reference Figure 3 Based on the same inventive concept, embodiments of the present invention also provide a recognition device for implementing the above-described method for recognizing mixed tooth positions from local oral images, including a processor and a memory, wherein the memory is electrically connected to the processor, and the processor executes the computer program stored in the memory to implement the above-described method for recognizing mixed tooth positions from local oral images.

[0059] In one possible design, the processor may include one or more processing units. The processor and memory may be implemented on the same chip or on separate chips. The processor may be a general-purpose processor, such as a central processing unit (CPU), digital signal processor, application-specific integrated circuit, field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the hybrid tooth position recognition method for local oral images disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

[0060] Memory, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory can include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic memory, magnetic disk, optical disk, etc. Memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. The memory in the embodiments of this application can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0061] By designing and programming the processor, the code corresponding to the mixed tooth position recognition method of local oral images described in the foregoing embodiments can be embedded into the chip, thereby enabling the chip to execute the steps of the mixed tooth position recognition method of local oral images shown in the embodiments of the present invention during operation. How to design and program the processor is a technique known to those skilled in the art, and will not be elaborated here.

[0062] Reference Figure 4In some possible implementations, the recognition device includes an acquisition module, a tooth position recognition model, a classification recognition model, and a correction module. The acquisition module acquires local oral images using an oral observation instrument. The tooth position recognition model identifies the tooth position numbers in the acquired local oral images. This model is trained using a tooth position annotation dataset, formed by manually annotating the tooth position numbers of each tooth in the local oral images and replacing the position numbers of deciduous teeth with the corresponding permanent tooth numbers according to the positional mapping relationship between deciduous and permanent teeth. The classification recognition model identifies the category of the teeth in the acquired local oral images. This model is trained using a classification annotation dataset, formed by manually annotating the category information of each tooth and quantifying the category information of each tooth. The correction module corrects the identified tooth position numbers that are spatially discontinuous using prior knowledge of tooth positions.

[0063] In some possible implementations, in order to improve the accuracy and stability of the tooth position recognition model, during the training process of the tooth position recognition model, the images in the tooth position annotation dataset are first randomly occluded to obtain a relative position dataset, and then the tooth position recognition model is trained using the relative position dataset.

[0064] In some possible implementations, the correction module includes a tooth position matrix corresponding to the actual spatial position of each tooth position number, a tooth position filling module, and a tooth position moving module. The tooth position filling module is used to fill the number of each tooth position number identified by the tooth position recognition model into the corresponding position in the tooth position matrix shown in Table 1. If a tooth position number in the tooth position matrix is ​​repeatedly identified, the tooth position moving module is used to compare the number of identified teeth on both sides of the repeatedly identified tooth position number, shift the segment with fewer teeth positions in the direction with fewer teeth positions to the direction with fewer teeth positions, and fill the gaps created by the shift, thus correcting the tooth position matrix shown in Table 2 to the tooth position matrix shown in Table 4. If there are gaps in the identified tooth position segments in the tooth position matrix, the tooth position moving module is used to compare the number of identified teeth on both sides of the gap, and shift the segment with fewer teeth positions to the segment with more teeth positions to fill the gaps, thus correcting the tooth position matrix shown in Table 3 to the tooth position matrix shown in Table 5.

[0065] It should be noted that this application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A method of identifying a mixed tooth site of a partial oral image, characterized by, The method comprises the following steps: Collecting a large number of clinical oral local images, manually labeling the tooth position number and the category information of the permanent teeth or deciduous teeth in each tooth in the oral local images; Replacing the tooth position number of the deciduous teeth in the labeled oral local images with the corresponding permanent tooth number according to the position mapping relationship between the deciduous teeth and the permanent teeth, and deleting the category information of each tooth to form a tooth position labeling data set; Deleting the tooth position number in the manually labeled oral local images and numerically representing the category information of each tooth to form a classification labeling data set; Training a model through the tooth position labeling data set to obtain a tooth position recognition model for the relative position information of the concerned teeth; Training a model through the classification labeling data set to obtain a classification recognition model for the category information of the concerned teeth; Obtaining an oral image to be recognized, recognizing the tooth position number of the teeth in the oral image through the tooth position recognition model, and recognizing the category of the teeth in the oral image through the classification recognition model; Correcting the recognized tooth position number with discontinuous spatial positions by using the prior knowledge of the tooth positions; The step of correcting the recognized tooth position number with discontinuous spatial positions by using the prior knowledge of the tooth positions comprises: Establishing a tooth position matrix corresponding to the actual spatial positions of each tooth position number; Filling the number of each tooth position number recognized from the oral image into the corresponding position of the tooth position matrix; If there is a repeated recognition of a tooth position number in the tooth position matrix, comparing the number of the recognized tooth positions on both sides of the repeated tooth position number, translating the section with fewer tooth positions to the direction of the section with more tooth positions by the number of the repeated tooth position, and filling the gap generated after the translation; If there is a gap in the recognized tooth section in the tooth position matrix, comparing the number of the recognized tooth positions on both sides of the gap, and translating the section with fewer tooth positions to the section with more tooth positions to fill the gap; The step of training a model through the tooth position labeling data set to obtain a tooth position recognition model for the relative position information of the concerned teeth comprises: Obtaining a relative position data set by randomly masking the images in the tooth position labeling data set; Training a model using the relative position data set to obtain a tooth position recognition model for the relative position information of the concerned teeth; The random masking refers to randomly setting half of the pixels of the teeth in the image to zero, but still keeping the tooth position number of the teeth with the pixels set to zero.

2. A storage medium storing a computer program, characterized by The computer program is called and executed by the processor to implement the mixed tooth position recognition method of the local oral image according to claim 1.

3. A mixed tooth site recognition apparatus of a partial oral image, characterized by, The computer program is called and executed by the processor to implement the mixed tooth position recognition method of the local oral image according to claim 1.

4. A mixed tooth site recognition apparatus of a partial oral image, characterized by, The method comprises: An acquisition module configured to acquire a local oral image of an oral cavity; The tooth position recognition model is trained by a tooth position annotation data set formed by using artificial annotation of tooth position numbers of each tooth in a partial oral cavity image and replacing tooth position numbers of deciduous teeth in the partial oral cavity image into corresponding permanent tooth numbers according to a position mapping relationship between the deciduous teeth and the permanent teeth. The classification recognition model is trained by a classification annotation data set formed by using artificial annotation of permanent tooth or deciduous tooth classification information of each tooth and numerical value of the classification information of each tooth. The correction module is used for correcting the discontinuous tooth position numbers recognized by using prior knowledge of the tooth position. The correction module includes a tooth position matrix corresponding to actual spatial positions of each tooth number, a tooth position filling module and a tooth position moving module. The tooth position filling module is used for filling the number of each tooth number recognized by the tooth position recognition model into a corresponding position of the tooth position matrix. If there is a repeated recognition of a tooth number in the tooth position matrix, the tooth position moving module is used for comparing the number of recognized tooth positions on both sides of the repeated tooth number, moving a tooth position with less number of tooth positions to a direction with more number of tooth positions by a number of repeated recognitions, and filling a vacancy generated after the moving. If there is a vacancy in a recognized tooth segment in the tooth position matrix, the tooth position moving module is used for comparing the number of recognized tooth positions on both sides of the vacancy, and moving a tooth position with less number of tooth positions to a tooth position with more number of tooth positions to fill the vacancy. In the training process of the tooth position recognition model, the image in the tooth position annotation data set is randomly occluded to obtain a relative position data set, and the tooth position recognition model is trained by using the relative position data set. The random occlusion refers to randomly setting pixels of half of the teeth in the image to zero, but still keeping the tooth numbers of the teeth with the pixels set to zero.

Images