CN115937153A - Model training method and device, electronic equipment and computer storage medium - Google Patents
Model training method and device, electronic equipment and computer storage medium Download PDFInfo
- Publication number
- CN115937153A CN115937153A CN202211594026.4A CN202211594026A CN115937153A CN 115937153 A CN115937153 A CN 115937153A CN 202211594026 A CN202211594026 A CN 202211594026A CN 115937153 A CN115937153 A CN 115937153A
- Authority
- CN
- China
- Prior art keywords
- image
- image set
- rule
- data enhancement
- oral cavity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012549 training Methods 0.000 title claims abstract description 101
- 238000000034 method Methods 0.000 title claims abstract description 94
- 238000012545 processing Methods 0.000 claims abstract description 44
- 238000007781 pre-processing Methods 0.000 claims abstract description 28
- 238000012952 Resampling Methods 0.000 claims abstract description 12
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 210000000214 mouth Anatomy 0.000 claims description 59
- 230000011218 segmentation Effects 0.000 claims description 23
- 238000004891 communication Methods 0.000 claims description 14
- 230000009466 transformation Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 2
- 230000008569 process Effects 0.000 description 15
- 238000010586 diagram Methods 0.000 description 6
- 238000007408 cone-beam computed tomography Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000002372 labelling Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000010146 3D printing Methods 0.000 description 1
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011022 operating instruction Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000001915 proofreading effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 210000003625 skull Anatomy 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The application provides a model training method, a device, an electronic device and a computer storage medium, wherein the model training method comprises the following steps: acquiring a first image set, wherein the first image set comprises at least two images; preprocessing the first image set to obtain a second image set, wherein the preprocessing comprises at least one of data attribute matching, image cropping, image resampling and image normalization; performing data enhancement processing on the second image set to obtain a third image set, wherein the third image set comprises at least three images; and training a model to be trained according to the third image set. The model trained by the model training method provided by the application has high applicability.
Description
Technical Field
The present application relates to the field of model training technologies, and in particular, to a model training method and apparatus, an electronic device, and a computer storage medium.
Background
With the rapid innovation of artificial intelligence and the development of cone beam computed tomography, intraoral and facial scanners, and three-dimensional printing of teeth, digital dentistry is rapidly developing. Digital dentistry has improved the efficiency of dentists, has improved the accuracy of orthodontic diagnosis, treatment and operation manual. An essential component of digital dentistry is the three-dimensional segmentation of teeth, jaw bones and skull from CBCT images, in addition to the precise tooth shape that facilitates the evaluation simulation in medicine.
At present, a trained three-dimensional segmentation model is adopted to segment a CBCT image of an oral cavity to obtain a three-dimensional model of the oral cavity.
However, the three-dimensional segmentation model trained by the existing training method has poor robustness and unsatisfactory segmentation effect on singular data, so that the segmented oral three-dimensional model needs to be manually adjusted when the model is used, and the model trained by the existing model training method has low applicability.
Disclosure of Invention
In view of the above, the present application provides a model training method, apparatus, electronic device and computer storage medium to at least partially solve the above problems.
According to a first aspect of the present application, there is provided a model training method, the method comprising: acquiring a first image set, wherein the first image set comprises at least two images; preprocessing the first image set to obtain a second image set, wherein the preprocessing comprises at least one of data attribute matching, image cropping, image resampling and image normalization; performing data enhancement processing on the second image set to obtain a third image set, wherein the third image set comprises at least three images; and training a model to be trained according to the third image set.
In a possible implementation manner, the performing data enhancement processing on the second image set to obtain a third image set includes: acquiring a data enhancement rule set, wherein the data enhancement rule set comprises at least two image set expansion rules, each image set expansion rule has a corresponding selected probability, and the sum of the selected probabilities corresponding to the image set expansion rules is equal to 1; randomly extracting at least one image from the second image set to serve as a first source image, determining a first image set expansion rule from the data enhancement rule set according to the selected probability corresponding to each image set expansion rule in the data enhancement rule set, and processing each extracted first source image according to the first image set expansion rule to obtain a first target image; updating the second image set to enable the first target image to be used as an element of the updated second image set; generating the third image set comprising the images in the updated second image set.
In one possible implementation, the obtaining data enhancement rule set includes: grouping at least two image set expansion rules included in an expansion rule base to obtain at least two expansion rule groups, wherein each expansion rule group comprises at least two image set expansion rules; determining a random code corresponding to each extended rule group, wherein different extended rule groups correspond to different random codes; generating a target random code by a random code generator created in advance; and determining the extended rule group corresponding to the random code as the target random code as the data enhancement rule set.
In one possible implementation, the method further includes: generating a corresponding number of random numbers according to the number of the image set expansion rules in the data enhancement rule set, wherein the sum of the generated random numbers is equal to 1; and according to a preset distribution rule, distributing each generated random number to each image set expansion rule in the data enhancement rule set as a selected probability corresponding to the image set expansion rule.
In one possible implementation, the method further includes: randomly extracting at least one image from the updated second image set to serve as a second source image, determining a second image set expansion rule from the data enhancement rule set according to the selected probability corresponding to each image set expansion rule in the data enhancement rule set, and respectively processing each extracted second source image according to the second image set expansion rule to obtain a second target image; updating the second image set including the first target image, so that the second target image is used as an element of the updated second image set; generating the third image set comprising the updated images in the second image set.
In one possible implementation, the image set expansion rules include image rotation, image scaling, image mirroring, gao Sijia noise, gaussian blur, luminance adjustment, contrast transformation, or grayscale transformation.
According to a second aspect of the present application, there is provided a tooth segmentation model training method, the method comprising: acquiring a first oral cavity image set, wherein the first oral cavity image set comprises at least two oral cavity images shot by at least two shooting sources; preprocessing the first oral cavity image set to obtain a second oral cavity image set, wherein the preprocessing comprises at least one of data attribute matching, oral cavity image cutting, oral cavity image resampling and oral cavity image standardization; performing data enhancement processing on the second oral cavity image set to obtain a third oral cavity image set, wherein the third oral cavity image set comprises at least three oral cavity images; and training the model to be trained according to the third oral cavity image set.
According to a third aspect of the present application, there is provided a model training apparatus comprising: a first acquisition module configured to acquire a first image set, wherein the first image set comprises at least two images; a first processing module, configured to perform preprocessing on the first image set to obtain a second image set, where the preprocessing includes at least one of data attribute matching, image cropping, image resampling, and image normalization; a second processing module, configured to perform data enhancement processing on the second image set to obtain a third image set, where the third image set includes at least three images; and the first training module is used for training a model to be trained according to the third image set.
According to a fourth aspect of the present application, there is provided an electronic device comprising: the system comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface are communicated with each other through the communication bus; the memory is used for storing at least one executable instruction, and the executable instruction causes the processor to execute the operation corresponding to the method according to the first aspect or the method according to the second aspect.
According to a fifth aspect of the present application, there is provided a computer storage medium having stored thereon a computer program which, when executed by a processor, performs the method according to the first aspect or the method according to the second aspect.
According to the model training method provided by the application, the model to be trained is trained according to the image set after preprocessing and data enhancement, so that the model to be trained is trained, and the image set used by the training model is preprocessed, so that the images in the image set used by the training model are more standard, the convergence rate of the model during model training can be improved, the model is trained to be Cheng Xiaolv more highly, and the model after training is higher in precision. Because the image set used by the training model is subjected to data enhancement, the diversity of the image set is expanded, and the model trained by the image set has higher robustness, so that the model trained by the model training method can process singular data, and the model trained by the model training method has higher applicability.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the description below are only some embodiments described in the embodiments of the present application, and other drawings can be obtained by those skilled in the art according to these drawings.
FIG. 1 is a flow chart of a model training method provided in an embodiment of the present application;
fig. 2 is a flowchart of a data enhancement method provided in an embodiment of the present application;
FIG. 3 is a flowchart of a tooth segmentation model training method provided by an embodiment of the present application;
FIG. 4 is a schematic diagram of a model training apparatus according to an embodiment of the present disclosure;
FIG. 5 is a schematic diagram of a training apparatus for a tooth segmentation model provided in an embodiment of the present application;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
In order to make those skilled in the art better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, but not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application shall fall within the scope of the protection of the embodiments in the present application.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this application and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
It should be understood that although the terms first, second, third, etc. may be used herein to describe various information, such information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present application. The word "if," as used herein, may be interpreted as "at … …" or "at … …" or "in response to a determination," depending on the context.
Fig. 1 is a flowchart of a model training method provided in an embodiment of the present application, and as shown in fig. 1, the method includes the following steps 101 to 104:
The first set of images includes at least two images used to train the model, such as: if the street view labeling capability of the model is trained, the image is a street view image, and the first image set is a street view image set; if the recognition capability of the model is to be trained, the image is a person or animal image, the first image set is a person or animal image set, and the like.
And preprocessing the first image set, wherein the preprocessing comprises at least one step of data attribute matching, image cropping, image resampling and image standardization, and each image in the preprocessed first image set is determined as a second image set.
Illustratively, the specific preprocessing is as follows, the data attribute matching is the consistency of the information of the geometric attributes of the proofreading image data (X) and the label data (Y), such as coordinate system, origin, direction, size and size, etc., and the label abnormal value is checked and deleted. And the image is cut to a non-zero area by image cutting, so that the image size is reduced, the calculation speed is improved, and the segmentation precision is not influenced. The image resampling is to resample all image data to the same space size (l, m, n), the value can be selected as a median value of the data space size after respectively removing the maximum and minimum 5% data, the image data can adopt a 3-order Spline interpolation method, and the image labeling data sampling nearest neighbor interpolation. Image normalization can cause all data gray values to have the same distribution. It should be understood that the above description is intended as an example only, and not as a limitation on the present application in any way.
And 103, performing data enhancement processing on the second image set to obtain a third image set.
And performing data enhancement on the second image set, wherein the data enhancement process is a process for expanding the image set, so that the number of images in the third image set is greater than that in the second image set, namely, the third image set comprises at least three images.
And 104, training the model to be trained according to the third image set.
And training the model to be trained according to the image set subjected to preprocessing and data enhancement, so that the trained model can complete tasks to be completed.
For example, during training, the following parameters can be set, and the model training mode is as follows: using five-fold cross validation, four-fold for training and one-fold for testing; loss function: adopting weighting sum of dice loss and cross entropy loss; setting training parameters: adam optimizer, learning Rate 3 -4 Training for 500-1000 rounds; learning rate adjustment strategy: calculating the exponential moving average loss of the training set and the verification set, if the exponential moving average loss of the training set is reduced by not enough 5 in 30 rounds -3 The learning rate is attenuated by 5 times; training stopping conditions: when the exponential moving average loss of the verification set is reduced by not enough 5 in 60 rounds -3 Or learning rate less than e -6 The training is stopped, and it should be noted that this example is used only for illustration and does not limit the present application at all, and the specific model training mode is not limited in the present application.
In the embodiment of the application, the model to be trained is trained according to the image set after preprocessing and data enhancement, so that the model to be trained is trained, and the image set used by the training model is preprocessed, so that the images in the image set used by the training model are more standard, the convergence rate of the model during model training can be improved, the model trained is Cheng Xiaolv is higher, and the trained model is higher in precision. Because the image set used by the training model is subjected to data enhancement, the diversity of the image set is expanded, and the model trained by the image set has higher robustness, so that the model trained by the model training method can process singular data, and the model trained by the model training method has higher applicability.
Fig. 2 is a flowchart of a data enhancement method according to an embodiment of the present application, and as shown in fig. 2, when performing data enhancement processing on a second image set to obtain a third image set, the following steps 201 to 204 may be performed:
The data enhancement rule set includes at least two image set expansion rules indicating an operation method for expanding the image data, each image set expansion rule having a corresponding selected probability, and a sum of the selected probabilities for each image set expansion rule is equal to 1, for example: if the data enhancement rule set includes five image set expansion rules, the sum of the five corresponding selected probabilities is 1, e.g., each image set expansion rule is 0.2, etc.
And randomly extracting one or more images in the second image set, performing data enhancement operation by using the extracted images as the first source images to obtain the first target image, wherein the specific data enhancement operation can be to select an image set expansion rule from the data enhancement rule set according to the selected probability corresponding to the image set expansion rule, and perform enhancement operation on the first source image according to the selected image set expansion rule.
And step 203, updating the second image set to enable the first target image to be used as an element of the updated second image set.
And the first target image obtained after the data enhancement is carried out on the image is placed back to the second image set to achieve the purpose of updating the second image set, and at the moment, the second image set comprises a source image without the data enhancement, the first target image and a first source image corresponding to the first target image and before the data enhancement.
And taking the images included in the updated second image set, namely the source image without data enhancement, the first target image and the first source image corresponding to the first target image before data enhancement as a third image set.
In the embodiment of the application, the data enhancement rule set is obtained, and the image set extension rule included in the rule set is used for performing data enhancement on the image in the second image set to obtain the third image set, so that the data enhancement operation is realized, and the image in the image set is extended according to the image set extension rule, so that the trained model has higher robustness, and the model can be suitable for processing singular data, and therefore, the model trained by the model training method has higher applicability.
In a possible implementation manner, when the data enhancement rule set is obtained, at least two image set extension rules included in the extension rule base may be grouped to obtain at least two extension rule sets, where each extension rule set includes at least two image set extension rules, and then a random code corresponding to each extension rule set is determined, where different extension rule sets correspond to different random codes, and a target random code is generated by a pre-created random code generator, and the corresponding random code is determined as the extension rule set of the target random code and determined as the data enhancement rule set.
All the image set expansion rules included in the expansion rule base are grouped, each group comprises at least two image set expansion rules, and the image set expansion rules in different groups are different, namely the number is different or the image set expansion rules are different.
When grouping, the expansion rules of the image set are not divided into sequence, each expansion rule group generates random codes randomly when grouping is completed, and the random codes corresponding to each expansion rule group are different. Randomly extracting a random code from a random code library including random codes of all expansion rule groups according to a preset random code generator, and determining the expansion rule group corresponding to the random code as a data enhancement rule set, wherein the data enhancement rule set comprises at least two image set expansion rules.
In the embodiment of the application, the expansion rules of the image sets in the expansion rule base are grouped, and random codes are generated for each group, so that the acquisition of the data enhancement rule set is realized.
In a possible implementation manner, when the data enhancement rule set is obtained, a corresponding number of random numbers may be generated according to the number of image set extension rules in the data enhancement rule set, wherein the sum of the generated random numbers is equal to 1, and then the generated random numbers are assigned to the image set extension rules in the data enhancement rule set according to a preset assignment rule as the selected probabilities corresponding to the image set extension rules.
According to the number of the image set expansion rules, generating a corresponding number of random numbers, such as: four random numbers are generated if the data enhancement rule set includes 4 image set expansion rules. The sum between all the random numbers generated is 1, for example: 4 random numbers, each of which may be 0.25, such that the sum is 1; alternatively, the four random numbers are 0.1, 0.3, etc.
The generated random number is assigned to each image set expansion rule. For example, the assignment process may order the assigned IDs included in the image expansion rules from large to small, order the random numbers from large to small, and then assign the random numbers sequentially, or may randomly generate a random number for each image set expansion rule, where the random number is a positive integer in the range of [1,n ], where n is the number of image set expansion rules, and different image set expansion rules correspond to different random numbers, and order the random numbers from large to small, and then assign the random numbers sequentially, and so on. It should be noted that the above examples only show two methods for implementing the process, and do not set any limit to the present application. And determining the distributed random number as the selected probability of the corresponding image set expansion rule.
In the embodiment of the application, a plurality of random numbers with the corresponding number and the sum of 1 are generated according to the number of the image set expansion rules, and the random numbers are used as corresponding selection probabilities, so that data enhancement can be performed on the images in the second image set according to the selection probabilities, and the selection probabilities of the image set expansion rules are random because the selection probabilities are random numbers generated at random, so that the processing methods of all target images in the generated third image set are different, the diversity of images used for training the model is increased, and the robustness of the model can be improved.
In a possible implementation manner, the data enhancement method further includes randomly extracting at least one image from the updated second image set as a second source image, determining a second image set expansion rule from the data enhancement rule set according to a selected probability corresponding to each image set expansion rule in the data enhancement rule set, respectively processing each extracted second source image according to the second image set expansion rule to obtain a second target image, then updating the second image set including the first target image, enabling the second target image to serve as an element of the updated second image set, and generating a third image set including each image in the updated second image set.
And randomly extracting a second source image from a second image set which comprises the updated first target image, wherein the second source image can be the first target image, can be the first source image corresponding to the first target image, and can also be a source image which is not subjected to data enhancement processing in the second image set, namely, the same image can be subjected to data enhancement processing for multiple times.
And putting a second target image obtained by data processing of the second source image back to a second image set comprising the first target image, and obtaining a third image set, wherein the third image set comprises the first source image, the first target image, the second target image and a source image without data enhancement.
In the data enhancement process, data enhancement can be performed on the same image for multiple times, data enhancement can also be performed on the target image subjected to data enhancement again, and the second image set is updated according to the image obtained after the data enhancement is performed, so that the second image set comprises the target image obtained after the data enhancement, namely the third image set.
In the embodiment of the application, the updated second image set including the first target image is subjected to data enhancement processing continuously to obtain the third image set, so that the diversity of training samples in the third image set is further improved, the accuracy and the robustness of the trained model are improved, and the model can be suitable for singular data.
In one possible implementation, the image set expansion rules include image rotation, image scaling, image mirroring, gao Sijia noise, gaussian blur, luminance adjustment, contrast transformation, or grayscale transformation.
Illustratively, the specific operation may be an operation of image rotation: the rotation angles of the x axis, the y axis and the z axis of the image are uniformly and randomly taken between (-30, 30). Image zooming: by multiplying the image coordinates with a scaling factor in the voxel grid. A scale factor less than 1 produces a "zoom out" effect, and a scale factor greater than 1 produces a "zoom in" effect. The scale factors are sampled from (0.7,1.4) uniformly and randomly. Gaussian noise: additive gaussian noise centered at zero is added independently to each voxel in the sample, and the variance of the noise is extracted from the uniform distribution U (0,0.1). Gaussian blur: the width of the gaussian kernel is uniformly randomly sampled (0.5,1.5). And (3) brightness processing: the voxel intensities are multiplied by the values sampled uniformly and randomly at (0.7,1.3). Contrast processing: the voxel intensities are multiplied by the uniformly randomly sampled value at (0.65,1.5) and then clipped to their original intensity range. Gray value transformation: and carrying out nonlinear operation on the gray value of the input image to enable the gray value of the output image to be in an exponential relation with the gray value of the input image.
It should be noted that the above example only gives an example of each operation, and should not cause any limitation to the present application, and the specific operation process is not limited in the present application.
In the embodiment of the application, the image set expansion rule includes image rotation, image scaling, image mirroring, gao Sijia noise, gaussian blur, brightness adjustment, contrast transformation or gray value transformation, and a rule for enhancing the data of the image data is defined, so that the image data included in the image set is more diverse, and the sample data set is expanded, so that the model trained by the image set has higher robustness.
Fig. 3 is a flowchart of a tooth segmentation model training method provided in an embodiment of the present application, and as shown in fig. 3, the method includes the following steps 301 to 304:
The method comprises the steps of acquiring oral cavity images shot by shooting equipment of multiple brands in multiple hospitals, wherein the oral cavity images comprise but are not limited to CBCT images of tooth parts, and the shooting equipment comprises but is not limited to CT cameras of different brands.
And 303, performing data enhancement processing on the second oral cavity image set to obtain a third oral cavity image set, wherein the third oral cavity image set comprises at least three oral cavity images.
And step 304, training the model to be trained according to the third oral cavity image set.
In this application embodiment, oral cavity image set after according to preliminary treatment and data enhancement trains waiting to train the model, the training of waiting to train the model has been realized, because oral cavity image set is concentrated including the oral cavity image that a plurality of shooting sources were shot, consequently, the tooth segmentation model adaptability of training out is higher, and carried out the preliminary treatment to the oral cavity image set that trains the model and use, thereby the oral cavity image that trains the model and uses is more standard, convergence rate when can promoting the model training, make the efficiency of model training process higher, the model precision after the training is higher. The oral cavity image set used by the training model is subjected to data enhancement, the oral cavity image set is expanded, the robustness of the tooth segmentation model trained by the oral cavity image set is high, the model can be suitable for processing singular tooth image data, and the condition that a doctor manually segments and draws the tooth image is avoided, so that the model training method has high applicability.
Fig. 4 is a schematic diagram of a model training apparatus provided in an embodiment of the present application, and as shown in fig. 4, the apparatus 400 includes:
a first obtaining module 401, configured to obtain a first image set, where the first image set includes at least two images;
a first processing module 402, configured to perform preprocessing on the first image set to obtain a second image set, where the preprocessing includes at least one of data attribute matching, image cropping, image resampling, and image normalization;
a second processing module 403, configured to perform data enhancement processing on the second image set to obtain a third image set, where the third image set includes at least three images;
and a first training module 404, configured to train the model to be trained according to the third image set.
In this embodiment, the first obtaining module 401 may be configured to perform step 101 in the above-described method embodiment, the first processing module 402 may be configured to perform step 102 in the above-described method embodiment, the second processing module 403 may be configured to perform step 103 in the above-described method embodiment, and the first training module 404 may be configured to perform step 104 in the above-described method embodiment.
In a possible implementation manner, the second processing module 403 may be further configured to obtain a data enhancement rule set, where the data enhancement rule set includes at least two image set extension rules, each image set extension rule has a corresponding selected probability, and a sum of the selected probabilities corresponding to each image set extension rule is equal to 1; randomly extracting at least one image from a second image set as a first source image, determining a first image set expansion rule from the data enhancement rule set according to the selected probability corresponding to each image set expansion rule in the data enhancement rule set, and respectively processing each extracted first source image according to the first image set expansion rule to obtain a first target image; updating the second image set to enable the first target image to be used as an element of the updated second image set; a third image set is generated that includes images in the updated second image set.
In a possible implementation manner, the first obtaining module 401 may be further configured to group at least two image set expansion rules included in the expansion rule base to obtain at least two expansion rule groups, where each expansion rule group includes at least two image set expansion rules; determining random codes corresponding to each expansion rule group, wherein different expansion rule groups correspond to different random codes; generating a target random code by a random code generator created in advance; and determining the expansion rule group corresponding to the random code as the target random code as a data enhancement rule set.
In a possible implementation manner, the first obtaining module 401 may be further configured to generate a corresponding number of random numbers according to the number of the image set expansion rules in the data enhancement rule set, where a sum of the generated random numbers is equal to 1; and according to a preset allocation rule, allocating each generated random number to each image set expansion rule in the data enhancement rule set as a selected probability corresponding to the image set expansion rule.
In a possible implementation manner, the second processing module 403 may be further configured to randomly extract at least one image from the updated second image set as a second source image, determine a second image set expansion rule from the data enhancement rule set according to the selected probability corresponding to each image set expansion rule in the data enhancement rule set, and process each extracted second source image according to the second image set expansion rule to obtain a second target image; updating a second image set comprising the first target image, and enabling the second target image to be used as an element of the updated second image set; a third image set is generated that includes images in the updated second image set.
In one possible implementation, the image set expansion rules include image rotation, image scaling, image mirroring, gao Sijia noise, gaussian blur, luminance adjustment, contrast transformation, or grayscale transformation.
Fig. 5 is a schematic diagram of a tooth segmentation model training device provided in an embodiment of the present application, and as shown in fig. 5, the device 500 includes:
a second obtaining module 501, configured to obtain a first oral cavity image set, where the first oral cavity image set includes at least two oral cavity images captured by at least two capturing sources;
a third processing module 502, configured to perform preprocessing on the first oral image set to obtain a second oral image set, where the preprocessing includes at least one of data attribute matching, oral image clipping, oral image resampling, and oral image normalization;
a fourth processing module 503, configured to perform data enhancement processing on the second oral cavity image set to obtain a third oral cavity image set, where the third oral cavity image set includes at least three oral cavity images;
and a second training module 504, configured to train the model to be trained according to the third oral cavity image set.
In this embodiment, the second obtaining module 501 may be configured to perform step 301 in the foregoing method embodiment, the third processing module 502 may be configured to perform step 302 in the foregoing method embodiment, the fourth processing module 503 may be configured to perform step 103 in the foregoing method embodiment, and the second training module 504 may be configured to perform step 104 in the foregoing method embodiment.
It should be noted that, because the contents of information interaction, execution process, and the like between the modules in the model training device and the tooth segmentation model training device are based on the same concept as the aforementioned embodiment of the model training method and the tooth segmentation model training device, specific contents can be referred to the description in the aforementioned embodiment of the model training method, and are not repeated here.
Referring to fig. 6, a schematic structural diagram of an electronic device according to an embodiment of the present application is shown, and the specific embodiment of the present application does not limit a specific implementation of the electronic device.
As shown in fig. 6, the electronic device may include: a processor (processor) 602, a communication Interface 604, a memory 606, and a communication bus 608.
Wherein:
processor 602, communication interface 604, and memory 606 communicate with one another via a communication bus 608.
A communication interface 604 for communicating with other electronic devices or servers.
The processor 602, configured to execute the program 610, may specifically execute relevant steps in the above-described model training method or the tooth segmentation model training method embodiment.
In particular, program 610 may include program code comprising computer operating instructions.
Processor 602 may be a Central Processing Unit (CPU), or a Graphics Processing Unit (GPU), or an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits configured to implement embodiments of the present Application. The intelligent device comprises one or more processors which can be the same type of processor, such as one or more CPUs; one or more GPUs; or may be different types of processors, such as one or more CPUs and one or more GPUs and one or more ASICs.
And a memory 606 for storing a program 610. Memory 606 may comprise high-speed RAM memory, and may also include non-volatile memory (non-volatile memory), such as at least one disk memory.
The program 610 may specifically be configured to cause the processor 602 to perform a model training method or a tooth segmentation model training method in any of the embodiments described above.
For specific implementation of each step in the program 610, reference may be made to corresponding descriptions in corresponding steps and units in any one of the foregoing embodiments of the model training method or the tooth segmentation model training method, which are not described herein again. It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described devices and modules may refer to the corresponding process descriptions in the foregoing method embodiments, and are not described herein again.
In the embodiment of the application, the model to be trained is trained according to the image set after preprocessing and data enhancement, so that the model to be trained is trained, and the image set used by the training model is preprocessed, so that the images in the image set used by the training model are more standard, the convergence rate of the model during model training can be improved, the model trained is Cheng Xiaolv is higher, and the trained model is higher in precision. Because the image set used by the training model is subjected to data enhancement, the diversity of the image set is expanded, and the model trained by the image set has higher robustness, so that the model trained by the model training method can process singular data, and the model trained by the model training method has higher applicability.
The embodiment of the present application further provides a computer program product, which includes computer instructions for instructing a computing device to execute an operation corresponding to any one of the methods in the foregoing method embodiments.
It should be noted that, according to the implementation requirement, each component/step described in the embodiment of the present application may be divided into more components/steps, and two or more components/steps or partial operations of the components/steps may also be combined into a new component/step to achieve the purpose of the embodiment of the present application.
The above-described methods according to embodiments of the present application may be implemented in hardware, firmware, or as software or computer code storable in a recording medium such as a CD ROM, a RAM, a floppy disk, a hard disk, or a magneto-optical disk, or as computer code originally stored in a remote recording medium or a non-transitory machine-readable medium downloaded through a network and to be stored in a local recording medium, so that the methods described herein may be stored in such software processes on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware such as an ASIC or FPGA. It will be appreciated that the computer, processor, microprocessor controller or programmable hardware includes memory components (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code that, when accessed and executed by the computer, processor or hardware, implements the model training method or tooth segmentation model training method described herein. Further, when a general-purpose computer accesses code for implementing the model training method or the tooth segmentation model training method illustrated herein, execution of the code transforms the general-purpose computer into a special-purpose computer for performing the model training method or the tooth segmentation model training method illustrated herein.
Those of ordinary skill in the art will appreciate that the various illustrative elements and method steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the technical solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the embodiments of the present application.
The above embodiments are only used for illustrating the embodiments of the present application, and not for limiting the embodiments of the present application, and those skilled in the relevant art can make various changes and modifications without departing from the spirit and scope of the embodiments of the present application, so that all equivalent technical solutions also belong to the scope of the embodiments of the present application, and the scope of patent protection of the embodiments of the present application should be defined by the claims.
Claims (10)
1. A method of model training, comprising:
acquiring a first image set, wherein the first image set comprises at least two images;
preprocessing the first image set to obtain a second image set, wherein the preprocessing comprises at least one of data attribute matching, image cropping, image resampling and image standardization;
performing data enhancement processing on the second image set to obtain a third image set, wherein the third image set comprises at least three images;
and training a model to be trained according to the third image set.
2. The method of claim 1, wherein said enhancing the data of the second image set to obtain a third image set comprises:
acquiring a data enhancement rule set, wherein the data enhancement rule set comprises at least two image set extension rules, each image set extension rule has a corresponding selected probability, and the sum of the selected probabilities corresponding to each image set extension rule is equal to 1;
randomly extracting at least one image from the second image set to serve as a first source image, determining a first image set expansion rule from the data enhancement rule set according to the selected probability corresponding to each image set expansion rule in the data enhancement rule set, and processing each extracted first source image according to the first image set expansion rule to obtain a first target image;
updating the second image set to enable the first target image to be used as an element of the updated second image set;
generating the third image set comprising the updated images in the second image set.
3. The method of claim 2, wherein obtaining the data enhancement rule set comprises:
grouping at least two image set expansion rules included in an expansion rule base to obtain at least two expansion rule groups, wherein each expansion rule group comprises at least two image set expansion rules;
determining a random code corresponding to each extended rule group, wherein different extended rule groups correspond to different random codes;
generating a target random code by a random code generator created in advance;
and determining the expansion rule group corresponding to the random code as the target random code as the data enhancement rule set.
4. The method of claim 3, further comprising:
generating a corresponding number of random numbers according to the number of the image set expansion rules in the data enhancement rule set, wherein the sum of the generated random numbers is equal to 1;
and according to a preset distribution rule, distributing each generated random number to each image set expansion rule in the data enhancement rule set as a selected probability corresponding to the image set expansion rule.
5. The method of claim 2, further comprising:
randomly extracting at least one image from the updated second image set to serve as a second source image, determining a second image set expansion rule from the data enhancement rule set according to the selected probability corresponding to each image set expansion rule in the data enhancement rule set, and respectively processing each extracted second source image according to the second image set expansion rule to obtain a second target image;
updating the second image set comprising the first target image, so that the second target image is used as an element of the updated second image set;
generating the third image set comprising the updated images in the second image set.
6. The method of any of claims 2-5, wherein the image set expansion rules include image rotation, image scaling, image mirroring, gao Sijia noise, gaussian blur, luminance adjustment, contrast transformation, or grayscale transformation.
7. A tooth segmentation model training method is characterized by comprising the following steps:
acquiring a first oral cavity image set, wherein the first oral cavity image set comprises at least two oral cavity images shot by at least two shooting sources;
preprocessing the first oral cavity image set to obtain a second oral cavity image set, wherein the preprocessing comprises at least one of data attribute matching, oral cavity image cutting, oral cavity image resampling and oral cavity image standardization;
performing data enhancement processing on the second oral cavity image set to obtain a third oral cavity image set, wherein the third oral cavity image set comprises at least three oral cavity images;
and training the model to be trained according to the third oral cavity image set.
8. A model training apparatus, comprising:
a first acquisition module configured to acquire a first image set, wherein the first image set comprises at least two images;
a first processing module, configured to perform preprocessing on the first image set to obtain a second image set, where the preprocessing includes at least one of data attribute matching, image cropping, image resampling, and image normalization;
a second processing module, configured to perform data enhancement processing on the second image set to obtain a third image set, where the third image set includes at least three images;
and the first training module is used for training a model to be trained according to the third image set.
9. An electronic device, comprising: the system comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete mutual communication through the communication bus;
the memory is configured to store at least one executable instruction that causes the processor to perform the model training method of any one of claims 1-6 or the tooth segmentation model training method of claim 7.
10. A computer storage medium, characterized in that a computer program is stored thereon which, when being executed by a processor, carries out a model training method according to any one of claims 1 to 6 or a tooth segmentation model training method according to claim 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211594026.4A CN115937153A (en) | 2022-12-13 | 2022-12-13 | Model training method and device, electronic equipment and computer storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211594026.4A CN115937153A (en) | 2022-12-13 | 2022-12-13 | Model training method and device, electronic equipment and computer storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115937153A true CN115937153A (en) | 2023-04-07 |
Family
ID=86650544
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211594026.4A Pending CN115937153A (en) | 2022-12-13 | 2022-12-13 | Model training method and device, electronic equipment and computer storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115937153A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116580282A (en) * | 2023-07-12 | 2023-08-11 | 四川大学华西医院 | Neural network model-based pressure injury staged identification system and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111242063A (en) * | 2020-01-17 | 2020-06-05 | 江苏大学 | Small sample classification model construction method based on transfer learning and iris classification application |
| US20210166066A1 (en) * | 2019-01-15 | 2021-06-03 | Olympus Corporation | Image processing system and image processing method |
| CN113191431A (en) * | 2021-04-29 | 2021-07-30 | 武汉工程大学 | Fine-grained vehicle type identification method and device and storage medium |
| CN113420792A (en) * | 2021-06-03 | 2021-09-21 | 阿波罗智联(北京)科技有限公司 | Training method of image model, electronic equipment, road side equipment and cloud control platform |
| CN114511041A (en) * | 2022-04-01 | 2022-05-17 | 北京世纪好未来教育科技有限公司 | Model training method, image processing method, device, equipment and storage medium |
| CN114511758A (en) * | 2022-01-28 | 2022-05-17 | 北京百度网讯科技有限公司 | Image recognition method and device, electronic device and medium |
| KR20220085737A (en) * | 2020-12-15 | 2022-06-22 | 박지환 | Method of auto generating and analysis data based on rule-set |
| US20220261965A1 (en) * | 2020-05-18 | 2022-08-18 | Tencent Technology (Shenzhen) Company Limited | Training method of image processing model, image processing method, apparatus, and device |
-
2022
- 2022-12-13 CN CN202211594026.4A patent/CN115937153A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210166066A1 (en) * | 2019-01-15 | 2021-06-03 | Olympus Corporation | Image processing system and image processing method |
| CN111242063A (en) * | 2020-01-17 | 2020-06-05 | 江苏大学 | Small sample classification model construction method based on transfer learning and iris classification application |
| US20220261965A1 (en) * | 2020-05-18 | 2022-08-18 | Tencent Technology (Shenzhen) Company Limited | Training method of image processing model, image processing method, apparatus, and device |
| KR20220085737A (en) * | 2020-12-15 | 2022-06-22 | 박지환 | Method of auto generating and analysis data based on rule-set |
| CN113191431A (en) * | 2021-04-29 | 2021-07-30 | 武汉工程大学 | Fine-grained vehicle type identification method and device and storage medium |
| CN113420792A (en) * | 2021-06-03 | 2021-09-21 | 阿波罗智联(北京)科技有限公司 | Training method of image model, electronic equipment, road side equipment and cloud control platform |
| CN114511758A (en) * | 2022-01-28 | 2022-05-17 | 北京百度网讯科技有限公司 | Image recognition method and device, electronic device and medium |
| CN114511041A (en) * | 2022-04-01 | 2022-05-17 | 北京世纪好未来教育科技有限公司 | Model training method, image processing method, device, equipment and storage medium |
Non-Patent Citations (3)
| Title |
|---|
| ZHUN ZHONG 等: "Random Erasing Data Augmentation", ARXIV, 16 August 2017 (2017-08-16), pages 1 - 10 * |
| 刘瑞挺: "《计算机系统导论》", 30 June 1993, 高等教育出版社, pages: 82 - 86 * |
| 吴天雨;许英朝;晁鹏飞;: "一种提高目标图像识别准确率的数据增强技术", 激光杂志, no. 05, 25 May 2020 (2020-05-25), pages 100 - 104 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116580282A (en) * | 2023-07-12 | 2023-08-11 | 四川大学华西医院 | Neural network model-based pressure injury staged identification system and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP3591616A1 (en) | Automated determination of a canonical pose of a 3d dental structure and superimposition of 3d dental structures using deep learning | |
| EP3503038A1 (en) | Automated 3d root shape prediction using deep learning methods | |
| US20200320685A1 (en) | Automated classification and taxonomy of 3d teeth data using deep learning methods | |
| EP3111422B1 (en) | System and method for auto-contouring in adaptive radiotherapy | |
| CN112515787B (en) | Three-dimensional dental data analysis method | |
| CN110807775A (en) | Traditional Chinese medicine tongue image segmentation device and method based on artificial intelligence and storage medium | |
| CN115937153A (en) | Model training method and device, electronic equipment and computer storage medium | |
| CN111724389B (en) | Method, device, storage medium and computer equipment for segmenting CT image of hip joint | |
| CN113262070A (en) | Dental surgery equipment positioning method and system based on image recognition and storage medium | |
| US11715279B2 (en) | Weighted image generation apparatus, method, and program, determiner learning apparatus, method, and program, region extraction apparatus, method, and program, and determiner | |
| CN111723836A (en) | Image similarity calculation method and device, electronic equipment and storage medium | |
| CN110211200A (en) | A kind of arch wire generation method and its system based on nerual network technique | |
| KR20230164633A (en) | Apparatus and method for displaying three dimensional tooth image data and method for training same | |
| JP7202739B2 (en) | Device, method and recording medium for determining bone age of teeth | |
| CN117011318A (en) | Tooth CT image three-dimensional segmentation method, system, equipment and medium | |
| CN115294426B (en) | Method, device and equipment for tracking interventional medical equipment and storage medium | |
| CN115359257B (en) | Spine image segmentation method and operation navigation positioning system based on deep learning | |
| CN115760846A (en) | Medical image processing method, apparatus, electronic device and storage medium | |
| CN113658198A (en) | Interactive emphysema focus segmentation method, device, storage medium and equipment | |
| CN117315378B (en) | Grading judgment method for pneumoconiosis and related equipment | |
| CN113096117A (en) | Ectopic ossification CT image segmentation method, three-dimensional reconstruction method and device | |
| JP2022147713A (en) | Image generation device, learning device, and image generation method | |
| CN111161256A (en) | Image segmentation method, image segmentation device, storage medium, and electronic apparatus | |
| CN117058309B (en) | Image generation method and system based on oral imaging | |
| CN117726815B (en) | Small sample medical image segmentation method based on anomaly detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |