Deep learning-based children's deciduous permanent tooth replacement period anomaly analysis system

Abstract

The invention relates to the general technical field of artificial intelligence, in particular to a children's deciduous permanent tooth replacement period anomaly analysis system based on deep learning, which comprises an image acquisition module, an image screening module, an image processing module, a deciduous permanent tooth prediction module and an anomaly detection module, the image screening module is used for screening target images, the image processing module is used for acquiring analog images, the deciduous permanent tooth prediction module is used for acquiring deciduous permanent tooth detection results, and the anomaly detection module is used for acquiring deciduous permanent tooth anomaly conditions. According to the invention, deep learning and anomaly analysis are carried out on the child tooth image through cooperation of the modules, so that detection and anomaly risk assessment are carried out on the child deciduous permanent teeth through the child tooth mobile phone image, and the child oral image processing efficiency and the child deciduous permanent tooth anomaly analysis efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the general technical field of artificial intelligence, and in particular to a deep learning-based abnormality analysis system for children's deciduous and permanent teeth replacement period. Background Art

[0002] Traditional monitoring of children's dental development usually relies on professional oral examination equipment, which is bulky, expensive, and requires professional operation. It can only be performed in hospitals or professional dental clinics. This makes it difficult for parents to monitor their children's dental development in a timely and frequent manner in daily life. Although mobile phone photography has become popular as a non-invasive collection method, oral images taken by mobile phones are affected by many factors and the quality is uneven: poor lighting conditions during shooting may cause the image to be too dark, too bright, or shadows to appear, affecting the observation of dental details; there are few existing analysis models for children's dental images taken by mobile phones, and most of them are not perfect. On the one hand, due to the large differences in resolution, imaging principles, etc. between mobile phone images and images obtained by professional medical equipment, directly applying models trained on professional medical images to mobile phone image analysis is not effective.

[0003] Chinese Patent Publication No. CN117918871B discloses a method and system for predicting the growth of children's deciduous teeth. The method includes extracting tooth edges from dental X-ray images, determining the central axis and theoretical growth area of ​​permanent teeth based on the tooth edges, and determining the co-periodic and non-co-periodic permanent teeth of each permanent tooth according to the natural eruption pattern of teeth. This solution still cannot detect and assess the risk of abnormalities in children's deciduous and permanent teeth based on mobile phone images of children's teeth, resulting in low efficiency in children's oral image processing and low efficiency in abnormal analysis of children's deciduous and permanent teeth. Summary of the Invention

[0004] To this end, the present invention provides a deep learning-based abnormality analysis system for children's deciduous and permanent teeth replacement period, which is used to overcome the problem that the existing technology is unable to detect and assess abnormality risks of children's deciduous and permanent teeth based on mobile phone images of children's teeth, resulting in low efficiency in children's oral image processing and low efficiency in abnormality analysis of children's deciduous and permanent teeth.

[0005] To achieve the above objectives, the present invention provides a deep learning-based abnormality analysis system for children's deciduous and permanent teeth replacement period, the system comprising: An image acquisition module, used to acquire target images; An image screening module is used to screen the target image and obtain a usable target image; an image processing module, configured to perform image extraction processing on the mobile phone image of children's teeth in the target available image according to the mobile phone image processing method to obtain a processed mobile phone image of children's teeth, further configured to adjust the pixel values ​​of the mobile phone image processing method, further configured to perform migration adjustment on the pixel value adjustment process, and further configured to perform professional image migration on the professional image of children's teeth in the target available image according to the processed mobile phone image of children's teeth to obtain a simulated image; The deciduous and permanent teeth prediction module is used to obtain the deciduous and permanent teeth detection results based on the simulated image, obtain the deciduous and permanent teeth detection results, and also to update the results of the deciduous and permanent teeth detection result acquisition process; The abnormality detection module is used to obtain the abnormal conditions of deciduous and permanent teeth based on the detection results of deciduous and permanent teeth, and to obtain the abnormality risk level based on the abnormal conditions of deciduous and permanent teeth, and to adjust the level of the abnormality risk level acquisition process.

[0006] Furthermore, when the image screening module screens the target image, the target image gradient amplitude ERU is compared with the preset gradient amplitude ERU0, 12≤ERU0≤15 is set, and the target image is judged according to the comparison result to determine whether it meets the standard, and the target usable image is output according to the judgment result, wherein: When ERU≥ERU0, the image screening module determines that the target image meets the standards and outputs the target image as a target available image; When ERU<ERU0, the image screening module determines that the target image does not meet the standards, does not output the target image as a target available image, and removes the target image; When the image screening module screens the target image, it compares the image entropy value ERB with the preset image entropy value ERB0, sets 6.5≤ERB0≤7.5, judges whether the image entropy value meets the standard based on the comparison result, and adjusts the clarity of the preset gradient amplitude ERU0 based on the judgment result, wherein: When ERB≥ERB0, the image screening module determines that the image entropy value meets the standard and does not adjust the clarity of the preset gradient amplitude ERU0; When ERB<ERB0, the image screening module determines that the image entropy value does not meet the standard, and adjusts the clarity of the preset gradient amplitude ERU0 according to the clarity adjustment coefficient asx, setting asx=1.35-0.15×e -0.69×（ERB-ERB0）, where e is the base of the natural logarithm, and the adjusted preset gradient amplitude ERU0` is obtained. ERU0`=ERU0×asx is set, and the preset gradient amplitude ERU0 is replaced with the adjusted preset gradient amplitude ERU0`. The target image gradient amplitude is then re-compared with the adjusted preset gradient amplitude ERU0`.

[0007] Furthermore, the image processing module performs image extraction processing on the target available image according to a mobile phone image processing method, and the mobile phone image processing method includes: Step A01, dividing the mobile phone image of children's teeth into a set of rectangular sub-blocks aol={aol1, aol2, aol3, ..., aolh}; Step A02, obtaining sub-block pixel grayscale values ​​aom={aom1, aom2, aom3, ..., aomh} according to the rectangular sub-block set aol={aol1, aol2, aol3, ..., aolh}; Step A03: Compare the first sub-block pixel grayscale value aom1 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the first sub-block pixel grayscale value based on the comparison result, and output the first rectangular sub-block attribute aok1 based on the judgment result, where: When aom1≤ao1, the image processing module determines that the grayscale value of the first sub-block pixel is low, and outputs the enamel reflective sub-block as the first rectangular sub-block attribute aok1; When ao1<aom1≤ao2, the image processing module determines that the grayscale value of the first sub-block pixel is moderate, and outputs the enamel essence sub-block as the first rectangular sub-block attribute aok1; When aom1>ao2, the image processing module determines that the grayscale value of the first sub-block pixel is high, and outputs the tooth gap shadow sub-block as the first rectangular sub-block attribute aok1; Compare the second sub-block pixel grayscale value aom2 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the second sub-block pixel grayscale value based on the comparison result, and output the second rectangular sub-block attribute aok2 based on the judgment result, where: When aom2≤ao1, the image processing module determines that the grayscale value of the second sub-block pixel is low, and outputs the enamel reflective sub-block as the second rectangular sub-block attribute aok2; When ao1<aom2≤ao2, the image processing module determines that the grayscale value of the second sub-block pixel is moderate, and outputs the enamel essence sub-block as the second rectangular sub-block attribute aok2; When aom2>ao2, the image processing module determines that the grayscale value of the second sub-block pixel is high, and outputs the tooth gap shadow sub-block as the second rectangular sub-block attribute aok2; Compare the third sub-block pixel grayscale value aom3 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the third sub-block pixel grayscale value based on the comparison result, and output the third rectangular sub-block attribute aok3 based on the judgment result, where: When aom3≤ao1, the image processing module determines that the grayscale value of the third sub-block pixel is low, and outputs the enamel reflective sub-block as the third rectangular sub-block attribute aok3; When ao1<aom3≤ao2, the image processing module determines that the grayscale value of the third sub-block pixel is moderate, and outputs the enamel essence sub-block as the third rectangular sub-block attribute aok3; When aom3>ao2, the image processing module determines that the grayscale value of the third sub-block pixel is high, and outputs the tooth gap shadow sub-block as the third rectangular sub-block attribute aok3; … Compare the hth sub-block pixel grayscale value aomh with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the hth sub-block pixel grayscale value based on the comparison result, and output the hth rectangular sub-block attribute aokh based on the judgment result, where: When aomh≤ao1, the image processing module determines that the grayscale value of the h-th sub-block pixel is low, and outputs the enamel reflective sub-block as the h-th rectangular sub-block attribute aokh; When ao1<aomh≤ao2, the image processing module determines that the grayscale value of the h-th sub-block pixel is moderate, and outputs the enamel intrinsic sub-block as the h-th rectangular sub-block attribute aokh; When aomh>ao2, the image processing module determines that the grayscale value of the h-th sub-block pixel is high, and outputs the tooth gap shadow sub-block as the h-th rectangular sub-block attribute aokh; Step A04: Output the first rectangular sub-block attribute aok1, the second rectangular sub-block attribute aok2, the third rectangular sub-block attribute aok3, ..., the hth rectangular sub-block attribute aokh as the light and dark annotated mobile phone image of the child's teeth; Step A05, performing specific processing on the light and dark marked mobile phone image of the child's teeth to obtain a processed mobile phone image of the child's teeth; When the image processing module adjusts the pixel value of the mobile phone image processing method according to the image extraction accuracy, the image extraction accuracy IoU is compared with the preset model accuracy IoU0, and 0.85≤IoU0≤0.90 is set. The compliance of the image extraction accuracy is judged according to the comparison result, and the pixel value of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2 is adjusted according to the judgment result, wherein: When IoU≥IoU0, the image processing module determines that the image extraction accuracy rate meets the standard, and does not adjust the pixel values ​​of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2; When IoU<IoU0, the image processing module determines that the image extraction accuracy rate does not meet the standard, and adjusts the pixel value of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2. The pixel value of the first preset pixel grayscale value ao1 is adjusted according to the pixel value adjustment coefficient au, and au=0.77+0.14×e -0.7×（IoU0-IoU） , where e is the base of the natural logarithm, obtain the first preset pixel grayscale value ao1` after pixel value adjustment, set ao1`=ao1×au, replace the first preset pixel grayscale value ao1 with the first preset pixel grayscale value ao1` after pixel value adjustment, adjust the second preset pixel grayscale value ao2 according to the pixel value adjustment coefficient au, obtain the second preset pixel grayscale value ao2` after pixel value adjustment, set ao2`=ao2×au, replace the second preset pixel grayscale value ao2 with the second preset pixel grayscale value ao2` after pixel value adjustment, and re-compare the first rectangular sub-block attribute aok1, the second rectangular sub-block attribute aok2, the third rectangular sub-block attribute aok3,..., the hth rectangular sub-block attribute aokh with the first preset pixel grayscale value ao1` and the second preset pixel grayscale value ao2` after pixel value adjustment.

[0008] Furthermore, when the image processing module performs migration adjustment on the pixel value adjustment process according to the number of remote diagnoses, the migration factor Yj is calculated using the number of remote diagnoses Yd and the number of cases in the new scenario Yx to obtain the migration factor Yj, and Yj=0.45×Yd+0.55×Yx is set. The migration factor Yj is compared with the preset migration factor Yj0, and 0.56≤Yj0≤0.78 is set. The compliance of the migration factor is judged based on the comparison result, and the preset model accuracy IoU0 is migrated and adjusted based on the judgment result, wherein: When Yj≥Yj0, the image processing module determines that the migration factor meets the standard and does not perform migration adjustment on the preset model accuracy IoU0; When Yj<Yj0, the image processing module determines that the migration factor is not up to standard, and performs migration adjustment on the preset model accuracy IoU0. The preset model accuracy IoU0 is adjusted according to the migration coefficient yu, and yu=1.31-0.16×e -0.61×（Yj0-Yj） , obtain the adjusted preset model accuracy IoU0`, set IoU0`=IoU0×yu, replace the preset model accuracy IoU0 with the adjusted preset model accuracy IoU0`, and re-compare the image extraction accuracy IoU with the adjusted preset model accuracy IoU0`.

[0009] Furthermore, the image processing module performs professional image migration on the professional image of children's teeth based on the processed mobile phone image of children's teeth by a professional image migration method, and the professional image migration method includes: Step B01, initialize the network parameters of the generator and the discriminator to obtain the initialized generator and the initialized discriminator; Step B02: Input the professional image of children's teeth into the initialized generator, extract the illumination features through the dynamic illumination compensation layer, and generate an illumination mask based on the illumination features; Step B03, performing illumination compensation on the professional image of children's teeth according to the illumination mask to obtain the professional image of children's teeth after illumination compensation; Step B04, performing upsampling and convolution operations on the professional children's dental image after illumination compensation to obtain a preliminary simulated image; In step B05, the preliminary simulation image and the processed children's dental mobile phone image are input into the initialized discriminator, and a similarity value FK is output through multi-layer convolution. The preliminary simulation image with a similarity value of 90% is output as the simulation image.

[0010] Furthermore, when the deciduous and permanent teeth detection module obtains the deciduous and permanent teeth detection results according to the simulated image, the deciduous and permanent teeth detection model is constructed according to the deciduous and permanent teeth detection model construction method, and the deciduous and permanent teeth detection model construction method includes: Step J01, dividing the historical simulation training set into 50% deciduous and permanent tooth position annotation frames and 50% caries status annotation frames; Step J02, performing bias initialization on the EfficientNet-B4 detection model to obtain an initialized EfficientNet-B4 detection model; Step J03: Input the labeled boxes of the deciduous and permanent teeth positions and the labeled boxes of the caries plaque status into the initialized EfficientNet-B4 detection model for labeling training, and obtain the Intersection-Union (IU) ratio between the trained EfficientNet-B4 detection model and the target. Step J04, performing loss optimization on the trained EfficientNet-B4 detection model according to the bounding box loss function Lc; Step J05: compare the target intersection-and-union ratio IU with the preset target intersection-and-union ratio IU0, set 0.5≤IU0≤0.7, judge whether the target intersection-and-union ratio meets the standard based on the comparison result, and output the deciduous and permanent teeth detection model based on the judgment result, where: When IU≥IU0, the deciduous and permanent teeth detection module determines that the target intersection-over-union ratio meets the standard, and outputs the trained EfficientNet-B4 detection model as the deciduous and permanent teeth detection model; When IU<IU0, the deciduous and permanent teeth detection module determines that the target intersection-over-union ratio is not up to standard, and adjusts the loss of the bounding box loss function Lc according to the loss adjustment coefficient lm, setting lm=1.22-0.16×e -0.7×（IU0-IU） , obtain the adjusted bounding box loss function Lc`, set Lc`=Lc×lm, replace the bounding box loss function Lc with the adjusted bounding box loss function Lc`, and re-optimize the loss of the trained EfficientNet-B4 detection model according to the adjusted bounding box loss function Lc` until the target intersection-union ratio meets the standard; The processed mobile phone image of children's teeth is input into the deciduous and permanent teeth detection model to obtain the deciduous and permanent teeth detection results.

[0011] Furthermore, when the deciduous and permanent teeth detection module updates the result of the acquisition process of the deciduous and permanent teeth detection results according to the number of teeth, the number of teeth Ys is compared with the preset number of teeth Ys0, and 10≤Ys0≤16 is set. The state of the number of teeth is judged according to the comparison result, and the preset target intersection and union ratio IU0 is updated according to the judgment result, wherein: When Ys≥Ys0, the deciduous and permanent teeth detection module determines that the state of the number of teeth is large, and does not update the result of the preset target intersection and union ratio IU0; When Ys<Ys0, the deciduous and permanent teeth detection module determines that the number of teeth is small, updates the preset target intersection-and-union ratio IU0, and updates the preset target intersection-and-union ratio IU0 according to the result update coefficient yo, setting yo=1.37+0.14×e -1.24×（Ys / Ys0） , e is the base of the natural logarithm, and the updated preset target intersection-union ratio IU0` is obtained. IU0`=IU0×yo is set, and the preset target intersection-union ratio IU0 is replaced by the updated preset target intersection-union ratio IU0`, and the target intersection-union ratio IU is re-compared with the updated preset target intersection-union ratio IU0`.

[0012] Furthermore, when the abnormality detection module obtains abnormal conditions of deciduous and permanent teeth according to the detection results of deciduous and permanent teeth, the detection results of deciduous and permanent teeth are input into the deciduous and permanent teeth prediction model to obtain the deciduous tooth shedding time, permanent tooth eruption time and deciduous tooth unchanged time; When the abnormality detection module obtains the abnormality of the deciduous and permanent teeth according to the detection results of the deciduous and permanent teeth, the deciduous tooth shedding time Rt is compared with the first preset deciduous tooth shedding time Rt1 and the second preset deciduous tooth shedding time Rt2, and Rt1 is set to 3 months and Rt2 to 6 months. The deciduous tooth shedding time is judged according to the comparison result, and the deciduous tooth condition is output according to the judgment result, wherein: When Rt≤Rt1, the abnormality detection module determines that the deciduous tooth shedding time is abnormal and outputs the deciduous tooth abnormality as the deciduous tooth condition; When Rt1<Rt≤Rt2, the abnormality detection module determines that the deciduous tooth shedding time is normal and outputs the normal deciduous tooth as the deciduous tooth condition; When Rt>Rt2, the abnormality detection module determines that the deciduous tooth shedding time is abnormal and outputs the deciduous tooth abnormality as the deciduous tooth condition; The abnormality detection module compares the permanent tooth eruption time Rh with the first preset permanent tooth eruption time Rh1 and the second preset permanent tooth eruption time Rh2, sets Rh1=3 months and Rh2=6 months, judges the permanent tooth eruption time according to the comparison results, and outputs the permanent tooth status according to the judgment results, wherein: When Rh≤Rh1, the abnormality detection module determines that the permanent tooth eruption time is abnormal and outputs the permanent tooth abnormality as the permanent tooth condition; When Rh1<Rh≤Rh2, the abnormality detection module determines that the permanent tooth eruption time is normal and outputs the normal permanent tooth as the permanent tooth condition; When Rh>Rh2, the abnormality detection module determines that the permanent tooth eruption time is abnormal and outputs the permanent tooth abnormality as the permanent tooth condition; The abnormality detection module compares the unchanged time of deciduous teeth Rrw with the preset unchanged time Rrw0, sets Rrw0=2 years, judges the unchanged time of deciduous teeth according to the comparison result, and outputs the unchanged situation of deciduous teeth according to the judgment result, wherein: When Rrw≤Rrw0, the abnormality detection module determines that the situation of no change in deciduous teeth is normal, and outputs the normal situation of no change in deciduous teeth as the situation of no change in deciduous teeth; When Rrw>Rrw0, the abnormality detection module determines that the situation of no change in deciduous teeth is abnormal, and outputs the abnormality of no change in deciduous teeth as the situation of no change in deciduous teeth.

[0013] The deciduous teeth condition, permanent teeth condition and no change in deciduous teeth condition are output as abnormal conditions of deciduous and permanent teeth.

[0014] Furthermore, when the abnormality detection module obtains the abnormality risk level according to the abnormal conditions of the deciduous and permanent teeth, the abnormality risk score Rp is obtained according to the deciduous tooth abnormality parameter R1, the permanent tooth abnormality parameter R2, the unchanged abnormality parameter R3, the deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2 and the unchanged abnormality weight β3, and Rp=β1×R1+β2×R2+β3×R3 is set to obtain the abnormality risk score Rp. The abnormality risk score Rp is compared with the preset abnormality risk score Rp0, and the abnormality risk degree is judged according to the comparison result, and the abnormality risk level is output according to the judgment result, wherein: When Rp≤Rp0, the anomaly detection module determines that the abnormal risk level is normal and outputs no risk as the abnormal risk level; When Rp>Rp0, the anomaly detection module determines that the abnormal risk degree is abnormal and outputs risk as the abnormal risk level; Furthermore, when the abnormality detection module adjusts the abnormality risk level acquisition process according to the child's age, the child's age Nr is compared with the preset child age Nr0, and the child's age is judged according to the comparison result. The deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3 are adjusted according to the judgment result, wherein: When Nr≤Nr0, the anomaly detection module determines that the child is young and does not adjust the level of the deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3; When Nr>Nr0, the abnormality detection module determines that the child is old, and adjusts the grade of the deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2, and the unchanged abnormality weight β3. The deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3 are adjusted according to the grade adjustment coefficient uop, and 1.1<uop<1.4 is set to obtain the adjusted deciduous tooth abnormality weight β1` and the adjusted unchanged abnormality weight β3`, and β1`=β1×uop, β3`=(uop-1)×β3 are set. The permanent tooth abnormality weight β2 is graded and β2=1-β1-β3 to obtain the adjusted permanent tooth abnormality weight β2`. The deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2, and the changed abnormality weight β3 are replaced with the adjusted deciduous tooth abnormality weight β1`, the adjusted permanent tooth abnormality weight β2`, and the adjusted unchanged abnormality weight β3`, and the abnormality risk score is recalculated.

[0015] Compared with the existing technology, the beneficial effect of the present invention is that the various modules cooperate to perform deep learning and abnormal analysis on children's dental images, so as to detect and assess the abnormal risks of children's deciduous and permanent teeth through children's dental mobile phone images, thereby improving the efficiency of children's oral image processing. Among them, the children's deciduous and permanent teeth replacement period abnormality analysis system based on deep learning collects target images through the image acquisition module, so as to obtain the target available image according to the target image, so as to perform abnormal analysis on children's deciduous and permanent teeth, thereby improving the efficiency of abnormal analysis. The children's deciduous and permanent teeth replacement period abnormality analysis system based on deep learning also filters the target image through the image screening module to ensure that the quality of the target image entering the subsequent processing meets the requirements. The analysis requirements are met, thereby improving the efficiency of image data processing. The abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning also acquires simulated images through the image processing module, so as to perform abnormal analysis on the deciduous and permanent teeth of children according to the simulated images, thereby improving the efficiency of children's oral image processing. The abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning also acquires the detection results of deciduous and permanent teeth, so as to detect the deciduous and permanent teeth of children, thereby improving the accuracy of deciduous and permanent teeth detection. The abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning also acquires the abnormal risk level, so as to detect the deciduous and permanent teeth of children and conduct abnormal risk assessment, thereby improving the efficiency of abnormal analysis of deciduous and permanent teeth of children. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the structure of the abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning in this embodiment. DETAILED DESCRIPTION

[0017] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0018] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside", and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0020] See also Figure 1As shown in FIG, which is a structural diagram of the abnormal analysis system for the replacement period of primary and permanent teeth of children based on deep learning in this embodiment, the system includes: An image acquisition module, used to acquire target images; An image screening module is used to screen the target image and obtain a usable target image; an image processing module, configured to perform image extraction processing on the mobile phone image of children's teeth in the target available image according to the mobile phone image processing method to obtain a processed mobile phone image of children's teeth, further configured to adjust the pixel values ​​of the mobile phone image processing method, further configured to perform migration adjustment on the pixel value adjustment process, and further configured to perform professional image migration on the professional image of children's teeth in the target available image according to the processed mobile phone image of children's teeth to obtain a simulated image. The image processing module is connected to the image acquisition module; The deciduous and permanent teeth prediction module is used to obtain the deciduous and permanent teeth detection results based on the simulated image, obtain the deciduous and permanent teeth detection results, and is also used to update the results of the deciduous and permanent teeth detection result acquisition process. The deciduous and permanent teeth prediction module is connected to the image processing module; The abnormality detection module is used to obtain the abnormal conditions of deciduous and permanent teeth based on the detection results of deciduous and permanent teeth, and to obtain the abnormal risk level based on the abnormal conditions of deciduous and permanent teeth. It is also used to adjust the level of the abnormal risk level acquisition process. The abnormality detection module is connected to the deciduous and permanent teeth prediction module.

[0021] Specifically, the abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning is applied to the detection terminal of deciduous and permanent teeth of children, and the modules cooperate to perform deep learning and abnormal analysis on the images of children's teeth, so as to realize the detection and abnormal risk assessment of deciduous and permanent teeth of children through the mobile phone images of children's teeth, thereby improving the efficiency of children's oral image processing and the efficiency of abnormal analysis of deciduous and permanent teeth of children. Among them, the abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning collects the target image through the image acquisition module, so as to obtain the target available image according to the target image, so as to perform abnormal analysis on the deciduous and permanent teeth of children, thereby improving the efficiency of abnormal analysis. The abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning also screens the target image through the image screening module. Selection, ensuring that the quality of the target image entering subsequent processing meets the analysis requirements, thereby improving the efficiency of image data processing, the abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning also obtains simulated images through the image processing module, so as to facilitate abnormal analysis of the deciduous and permanent teeth of children based on the simulated images, thereby improving the efficiency of children's oral image processing, the abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning also obtains the detection results of deciduous and permanent teeth, so as to facilitate the detection of deciduous and permanent teeth of children, thereby improving the accuracy of deciduous and permanent teeth detection, the abnormal analysis system for the replacement period of deciduous and permanent teeth of children based on deep learning also obtains the abnormal risk level, so as to facilitate the detection and abnormal risk assessment of deciduous and permanent teeth of children, thereby improving the efficiency of abnormal analysis of deciduous and permanent teeth of children.

[0022] Specifically, the image acquisition module acquires the target image, which includes professional images of children's teeth and mobile phone images of children's teeth. The professional images of children's teeth refer to high-quality images that reflect the condition of children's teeth using professional oral medical equipment. This embodiment does not limit the specific acquisition method of professional images of children's teeth. Relevant technical personnel in this field can freely choose according to actual needs, such as acquiring through professional oral medical equipment. The mobile phone images of children's teeth refer to images of children's teeth taken using a mobile phone. This embodiment does not limit the acquisition method of mobile phone images of children's teeth. Relevant technical personnel in this field can freely choose according to actual needs, such as manual shooting.

[0023] Specifically, the image acquisition module acquires the target image so as to obtain the target usable image later, so as to perform abnormality analysis on the deciduous and permanent teeth of children and improve the efficiency of abnormality analysis.

[0024] Specifically, when the image screening module screens the target image, it compares the target image gradient amplitude ERU with the preset gradient amplitude ERU0, sets 12≤ERU0≤15, judges the compliance of the target image based on the comparison result, and outputs the target usable image based on the judgment result, wherein: When ERU≥ERU0, the image screening module determines that the target image meets the standards and outputs the target image as a target available image; When ERU<ERU0, the image screening module determines that the target image does not meet the standards, does not output the target image as a target available image, and removes the target image; When the image screening module screens the target image, it compares the image entropy value ERB with the preset image entropy value ERB0, sets 6.5≤ERB0≤7.5, judges whether the image entropy value meets the standard based on the comparison result, and adjusts the clarity of the preset gradient amplitude ERU0 based on the judgment result, wherein: When ERB≥ERB0, the image screening module determines that the image entropy value meets the standard and does not adjust the clarity of the preset gradient amplitude ERU0; When ERB<ERB0, the image screening module determines that the image entropy value does not meet the standard, and adjusts the clarity of the preset gradient amplitude ERU0 according to the clarity adjustment coefficient asx, setting asx=1.35-0.15×e -0.69×（ERB-ERB0） , where e is the base of the natural logarithm, and the adjusted preset gradient amplitude ERU0` is obtained. ERU0`=ERU0×asx is set, and the preset gradient amplitude ERU0 is replaced with the adjusted preset gradient amplitude ERU0`. The target image gradient amplitude is then re-compared with the adjusted preset gradient amplitude ERU0`.

[0025] Specifically, the target image gradient amplitude refers to a value used to reflect the degree of drastic change in pixel grayscale values ​​in the target image. This embodiment does not limit the specific method for obtaining the target image gradient amplitude. Relevant technicians in this field can freely choose according to actual needs, such as software acquisition. The preset gradient amplitude refers to a preset value for judging the compliance of the target image. The compliance of the target image refers to the degree of compliance of the target image quality. The compliance of the target image includes the compliance of the target image as being up to standard and the compliance of the target image as being not up to standard. The image entropy value refers to a value reflecting the information richness of the target image and the uniformity of the pixel grayscale distribution. This embodiment does not limit the specific method for obtaining the image entropy value. Relevant technicians in this field can freely choose according to actual needs, such as software acquisition. The preset image entropy value refers to a preset value for judging the compliance of the image entropy value. The compliance of the image entropy value refers to the degree of compliance of the image entropy value. The compliance of the image entropy value includes the compliance of the image entropy value as being up to standard and the compliance of the image entropy value as being not up to standard.

[0026] Specifically, the image screening module screens the target image to ensure that the quality of the target image entering subsequent processing meets the analysis requirements, thereby improving the efficiency of image data processing.

[0027] Specifically, the image processing module performs image extraction processing on the target available image according to a mobile phone image processing method, and the mobile phone image processing method includes: Step A01, dividing the mobile phone image of children's teeth into a set of rectangular sub-blocks aol={aol1, aol2, aol3, ..., aolh}; Step A02, obtaining sub-block pixel grayscale values ​​aom={aom1, aom2, aom3, ..., aomh} according to the rectangular sub-block set aol={aol1, aol2, aol3, ..., aolh}; Step A03: Compare the first sub-block pixel grayscale value aom1 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the first sub-block pixel grayscale value based on the comparison result, and output the first rectangular sub-block attribute aok1 based on the judgment result, where: When aom1≤ao1, the image processing module determines that the grayscale value of the first sub-block pixel is low, and outputs the enamel reflective sub-block as the first rectangular sub-block attribute aok1; When ao1<aom1≤ao2, the image processing module determines that the grayscale value of the first sub-block pixel is moderate, and outputs the enamel essence sub-block as the first rectangular sub-block attribute aok1; When aom1>ao2, the image processing module determines that the grayscale value of the first sub-block pixel is high, and outputs the tooth gap shadow sub-block as the first rectangular sub-block attribute aok1; Compare the second sub-block pixel grayscale value aom2 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the second sub-block pixel grayscale value based on the comparison result, and output the second rectangular sub-block attribute aok2 based on the judgment result, where: When aom2≤ao1, the image processing module determines that the grayscale value of the second sub-block pixel is low, and outputs the enamel reflective sub-block as the second rectangular sub-block attribute aok2; When ao1<aom2≤ao2, the image processing module determines that the grayscale value of the second sub-block pixel is moderate, and outputs the enamel essence sub-block as the second rectangular sub-block attribute aok2; When aom2>ao2, the image processing module determines that the grayscale value of the second sub-block pixel is high, and outputs the tooth gap shadow sub-block as the second rectangular sub-block attribute aok2; Compare the third sub-block pixel grayscale value aom3 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the third sub-block pixel grayscale value based on the comparison result, and output the third rectangular sub-block attribute aok3 based on the judgment result, where: When aom3≤ao1, the image processing module determines that the grayscale value of the third sub-block pixel is low, and outputs the enamel reflective sub-block as the third rectangular sub-block attribute aok3; When ao1<aom3≤ao2, the image processing module determines that the grayscale value of the third sub-block pixel is moderate, and outputs the enamel essence sub-block as the third rectangular sub-block attribute aok3; When aom3>ao2, the image processing module determines that the grayscale value of the third sub-block pixel is high, and outputs the tooth gap shadow sub-block as the third rectangular sub-block attribute aok3; … Compare the hth sub-block pixel grayscale value aomh with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the hth sub-block pixel grayscale value based on the comparison result, and output the hth rectangular sub-block attribute aokh based on the judgment result, where: When aomh≤ao1, the image processing module determines that the grayscale value of the h-th sub-block pixel is low, and outputs the enamel reflective sub-block as the h-th rectangular sub-block attribute aokh; When ao1<aomh≤ao2, the image processing module determines that the grayscale value of the h-th sub-block pixel is moderate, and outputs the enamel intrinsic sub-block as the h-th rectangular sub-block attribute aokh; When aomh>ao2, the image processing module determines that the grayscale value of the h-th sub-block pixel is high, and outputs the tooth gap shadow sub-block as the h-th rectangular sub-block attribute aokh; Step A04: Output the first rectangular sub-block attribute aok1, the second rectangular sub-block attribute aok2, the third rectangular sub-block attribute aok3, ..., the hth rectangular sub-block attribute aokh as the light and dark annotated mobile phone image of the child's teeth; Step A05: performing specific processing on the mobile phone image of the child's teeth after light and dark marking to obtain a processed mobile phone image of the child's teeth.

[0028] Specifically, the rectangular sub-block set aol={aol1, aol2, aol3, ..., aolh} refers to an image set obtained by dividing the children's teeth mobile phone image according to a preset size, and the preset size is set to 8×8 Pixels, where aol1 is the first image in the rectangular sub-block collection, aol2 is the second image in the rectangular sub-block collection, aol3 is the third image in the rectangular sub-block collection, aolh is the h-th image in the rectangular sub-block collection, and h is the order of the images in the rectangular sub-block collection. This embodiment does not limit the specific method of dividing the children's dental mobile phone images. Relevant technical personnel in this field can freely choose according to actual needs, such as through software division. The sub-block pixel grayscale value aom={aom1, aom2, aom3, ..., aomh} refers to a value used to reflect the brightness and darkness of each image in the rectangular sub-block collection, where aom1 is the first sub-block pixel grayscale value, and the first sub-block pixel grayscale value refers to the pixel grayscale value of the first image in the rectangular sub-block collection, aom2 is the second sub-block pixel grayscale value, and the second sub-block pixel grayscale value refers to the pixel grayscale value of the first image in the rectangular sub-block collection. The pixel grayscale value of the second image, aom3 is the pixel grayscale value of the third sub-block, and the pixel grayscale value of the third sub-block refers to the pixel grayscale value of the third image in the rectangular sub-block collection. aomh is the pixel grayscale value of the h-th sub-block, and the pixel grayscale value of the h-th sub-block refers to the pixel grayscale value of the h-th image in the rectangular sub-block collection. This embodiment does not limit the specific method for obtaining the sub-block pixel grayscale values. Relevant technicians in this field can freely choose a method based on actual needs, such as manual statistics. The first preset pixel grayscale value refers to the preset lower limit value for judging the pixel grayscale value of the first sub-block, the pixel grayscale value of the second sub-block, the pixel grayscale value of the third sub-block, ..., and the pixel grayscale value of the h-th sub-block. The second preset pixel grayscale value refers to the pixel grayscale value of the first sub-block, the pixel grayscale value of the second sub-block, the pixel grayscale value of the third sub-block, ..., the preset upper limit value for judging the situation of the h-th sub-block pixel grayscale value, the situation of the first sub-block pixel grayscale value refers to the high and low degree of the first sub-block pixel grayscale value, and the situation of the first sub-block pixel grayscale value includes the situation of the first sub-block pixel grayscale value being low, the situation of the first sub-block pixel grayscale value being medium, and the situation of the first sub-block pixel grayscale value being high, the situation of the second sub-block pixel grayscale value refers to the high and low degree of the second sub-block pixel grayscale value, and the situation of the second sub-block pixel grayscale value includes the situation of the second sub-block pixel grayscale value being low, the situation of the second sub-block pixel grayscale value being The grayscale value of the pixel of the third sub-block is medium and the grayscale value of the pixel of the second sub-block is high. The grayscale value of the pixel of the third sub-block refers to the high and low degree of the grayscale value of the pixel of the third sub-block. The grayscale value of the pixel of the third sub-block includes the grayscale value of the pixel of the third sub-block is low, the grayscale value of the pixel of the third sub-block is medium and the grayscale value of the pixel of the third sub-block is high. The grayscale value of the pixel of the h-th sub-block refers to the high and low degree of the grayscale value of the pixel of the h-th sub-block. The grayscale value of the pixel of the h-th sub-block includes the grayscale value of the pixel of the h-th sub-block is low, the grayscale value of the pixel of the h-th sub-block is medium and the grayscale value of the h-th sub-block is high. The grayscale value of the sub-block pixel is height, the first rectangular sub-block attribute refers to the tooth position reflected by the first image in the rectangular sub-block collection, the second rectangular sub-block attribute refers to the tooth position reflected by the second image in the rectangular sub-block collection, the third rectangular sub-block attribute refers to the tooth position reflected by the third image in the rectangular sub-block collection, the hth rectangular sub-block attribute refers to the tooth position reflected by the hth image in the rectangular sub-block collection, the tooth gap shadow sub-block refers to the tooth gap position of children's teeth, the enamel essence sub-block refers to the non-reflective position of children's enamel, and the enamel reflective sub-block refers to the non-reflective position of children's enamel. The block refers to the reflective area of ​​a child's tooth enamel. The specialized processing involves compressing the high-grayscale intervals of the enamel reflective sub-block, stretching the dark grayscale of the interdental shadow sub-block, and retaining the original pixel grayscale values ​​of the enamel intrinsic sub-block. Compressing the high-grayscale intervals refers to reducing the pixel grayscale values ​​of the enamel reflective sub-block, while stretching the dark grayscale refers to increasing the pixel grayscale values ​​of the interdental shadow sub-block. This embodiment does not limit the specific methods of compressing the high-grayscale intervals and stretching the dark grayscale. Those skilled in the art can freely select these methods based on actual needs, such as software processing.

[0029] Specifically, the image processing module divides the children's dental mobile phone image to improve the processing accuracy of the children's dental mobile phone image, and at the same time determines the grayscale value of each rectangular sub-block to mark the dental condition reflected by the children's dental mobile phone image, thereby improving the accuracy of image extraction processing.

[0030] Specifically, when the image processing module adjusts the pixel value of the mobile phone image processing method according to the image extraction accuracy, the image extraction accuracy IoU is compared with the preset model accuracy IoU0, and 0.85≤IoU0≤0.90 is set. The compliance of the image extraction accuracy is judged according to the comparison result, and the pixel value of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2 is adjusted according to the judgment result, wherein: When IoU≥IoU0, the image processing module determines that the image extraction accuracy rate meets the standard, and does not adjust the pixel values ​​of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2; When IoU<IoU0, the image processing module determines that the image extraction accuracy rate does not meet the standard, and adjusts the pixel value of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2. The pixel value of the first preset pixel grayscale value ao1 is adjusted according to the pixel value adjustment coefficient au, and au=0.77+0.14×e -0.7×（IoU0-IoU） , where e is the base of the natural logarithm, obtain the first preset pixel grayscale value ao1` after pixel value adjustment, set ao1`=ao1×au, replace the first preset pixel grayscale value ao1 with the first preset pixel grayscale value ao1` after pixel value adjustment, adjust the second preset pixel grayscale value ao2 according to the pixel value adjustment coefficient au, obtain the second preset pixel grayscale value ao2` after pixel value adjustment, set ao2`=ao2×au, replace the second preset pixel grayscale value ao2 with the second preset pixel grayscale value ao2` after pixel value adjustment, and re-compare the first rectangular sub-block attribute aok1, the second rectangular sub-block attribute aok2, the third rectangular sub-block attribute aok3,..., the hth rectangular sub-block attribute aokh with the first preset pixel grayscale value ao1` and the second preset pixel grayscale value ao2` after pixel value adjustment.

[0031] Specifically, the image extraction accuracy refers to the degree of match between the child's dental condition reflected by the processed mobile phone image of the child's teeth and the actual child's dental condition. This embodiment does not limit the specific method of obtaining the image extraction accuracy. Relevant technical personnel in this field can freely choose according to actual needs, such as manual comparison. The preset model accuracy refers to the preset value for judging the compliance of the image extraction accuracy. The compliance of the image extraction accuracy refers to the degree of compliance of the image extraction accuracy. The compliance of the image extraction accuracy includes the compliance of the image extraction accuracy as being up to standard and the compliance of the image extraction accuracy as being not up to standard.

[0032] Specifically, the image processing module judges whether the image extraction accuracy meets the standard. When the image extraction accuracy does not meet the standard, the first preset pixel grayscale value and the second preset pixel grayscale value are reduced as the image extraction accuracy decreases, so as to enhance the judgment accuracy of the dark area of ​​the children's dental mobile phone image, avoid the misjudgment of the specific position of the teeth due to insufficient image extraction accuracy, and thus improve the quality of the children's dental mobile phone image.

[0033] Specifically, when the image processing module performs migration adjustment on the pixel value adjustment process according to the number of remote diagnoses, the migration factor Yj is calculated using the number of remote diagnoses Yd and the number of new scenario cases Yx to obtain the migration factor Yj, and Yj=0.45×Yd+0.55×Yx is set. The migration factor Yj is compared with the preset migration factor Yj0, and 0.56≤Yj0≤0.78 is set. The compliance of the migration factor is judged based on the comparison result, and the preset model accuracy IoU0 is migrated and adjusted based on the judgment result, where: When Yj≥Yj0, the image processing module determines that the migration factor meets the standard and does not perform migration adjustment on the preset model accuracy IoU0; When Yj<Yj0, the image processing module determines that the migration factor is not up to standard, and performs migration adjustment on the preset model accuracy IoU0. The preset model accuracy IoU0 is adjusted according to the migration coefficient yu, and yu=1.31-0.16×e -0.61×（Yj0-Yj） , obtain the adjusted preset model accuracy IoU0`, set IoU0`=IoU0×yu, replace the preset model accuracy IoU0 with the adjusted preset model accuracy IoU0`, and re-compare the image extraction accuracy IoU with the adjusted preset model accuracy IoU0`.

[0034] Specifically, the remote diagnosis number refers to the ratio of the number of dental diagnosis cases completed by telemedicine within the preset diagnosis time to the total number of dental diagnosis cases. For example, if the preset diagnosis time is set to three months, the number of dental diagnosis cases completed by telemedicine is AS, and the total number of dental diagnosis cases is AS0, then Yd=AS / AS0. The telemedicine method refers to the method of diagnosing dental cases remotely, such as online image transmission and video consultation. The total number of dental diagnosis cases refers to the total number of dental cases diagnosed by the hospital within the preset diagnosis time. This embodiment does not limit the method for obtaining the remote diagnosis number. Related technologies in this field Surgeons can freely choose according to actual needs, such as those obtained through the hospital information system. The number of new scenario cases refers to new case scenarios in dental diagnosis that go beyond traditional clinical scenarios. The traditional clinical scenarios refer to cases that can be diagnosed through traditional medical means, such as routine clinics and standard equipment examinations. The preset migration factor refers to a preset value for judging the compliance of the migration factor. The compliance of the migration factor refers to the degree of compliance of the migration factor judged based on the migration factor and the preset migration factor. The compliance of the migration factor includes the compliance of the migration factor as being in compliance and the compliance of the migration factor as being not in compliance.

[0035] Specifically, the image processing module calculates the migration factor. The larger the migration factor, the stronger the adaptability of the new scene to the technology. When the migration factor does not meet the standard, the accuracy of the preset model is increased to reduce the impact of the new scene's inadaptability to the technology on image extraction, thereby increasing the accuracy of the image.

[0036] Specifically, the image processing module performs professional image migration on the professional image of children's teeth based on the processed mobile phone image of children's teeth using a professional image migration method, and the professional image migration method includes: Step B01, initialize the network parameters of the generator and the discriminator to obtain the initialized generator and the initialized discriminator; Step B02: Input the professional image of children's teeth into the initialized generator, extract the illumination features through the dynamic illumination compensation layer, and generate an illumination mask based on the illumination features; Step B03, performing illumination compensation on the professional image of children's teeth according to the illumination mask to obtain the professional image of children's teeth after illumination compensation; Step B04, performing upsampling and convolution operations on the professional children's dental image after illumination compensation to obtain a preliminary simulated image; In step B05, the preliminary simulation image and the processed children's dental mobile phone image are input into the initialized discriminator, and a similarity value FK is output through multi-layer convolution. The preliminary simulation image with a similarity value of 90% is output as the simulation image.

[0037] Specifically, the generator refers to a component that converts a professional image of children's teeth into a simulated image, the discriminator refers to a component used to distinguish between processed mobile phone images of children's teeth and simulated images, and the network parameter initialization refers to the process of setting the initial parameters of the generator and the discriminator. This embodiment does not limit the specific method of initializing the network parameters. Relevant technical personnel in this field can freely choose according to actual needs, such as random initialization. The dynamic lighting compensation layer refers to a functional module that extracts the lighting features of professional images of children's teeth and generates a corresponding lighting mask. The lighting features refer to characteristic information used to describe the lighting conditions of professional images of children's teeth, such as light distribution and intensity. This embodiment does not limit the specific method of extracting lighting features by the dynamic lighting compensation layer. Relevant technical personnel in this field can freely choose according to actual needs, such as software extraction. The lighting mask refers to a two-dimensional matrix used to adjust the lighting effect of professional images of children's teeth. The elements in the lighting mask correspond to pixels in the professional images of children's teeth. The lighting compensation refers to adjusting the lighting of the image according to the lighting mask. Distribution, the process of correcting visual distortion caused by lighting problems. This embodiment does not limit the specific method of lighting compensation. Relevant technical personnel in this field can freely choose according to actual needs, such as software processing. The upsampling refers to the process of restoring the professional image of children's teeth after lighting compensation from low resolution to the original size of the professional image of children's teeth. The convolution operation refers to the process of converting the professional image of children's teeth after lighting compensation processed by upsampling into a preliminary simulation image. This embodiment does not limit the specific methods of upsampling and convolution operations. Relevant technical personnel in this field can freely choose according to actual needs, such as software processing. The multi-layer convolution refers to the process of extracting image features from the preliminary simulation image step by step based on the processed children's teeth mobile phone image to determine whether the preliminary simulation image is the processed children's teeth mobile phone image. This embodiment does not limit the specific method of performing multi-layer convolution. Relevant technical personnel in this field can freely choose according to actual needs, such as software processing. The similarity value refers to a numerical value reflecting the similarity between the preliminary simulation image and the processed children's teeth mobile phone image.

[0038] Specifically, the image processing module performs professional image migration on professional children's dental images to obtain high-quality simulated images that retain dental diagnostic information and conform to the lighting characteristics of mobile phone images, thereby improving the accuracy of abnormality analysis of children's deciduous and permanent teeth through children's dental mobile phone images.

[0039] Specifically, when the deciduous and permanent teeth detection module obtains the deciduous and permanent teeth detection results according to the simulated image, the deciduous and permanent teeth detection model is constructed according to the deciduous and permanent teeth detection model construction method, and the deciduous and permanent teeth detection model construction method includes: Step J01, dividing the historical simulation training set into 50% deciduous and permanent tooth position annotation frames and 50% caries status annotation frames; Step J02, performing bias initialization on the EfficientNet-B4 detection model to obtain an initialized EfficientNet-B4 detection model; Step J03: Input the labeled boxes of the deciduous and permanent teeth positions and the labeled boxes of the caries plaque status into the initialized EfficientNet-B4 detection model for labeling training, and obtain the Intersection-Union (IU) ratio between the trained EfficientNet-B4 detection model and the target. Step J04, performing loss optimization on the trained EfficientNet-B4 detection model according to the bounding box loss function Lc; Step J05: compare the target intersection-and-union ratio IU with the preset target intersection-and-union ratio IU0, set 0.5≤IU0≤0.7, judge whether the target intersection-and-union ratio meets the standard based on the comparison result, and output the deciduous and permanent teeth detection model based on the judgment result, where: When IU≥IU0, the deciduous and permanent teeth detection module determines that the target intersection-over-union ratio meets the standard, and outputs the trained EfficientNet-B4 detection model as the deciduous and permanent teeth detection model; When IU<IU0, the deciduous and permanent teeth detection module determines that the target intersection-over-union ratio is not up to standard, and adjusts the loss of the bounding box loss function Lc according to the loss adjustment coefficient lm, setting lm=1.22-0.16×e -0.7×（IU0-IU） , obtain the adjusted bounding box loss function Lc`, set Lc`=Lc×lm, replace the bounding box loss function Lc with the adjusted bounding box loss function Lc`, and re-optimize the loss of the trained EfficientNet-B4 detection model according to the adjusted bounding box loss function Lc` until the target intersection-union ratio meets the standard; The processed mobile phone image of children's teeth is input into the deciduous and permanent teeth detection model to obtain the deciduous and permanent teeth detection results.

[0040] Specifically, the historical simulation training set refers to the simulation image and the deciduous and permanent teeth detection results corresponding to the simulation image. The deciduous and permanent teeth detection results refer to the images obtained according to the deciduous and permanent teeth detection model for marking the deciduous and permanent teeth and caries of children's teeth. The deciduous and permanent teeth position marking frame refers to a rectangular frame marking the specific positions of the deciduous and permanent teeth in the simulation image. The caries status marking frame refers to a rectangular frame used to mark the caries lesion area on the tooth surface in the simulation image. This embodiment does not limit the specific acquisition method of the deciduous and permanent teeth position marking frame and the caries status marking frame. Relevant technicians in this field can freely choose according to actual needs, such as software marking or manual marking. The EfficientNet-B4 detection model is Refers to the model infrastructure for constructing the deciduous and permanent teeth detection model. The bias initialization refers to the process of setting the initial value of the EfficientNet-B4 detection model. This embodiment does not limit the specific method of bias initialization. Relevant technical personnel in this field can freely choose according to actual needs, such as random initialization. The annotation training refers to the process of training the initialized EfficientNet-B4 detection model with the deciduous and permanent teeth position annotation box and the caries status annotation box to obtain the process of the intersection and union ratio of the trained EfficientNet-B4 detection model and the target. This embodiment does not limit the specific implementation method of annotation training. Relevant technical personnel in this field can freely choose according to actual needs. Freely select, such as through software training, the target intersection-over-union ratio refers to a value used to evaluate the degree of overlap between the deciduous and permanent teeth detection results output by the trained EfficientNet-B4 detection model and the actual deciduous and permanent teeth detection results. This embodiment does not limit the specific method for obtaining the target intersection-over-union ratio. Relevant technical personnel in this field can freely select according to actual needs, such as software calculation. The bounding box loss function refers to a function used to measure the difference between the deciduous and permanent teeth detection results output by the trained EfficientNet-B4 detection model and the actual deciduous and permanent teeth detection results. This embodiment does not limit the specific method for obtaining the bounding box loss function. Relevant technical personnel in this field can freely select according to actual needs. Free choice, such as software calculation, the loss optimization refers to the process of reducing the difference between the deciduous and permanent teeth detection results output by the trained EfficientNet-B4 detection model and the actual deciduous and permanent teeth detection results. This embodiment does not limit the specific method of loss optimization. Relevant technical personnel in this field can freely choose according to actual needs, such as software calculation. The preset target intersection-and-union ratio refers to a preset value for judging the compliance of the target intersection-and-union ratio. The compliance of the target intersection-and-union ratio refers to the degree of compliance of the target intersection-and-union ratio judged based on the target intersection-and-union ratio and the preset target intersection-and-union ratio. The compliance of the target intersection-and-union ratio includes the compliance of the target intersection-and-union ratio as being up to standard and the compliance of the target intersection-and-union ratio as being not up to standard.

[0041] Specifically, the deciduous and permanent teeth detection module constructs a deciduous and permanent teeth detection model so that abnormal analysis of children's teeth can be performed based on the processed mobile phone images of children's teeth, thereby improving the efficiency of abnormal analysis of children's teeth through mobile phone images. At the same time, when the target intersection-union ratio does not meet the standard, the bounding box loss function is increased to improve the accuracy of the output results of the deciduous and permanent teeth detection model.

[0042] Specifically, when the deciduous and permanent teeth detection module updates the result of the acquisition process of the deciduous and permanent teeth detection results according to the number of teeth, the number of teeth Ys is compared with the preset number of teeth Ys0, and 10≤Ys0≤16 is set. The state of the number of teeth is judged according to the comparison result, and the preset target intersection and union ratio IU0 is updated according to the judgment result, wherein: When Ys≥Ys0, the deciduous and permanent teeth detection module determines that the state of the number of teeth is large, and does not update the result of the preset target intersection and union ratio IU0; When Ys<Ys0, the deciduous and permanent teeth detection module determines that the number of teeth is small, updates the preset target intersection-and-union ratio IU0, and updates the preset target intersection-and-union ratio IU0 according to the result update coefficient yo, setting yo=1.37+0.14×e -1.24×（Ys / Ys0） , e is the base of the natural logarithm, and the updated preset target intersection-union ratio IU0` is obtained. IU0`=IU0×yo is set, and the preset target intersection-union ratio IU0 is replaced by the updated preset target intersection-union ratio IU0`, and the target intersection-union ratio IU is re-compared with the updated preset target intersection-union ratio IU0`.

[0043] Specifically, the number of teeth refers to the number of teeth that can be observed in the processed mobile phone image of children's teeth. This embodiment does not limit the specific method of obtaining the number of teeth. Relevant technical personnel in this field can freely choose according to actual needs, such as manual statistics. The preset number of teeth refers to the preset value for judging the status of the number of teeth. The status of the number of teeth refers to the number of teeth judged based on the number of teeth and the preset number of teeth. The status of the number of teeth includes the status of the number of teeth being large and the status of the number of teeth being small.

[0044] Specifically, the deciduous and permanent teeth detection module judges the status of the number of teeth. When the number of teeth is small, the preset target intersection-over-union ratio is increased as the number of teeth decreases, so as to reduce the impact of the number of teeth on the deciduous and permanent teeth detection results, thereby improving the accuracy of the deciduous and permanent teeth detection results.

[0045] Specifically, when the abnormality detection module obtains the abnormality of deciduous and permanent teeth based on the deciduous and permanent teeth detection results, the deciduous and permanent teeth detection results are input into the deciduous and permanent teeth prediction model to obtain the deciduous teeth shedding time, permanent teeth eruption time and deciduous teeth unchanged time.

[0046] Specifically, the deciduous and permanent teeth prediction model refers to a convolutional neural network model that takes the deciduous and permanent teeth detection results as input and takes the deciduous tooth shedding time, permanent tooth eruption time and deciduous tooth unchanged time as output. The abnormality detection module constructs the deciduous and permanent teeth prediction model through the deciduous and permanent teeth prediction model construction method. This embodiment does not limit the deciduous and permanent teeth prediction model construction method. Relevant technical personnel in this field can freely choose according to actual needs, such as constructing the deciduous and permanent teeth prediction model using the historical deciduous and permanent teeth detection results and the historical deciduous tooth shedding time, historical permanent tooth eruption time and historical deciduous tooth unchanged time corresponding to the historical deciduous and permanent teeth detection results. The deciduous tooth shedding time refers to the time point at which the deciduous teeth are predicted to fall out obtained according to the deciduous and permanent teeth prediction model, the permanent tooth eruption time refers to the time point at which the permanent teeth erupt obtained according to the deciduous and permanent teeth prediction model, and the deciduous tooth unchanged time refers to the length of time that the deciduous teeth remain unchanged obtained according to the deciduous and permanent teeth prediction model.

[0047] Specifically, the abnormality detection module obtains the time when deciduous teeth fall out, the time when permanent teeth erupt, and the time when deciduous teeth remain unchanged, so as to facilitate subsequent prediction of the condition of children's teeth, thereby improving treatment efficiency.

[0048] Specifically, when the abnormality detection module obtains the abnormality of the deciduous and permanent teeth according to the detection results of the deciduous and permanent teeth, the deciduous tooth shedding time Rt is compared with the first preset deciduous tooth shedding time Rt1 and the second preset deciduous tooth shedding time Rt2, and Rt1=3 months and Rt2=6 months are set. The deciduous tooth shedding time is judged according to the comparison results, and the deciduous tooth condition is output according to the judgment result, wherein: When Rt≤Rt1, the abnormality detection module determines that the deciduous tooth shedding time is abnormal and outputs the deciduous tooth abnormality as the deciduous tooth condition; When Rt1<Rt≤Rt2, the abnormality detection module determines that the deciduous tooth shedding time is normal and outputs the normal deciduous tooth as the deciduous tooth condition; When Rt>Rt2, the abnormality detection module determines that the deciduous tooth shedding time is abnormal and outputs the deciduous tooth abnormality as the deciduous tooth condition; The abnormality detection module compares the permanent tooth eruption time Rh with the first preset permanent tooth eruption time Rh1 and the second preset permanent tooth eruption time Rh2, sets Rh1=3 months and Rh2=6 months, judges the permanent tooth eruption time according to the comparison results, and outputs the permanent tooth status according to the judgment results, wherein: When Rh≤Rh1, the abnormality detection module determines that the permanent tooth eruption time is abnormal and outputs the permanent tooth abnormality as the permanent tooth condition; When Rh1<Rh≤Rh2, the abnormality detection module determines that the permanent tooth eruption time is normal and outputs the normal permanent tooth as the permanent tooth condition; When Rh>Rh2, the abnormality detection module determines that the permanent tooth eruption time is abnormal and outputs the permanent tooth abnormality as the permanent tooth condition; The abnormality detection module compares the unchanged time of deciduous teeth Rrw with the preset unchanged time Rrw0, sets Rrw0=2 years, judges the unchanged time of deciduous teeth according to the comparison result, and outputs the unchanged situation of deciduous teeth according to the judgment result, wherein: When Rrw≤Rrw0, the abnormality detection module determines that the situation of no change in deciduous teeth is normal, and outputs the normal situation of no change in deciduous teeth as the situation of no change in deciduous teeth; When Rrw>Rrw0, the abnormality detection module determines that the situation of no change in deciduous teeth is abnormal, and outputs the abnormality of no change in deciduous teeth as the situation of no change in deciduous teeth.

[0049] The deciduous teeth condition, permanent teeth condition and no change in deciduous teeth condition are output as abnormal conditions of deciduous and permanent teeth.

[0050] Specifically, the first preset deciduous tooth shedding time refers to the preset lower limit value for judging the situation of the deciduous tooth shedding time, the second preset deciduous tooth shedding time refers to the preset upper limit value for judging the situation of the deciduous tooth shedding time, the situation of the deciduous tooth shedding time refers to the normal degree of the deciduous tooth shedding time, and the situation of the deciduous tooth shedding time includes the situation of the deciduous tooth shedding time being normal and the situation of the deciduous tooth shedding time being abnormal, the first preset permanent tooth eruption time refers to the preset lower limit value for judging the situation of the permanent tooth eruption time, the second preset permanent tooth eruption time refers to the preset upper limit value for judging the situation of the permanent tooth eruption time, the situation of the permanent tooth eruption time refers to the normal degree of the permanent tooth eruption time, and the situation of the permanent tooth eruption time includes the situation of the permanent tooth eruption time being abnormal and the situation of the permanent tooth eruption time being normal, the preset no-change time refers to the preset value for judging the situation of the no-change time of deciduous teeth, the situation of the no-change time of deciduous teeth refers to the normal degree of the no-change time of deciduous teeth, and the situation of the no-change time of deciduous teeth includes the situation of the no-change time of deciduous teeth being normal and the situation of the no-change time of deciduous teeth being abnormal.

[0051] Specifically, the abnormality detection module obtains abnormal conditions of deciduous and permanent teeth in order to further predict the condition of children's teeth and improve prevention efficiency.

[0052] Specifically, when the abnormality detection module obtains the abnormality risk level according to the abnormality of deciduous and permanent teeth, the abnormality risk score Rp is obtained according to the deciduous tooth abnormality parameter R1, the permanent tooth abnormality parameter R2, the unchanged abnormality parameter R3, the deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2 and the unchanged abnormality weight β3, and Rp=β1×R1+β2×R2+β3×R3 is set to obtain the abnormality risk score Rp. The abnormality risk score Rp is compared with the preset abnormality risk score Rp0, and the abnormality risk degree is judged according to the comparison result, and the abnormality risk level is output according to the judgment result, wherein: When Rp≤Rp0, the anomaly detection module determines that the abnormal risk level is normal and outputs no risk as the abnormal risk level; When Rp>Rp0, the abnormality detection module determines that the abnormality risk degree is abnormal and outputs risk as the abnormality risk level.

[0053] Specifically, the deciduous teeth abnormality parameter refers to the numerical value measuring the degree of abnormality of deciduous teeth in the case of abnormalities of deciduous and permanent teeth, the permanent teeth abnormality parameter refers to the numerical value measuring the degree of abnormality of permanent teeth in the case of abnormalities of deciduous and permanent teeth, and the unchanged abnormality parameter refers to the numerical value measuring the degree of abnormality of deciduous teeth in the test results of deciduous and permanent teeth. This embodiment does not limit the specific acquisition method of the deciduous teeth abnormality parameter, the permanent teeth abnormality parameter and the unchanged abnormality parameter. Relevant technical personnel in this field can freely choose according to actual needs, such as expert evaluation. The deciduous teeth abnormality weight refers to the coefficient measuring the importance of the deciduous teeth abnormality parameter in the abnormal risk score, the permanent teeth abnormality weight refers to the coefficient measuring the importance of the permanent teeth abnormality parameter in the abnormal risk score, and the unchanged abnormality parameter The abnormal weight refers to a coefficient that measures the importance of the unchanged abnormal parameter in the abnormal risk score. This embodiment does not limit the deciduous tooth abnormal weight β1, the permanent tooth abnormal weight β2 and the unchanged abnormal weight β3. Relevant technical personnel in this field can freely choose according to actual needs, and only need to satisfy β1+β2+β3=1. For example, β1=0.4, β2=0.3, and β3=0.3 are set. The preset abnormal risk score refers to a preset value for judging the degree of abnormal risk. The abnormal risk degree refers to the normal degree of risk of children's deciduous and permanent teeth judged according to the abnormal risk score and the preset abnormal risk score. The abnormal risk degree includes a normal abnormal risk degree and an abnormal abnormal risk degree.

[0054] Specifically, the abnormality detection module obtains the abnormality risk level so as to take preventive and therapeutic measures for the child's deciduous and permanent teeth according to the abnormality risk level, thereby improving the efficiency of abnormality analysis of the child's deciduous and permanent teeth.

[0055] Specifically, when the abnormality detection module adjusts the abnormality risk level acquisition process according to the child's age, the child's age Nr is compared with the preset child age Nr0, and the child's age is judged according to the comparison result. The deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3 are adjusted according to the judgment result, where: When Nr≤Nr0, the anomaly detection module determines that the child is young and does not adjust the level of the deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3; When Nr>Nr0, the abnormality detection module determines that the child is old, and adjusts the grade of the deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2, and the unchanged abnormality weight β3. The deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3 are adjusted according to the grade adjustment coefficient uop, and 1.1<uop<1.4 is set to obtain the adjusted deciduous tooth abnormality weight β1` and the adjusted unchanged abnormality weight β3`, and β1`=β1×uop, β3`=(uop-1)×β3 are set. The permanent tooth abnormality weight β2 is graded and β2=1-β1-β3 to obtain the adjusted permanent tooth abnormality weight β2`. The deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2, and the changed abnormality weight β3 are replaced with the adjusted deciduous tooth abnormality weight β1`, the adjusted permanent tooth abnormality weight β2`, and the adjusted unchanged abnormality weight β3`, and the abnormality risk score is recalculated.

[0056] Specifically, the child's age refers to the child's age obtained based on the processed mobile phone image of the child's teeth. This embodiment does not limit the specific method of obtaining the child's age. Relevant technical personnel in this field can freely choose according to actual needs, such as obtaining it on the Internet. The preset child's age refers to a preset value for judging the child's age situation. The child's age situation refers to the age of the child. The child's age situation includes the child's age being young and the child's age being old.

[0057] Specifically, the abnormality detection module judges the age of the child. When the child is older, the weight of the abnormality of deciduous teeth is increased and the weight of the abnormality of no change is reduced, so as to avoid the impact of the child's old age on the determination of the abnormal risk level, thereby improving the accuracy of the abnormality analysis of children's deciduous and permanent teeth.

[0058] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. A deep learning-based abnormality analysis system for children's deciduous and permanent teeth replacement period, characterized by: The system comprises: An image acquisition module, used to acquire target images; An image screening module is used to screen the target image and obtain a usable target image; an image processing module, configured to perform image extraction processing on the mobile phone image of children's teeth in the target available image according to the mobile phone image processing method to obtain a processed mobile phone image of children's teeth, further configured to adjust the pixel values ​​of the mobile phone image processing method, further configured to perform migration adjustment on the pixel value adjustment process, and further configured to perform professional image migration on the professional image of children's teeth in the target available image according to the processed mobile phone image of children's teeth to obtain a simulated image; The deciduous and permanent teeth prediction module is used to obtain the deciduous and permanent teeth detection results based on the simulated image, obtain the deciduous and permanent teeth detection results, and also to update the results of the deciduous and permanent teeth detection result acquisition process; The abnormality detection module is used to obtain the abnormal conditions of deciduous and permanent teeth based on the detection results of deciduous and permanent teeth, and to obtain the abnormality risk level based on the abnormal conditions of deciduous and permanent teeth, and to adjust the level of the abnormality risk level acquisition process.

2. The deep learning-based children's deciduous and permanent tooth replacement abnormality analysis system according to claim 1 is characterized in that: When the image screening module screens the target image, it compares the target image gradient amplitude ERU with the preset gradient amplitude ERU0, sets 12≤ERU0≤15, judges the target image's compliance with the standard based on the comparison result, and outputs the target usable image based on the judgment result, wherein: When ERU≥ERU0, the image screening module determines that the target image meets the standards and outputs the target image as a target available image; When ERU<ERU0, the image screening module determines that the target image does not meet the standards, does not output the target image as a target available image, and removes the target image; When the image screening module screens the target image, it compares the image entropy value ERB with the preset image entropy value ERB0, sets 6.5≤ERB0≤7.5, judges whether the image entropy value meets the standard based on the comparison result, and adjusts the clarity of the preset gradient amplitude ERU0 based on the judgment result, wherein: When ERB≥ERB0, the image screening module determines that the image entropy value meets the standard and does not adjust the clarity of the preset gradient amplitude ERU0; When ERB<ERB0, the image screening module determines that the image entropy value does not meet the standard, and adjusts the clarity of the preset gradient amplitude ERU0 according to the clarity adjustment coefficient asx, setting asx=1.35-0.15×e -0.69×（ERB-ERB0） , where e is the base of the natural logarithm, and the adjusted preset gradient amplitude ERU0` is obtained. ERU0`=ERU0×asx is set, and the preset gradient amplitude ERU0 is replaced with the adjusted preset gradient amplitude ERU0`. The target image gradient amplitude is then re-compared with the adjusted preset gradient amplitude ERU0`.

3. The deep learning-based children's deciduous and permanent tooth replacement abnormality analysis system according to claim 2 is characterized in that: The image processing module performs image extraction processing on the target available image according to a mobile phone image processing method, and the mobile phone image processing method includes: Step A01, dividing the mobile phone image of children's teeth into a set of rectangular sub-blocks aol={aol1, aol2, aol3, ..., aolh}; Step A02, obtaining sub-block pixel grayscale values ​​aom={aom1, aom2, aom3, ..., aomh} according to the rectangular sub-block set aol={aol1, aol2, aol3, ..., aolh}; Step A03: Compare the first sub-block pixel grayscale value aom1 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the first sub-block pixel grayscale value based on the comparison result, and output the first rectangular sub-block attribute aok1 based on the judgment result, where: When aom1≤ao1, the image processing module determines that the grayscale value of the first sub-block pixel is low, and outputs the enamel reflective sub-block as the first rectangular sub-block attribute aok1; When ao1<aom1≤ao2, the image processing module determines that the grayscale value of the first sub-block pixel is moderate, and outputs the enamel essence sub-block as the first rectangular sub-block attribute aok1; When aom1>ao2, the image processing module determines that the grayscale value of the first sub-block pixel is high, and outputs the tooth gap shadow sub-block as the first rectangular sub-block attribute aok1; Compare the second sub-block pixel grayscale value aom2 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the second sub-block pixel grayscale value based on the comparison result, and output the second rectangular sub-block attribute aok2 based on the judgment result, where: When aom2≤ao1, the image processing module determines that the grayscale value of the second sub-block pixel is low, and outputs the enamel reflective sub-block as the second rectangular sub-block attribute aok2; When ao1<aom2≤ao2, the image processing module determines that the grayscale value of the second sub-block pixel is moderate, and outputs the enamel essence sub-block as the second rectangular sub-block attribute aok2; When aom2>ao2, the image processing module determines that the grayscale value of the second sub-block pixel is high, and outputs the tooth gap shadow sub-block as the second rectangular sub-block attribute aok2; Compare the third sub-block pixel grayscale value aom3 with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the third sub-block pixel grayscale value based on the comparison result, and output the third rectangular sub-block attribute aok3 based on the judgment result, where: When aom3≤ao1, the image processing module determines that the grayscale value of the third sub-block pixel is low, and outputs the enamel reflective sub-block as the third rectangular sub-block attribute aok3; When ao1<aom3≤ao2, the image processing module determines that the grayscale value of the third sub-block pixel is moderate, and outputs the enamel essence sub-block as the third rectangular sub-block attribute aok3; When aom3>ao2, the image processing module determines that the grayscale value of the third sub-block pixel is high, and outputs the tooth gap shadow sub-block as the third rectangular sub-block attribute aok3; …… Compare the hth sub-block pixel grayscale value aomh with the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2, set ao1=80, ao2=200, judge the hth sub-block pixel grayscale value based on the comparison result, and output the hth rectangular sub-block attribute aokh based on the judgment result, where: When aomh≤ao1, the image processing module determines that the grayscale value of the h-th sub-block pixel is low, and outputs the enamel reflective sub-block as the h-th rectangular sub-block attribute aokh; When ao1<aomh≤ao2, the image processing module determines that the grayscale value of the h-th sub-block pixel is moderate, and outputs the enamel intrinsic sub-block as the h-th rectangular sub-block attribute aokh; When aomh>ao2, the image processing module determines that the grayscale value of the h-th sub-block pixel is high, and outputs the tooth gap shadow sub-block as the h-th rectangular sub-block attribute aokh; Step A04: Output the first rectangular sub-block attribute aok1, the second rectangular sub-block attribute aok2, the third rectangular sub-block attribute aok3, ..., the hth rectangular sub-block attribute aokh as the light and dark annotated mobile phone image of the child's teeth; Step A05, performing specific processing on the light and dark marked mobile phone image of the child's teeth to obtain a processed mobile phone image of the child's teeth; When the image processing module adjusts the pixel value of the mobile phone image processing method according to the image extraction accuracy, the image extraction accuracy IoU is compared with the preset model accuracy IoU0, and 0.85≤IoU0≤0.90 is set. The compliance of the image extraction accuracy is judged according to the comparison result, and the pixel value of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2 is adjusted according to the judgment result, wherein: When IoU≥IoU0, the image processing module determines that the image extraction accuracy rate meets the standard, and does not adjust the pixel values ​​of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2; When IoU<IoU0, the image processing module determines that the image extraction accuracy rate does not meet the standard, and adjusts the pixel value of the first preset pixel grayscale value ao1 and the second preset pixel grayscale value ao2. The pixel value of the first preset pixel grayscale value ao1 is adjusted according to the pixel value adjustment coefficient au, and au=0.77+0.14×e -0.7×（IoU0-IoU） , where e is the base of the natural logarithm, obtain the first preset pixel grayscale value ao1` after pixel value adjustment, set ao1`=ao1×au, replace the first preset pixel grayscale value ao1 with the first preset pixel grayscale value ao1` after pixel value adjustment, adjust the second preset pixel grayscale value ao2 according to the pixel value adjustment coefficient au, obtain the second preset pixel grayscale value ao2` after pixel value adjustment, set ao2`=ao2×au, replace the second preset pixel grayscale value ao2 with the second preset pixel grayscale value ao2` after pixel value adjustment, and re-compare the first rectangular sub-block attribute aok1, the second rectangular sub-block attribute aok2, the third rectangular sub-block attribute aok3,..., the hth rectangular sub-block attribute aokh with the first preset pixel grayscale value ao1` and the second preset pixel grayscale value ao2` after pixel value adjustment.

4. The deep learning-based children's deciduous and permanent tooth replacement abnormality analysis system according to claim 3 is characterized in that: When the image processing module performs migration adjustment on the pixel value adjustment process according to the number of remote diagnoses, the migration factor Yj is calculated by the number of remote diagnoses Yd and the number of cases in the new scene Yx to obtain the migration factor Yj, and Yj=0.45×Yd+0.55×Yx is set. The migration factor Yj is compared with the preset migration factor Yj0, and 0.56≤Yj0≤0.78 is set. The compliance of the migration factor is judged according to the comparison result, and the preset model accuracy IoU0 is migrated and adjusted according to the judgment result, wherein: When Yj≥Yj0, the image processing module determines that the migration factor meets the standard and does not perform migration adjustment on the preset model accuracy IoU0; When Yj<Yj0, the image processing module determines that the migration factor is not up to standard, and performs migration adjustment on the preset model accuracy IoU0. The preset model accuracy IoU0 is adjusted according to the migration coefficient yu, and yu=1.31-0.16×e -0.61×（Yj0-Yj） , obtain the adjusted preset model accuracy IoU0`, set IoU0`=IoU0×yu, replace the preset model accuracy IoU0 with the adjusted preset model accuracy IoU0`, and re-compare the image extraction accuracy IoU with the adjusted preset model accuracy IoU0`.

5. The deep learning-based children's deciduous and permanent tooth replacement abnormality analysis system according to claim 4 is characterized in that: The image processing module performs professional image migration on the professional image of children's teeth based on the processed mobile phone image of children's teeth by a professional image migration method, and the professional image migration method includes: Step B01, initialize the network parameters of the generator and the discriminator to obtain the initialized generator and the initialized discriminator; Step B02: Input the professional image of children's teeth into the initialized generator, extract the illumination features through the dynamic illumination compensation layer, and generate an illumination mask based on the illumination features; Step B03, performing illumination compensation on the professional image of children's teeth according to the illumination mask to obtain the professional image of children's teeth after illumination compensation; Step B04, performing upsampling and convolution operations on the professional children's dental image after illumination compensation to obtain a preliminary simulated image; In step B05, the preliminary simulation image and the processed children's dental mobile phone image are input into the initialized discriminator, and a similarity value FK is output through multi-layer convolution. The preliminary simulation image with a similarity value of 90% is output as the simulation image.

6. The deep learning-based children's deciduous and permanent tooth replacement abnormality analysis system according to claim 5, characterized in that: When the deciduous and permanent teeth detection module obtains the deciduous and permanent teeth detection results according to the simulated image, the deciduous and permanent teeth detection model is constructed according to the deciduous and permanent teeth detection model construction method, and the deciduous and permanent teeth detection model construction method includes: Step J01, dividing the historical simulation training set into 50% deciduous and permanent tooth position annotation frames and 50% caries status annotation frames; Step J02, performing bias initialization on the EfficientNet-B4 detection model to obtain an initialized EfficientNet-B4 detection model; Step J03: Input the labeled boxes of the deciduous and permanent teeth positions and the labeled boxes of the caries plaque status into the initialized EfficientNet-B4 detection model for labeling training, and obtain the Intersection-Union (IU) ratio between the trained EfficientNet-B4 detection model and the target. Step J04, performing loss optimization on the trained EfficientNet-B4 detection model according to the bounding box loss function Lc; Step J05: compare the target intersection-and-union ratio IU with the preset target intersection-and-union ratio IU0, set 0.5≤IU0≤0.7, judge whether the target intersection-and-union ratio meets the standard based on the comparison result, and output the deciduous and permanent teeth detection model based on the judgment result, where: When IU≥IU0, the deciduous and permanent teeth detection module determines that the target intersection-over-union ratio meets the standard, and outputs the trained EfficientNet-B4 detection model as the deciduous and permanent teeth detection model; When IU<IU0, the deciduous and permanent teeth detection module determines that the target intersection-over-union ratio is not up to standard, and adjusts the loss of the bounding box loss function Lc according to the loss adjustment coefficient lm, setting lm=1.22-0.16×e -0.7×（IU0-IU） , obtain the adjusted bounding box loss function Lc`, set Lc`=Lc×lm, replace the bounding box loss function Lc with the adjusted bounding box loss function Lc`, and re-optimize the loss of the trained EfficientNet-B4 detection model according to the adjusted bounding box loss function Lc` until the target intersection-union ratio meets the standard; The processed mobile phone image of children's teeth is input into the deciduous and permanent teeth detection model to obtain the deciduous and permanent teeth detection results.

7. The deep learning-based children's deciduous and permanent tooth replacement abnormality analysis system according to claim 6, characterized in that: When the deciduous and permanent teeth detection module updates the result of the acquisition process of the deciduous and permanent teeth detection results according to the number of teeth, the number of teeth Ys is compared with the preset number of teeth Ys0, and 10≤Ys0≤16 is set. The state of the number of teeth is judged according to the comparison result, and the preset target intersection and union ratio IU0 is updated according to the judgment result, wherein: When Ys≥Ys0, the deciduous and permanent teeth detection module determines that the state of the number of teeth is large, and does not update the result of the preset target intersection and union ratio IU0; When Ys<Ys0, the deciduous and permanent teeth detection module determines that the number of teeth is small, updates the preset target intersection-and-union ratio IU0, and updates the preset target intersection-and-union ratio IU0 according to the result update coefficient yo, setting yo=1.37+0.14×e -1.24×（Ys / Ys0） , e is the base of the natural logarithm, and the updated preset target intersection-union ratio IU0` is obtained. IU0`=IU0×yo is set, and the preset target intersection-union ratio IU0 is replaced by the updated preset target intersection-union ratio IU0`, and the target intersection-union ratio IU is re-compared with the updated preset target intersection-union ratio IU0`.

8. The deep learning-based children's deciduous and permanent tooth replacement abnormality analysis system according to claim 7, characterized in that: When the abnormality detection module obtains the abnormality of the deciduous and permanent teeth according to the deciduous and permanent teeth detection results, the deciduous and permanent teeth detection results are input into the deciduous and permanent teeth prediction model to obtain the deciduous teeth shedding time, permanent teeth eruption time and deciduous teeth unchanged time; When the abnormality detection module obtains the abnormality of the deciduous and permanent teeth according to the detection results of the deciduous and permanent teeth, the deciduous tooth shedding time Rt is compared with the first preset deciduous tooth shedding time Rt1 and the second preset deciduous tooth shedding time Rt2, and Rt1 is set to 3 months and Rt2 to 6 months. The deciduous tooth shedding time is judged according to the comparison result, and the deciduous tooth condition is output according to the judgment result, wherein: When Rt≤Rt1, the abnormality detection module determines that the deciduous tooth shedding time is abnormal and outputs the deciduous tooth abnormality as the deciduous tooth condition; When Rt1<Rt≤Rt2, the abnormality detection module determines that the deciduous tooth shedding time is normal and outputs the normal deciduous tooth as the deciduous tooth condition; When Rt>Rt2, the abnormality detection module determines that the deciduous tooth shedding time is abnormal and outputs the deciduous tooth abnormality as the deciduous tooth condition; The abnormality detection module compares the permanent tooth eruption time Rh with the first preset permanent tooth eruption time Rh1 and the second preset permanent tooth eruption time Rh2, sets Rh1=3 months and Rh2=6 months, judges the permanent tooth eruption time according to the comparison results, and outputs the permanent tooth status according to the judgment results, wherein: When Rh≤Rh1, the abnormality detection module determines that the permanent tooth eruption time is abnormal and outputs the permanent tooth abnormality as the permanent tooth condition; When Rh1<Rh≤Rh2, the abnormality detection module determines that the permanent tooth eruption time is normal and outputs the normal permanent tooth as the permanent tooth condition; When Rh>Rh2, the abnormality detection module determines that the permanent tooth eruption time is abnormal and outputs the permanent tooth abnormality as the permanent tooth condition; The abnormality detection module compares the unchanged time of deciduous teeth Rrw with the preset unchanged time Rrw0, sets Rrw0=2 years, judges the unchanged time of deciduous teeth according to the comparison result, and outputs the unchanged situation of deciduous teeth according to the judgment result, wherein: When Rrw≤Rrw0, the abnormality detection module determines that the situation of no change in deciduous teeth is normal, and outputs the normal situation of no change in deciduous teeth as the situation of no change in deciduous teeth; When Rrw>Rrw0, the abnormality detection module determines that the deciduous teeth have no change time as abnormal, and outputs the abnormality of no change in deciduous teeth as the deciduous teeth have no change situation; The deciduous teeth condition, permanent teeth condition and no change in deciduous teeth condition are output as abnormal conditions of deciduous and permanent teeth.

9. The deep learning-based abnormal analysis system for children's deciduous and permanent teeth replacement period according to claim 8, characterized in that: When the abnormality detection module obtains the abnormality risk level according to the abnormality of the deciduous and permanent teeth, the abnormality risk score Rp is obtained according to the deciduous tooth abnormality parameter R1, the permanent tooth abnormality parameter R2, the unchanged abnormality parameter R3, the deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2 and the unchanged abnormality weight β3, and Rp=β1×R1+β2×R2+β3×R3 is set to obtain the abnormality risk score Rp. The abnormality risk score Rp is compared with the preset abnormality risk score Rp0, and the abnormality risk degree is judged according to the comparison result, and the abnormality risk level is output according to the judgment result, wherein: When Rp≤Rp0, the anomaly detection module determines that the abnormal risk level is normal and outputs no risk as the abnormal risk level; When Rp>Rp0, the abnormality detection module determines that the abnormality risk degree is abnormal and outputs risk as the abnormality risk level.

10. The deep learning-based children's deciduous and permanent tooth replacement abnormality analysis system according to claim 9, characterized in that: When the abnormality detection module adjusts the abnormality risk level acquisition process according to the child's age, the child's age Nr is compared with the preset child age Nr0, and the child's age is judged according to the comparison result. The deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3 are adjusted according to the judgment result, wherein: When Nr≤Nr0, the anomaly detection module determines that the child is young and does not adjust the level of the deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3; When Nr>Nr0, the abnormality detection module determines that the child is old, and adjusts the grade of the deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2, and the unchanged abnormality weight β3. The deciduous tooth abnormality weight β1 and the unchanged abnormality weight β3 are adjusted according to the grade adjustment coefficient uop, and 1.1<uop<1.4 is set to obtain the adjusted deciduous tooth abnormality weight β1` and the adjusted unchanged abnormality weight β3`, and β1`=β1×uop, β3`=(uop-1)×β3 are set. The permanent tooth abnormality weight β2 is graded and β2=1-β1-β3 to obtain the adjusted permanent tooth abnormality weight β2`. The deciduous tooth abnormality weight β1, the permanent tooth abnormality weight β2, and the changed abnormality weight β3 are replaced with the adjusted deciduous tooth abnormality weight β1`, the adjusted permanent tooth abnormality weight β2`, and the adjusted unchanged abnormality weight β3`, and the abnormality risk score is recalculated.

Images