CN115049546A - Sample data processing method and device, electronic equipment and storage medium - Google Patents
Sample data processing method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN115049546A CN115049546A CN202210637331.0A CN202210637331A CN115049546A CN 115049546 A CN115049546 A CN 115049546A CN 202210637331 A CN202210637331 A CN 202210637331A CN 115049546 A CN115049546 A CN 115049546A
- Authority
- CN
- China
- Prior art keywords
- sample
- data set
- processed
- image
- sample data
- 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
- 238000003672 processing method Methods 0.000 title claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 168
- 238000012360 testing method Methods 0.000 claims abstract description 89
- 238000013145 classification model Methods 0.000 claims abstract description 85
- 238000012545 processing Methods 0.000 claims abstract description 52
- 238000000034 method Methods 0.000 claims abstract description 45
- 238000004590 computer program Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 description 16
- 230000008569 process Effects 0.000 description 12
- 208000026350 Inborn Genetic disease Diseases 0.000 description 11
- 208000016361 genetic disease Diseases 0.000 description 11
- 208000026928 Turner syndrome Diseases 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 8
- 230000003287 optical effect Effects 0.000 description 6
- 238000007781 pre-processing Methods 0.000 description 6
- 238000013136 deep learning model Methods 0.000 description 5
- 230000001815 facial effect Effects 0.000 description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 4
- 238000012790 confirmation Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 3
- 239000002131 composite material Substances 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000009499 grossing Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000013519 translation Methods 0.000 description 3
- 230000007423 decrease Effects 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 239000013307 optical fiber Substances 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 235000017965 Asarum canadense Nutrition 0.000 description 1
- 235000000385 Costus speciosus Nutrition 0.000 description 1
- 241000606265 Valeriana jatamansi Species 0.000 description 1
- 235000014687 Zingiber zerumbet Nutrition 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The embodiment of the invention discloses a sample data processing method, a sample data processing device, electronic equipment and a storage medium, wherein the method comprises the following steps: acquiring a sample medical image and a sample original label corresponding to the sample medical image; performing super-resolution processing on the sample medical image with the resolution lower than the preset resolution, and updating the sample medical image according to the sample medical image after the super-resolution processing; determining at least two target categories, and respectively determining a sub-sample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image; determining a sample data set to be processed in the sub-sample data set, and updating the sample data set to be processed through two-stage classification model training and testing for each sample data set to be processed; a sample data set is determined based on the updated sub-sample data set. The embodiment of the invention improves the image quality of the sample medical image and the accuracy of the original label of the sample.
Description
Technical Field
The present invention relates to data processing technologies, and in particular, to a method and an apparatus for processing sample data, an electronic device, and a storage medium.
Background
In the field of image classification, deep learning model-based classification has been widely used. When deep learning model training is performed, the quality of sample data has a large influence on the model effect.
When the deep learning model is trained, the problem that the quality of a sample image is poor or a sample label is inaccurate usually occurs. If the deep learning model is trained based on the original samples, the obtained image classification model may have the problems of poor generalization capability and low classification accuracy.
Disclosure of Invention
The embodiment of the invention provides a sample data processing method and device, electronic equipment and a storage medium, which are used for realizing the effects of improving the image quality of a sample medical image and the accuracy of a sample original label so as to better perform model training in the following process.
In a first aspect, an embodiment of the present invention provides a sample data processing method, where the method includes:
acquiring a sample medical image and a sample original label corresponding to the sample medical image;
performing super-resolution processing on the sample medical image with the resolution lower than the preset resolution, and updating the sample medical image according to the sample medical image after the super-resolution processing;
determining at least two target categories, and determining a sub-sample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image;
determining a sample data set to be processed in the sub-sample data set, and updating the sample data set to be processed through two-stage classification model training and testing aiming at each sample data set to be processed;
based on the updated subsample dataset, a sample dataset is determined.
In a second aspect, an embodiment of the present invention further provides a sample data processing apparatus, where the apparatus includes:
the system comprises a sample original data acquisition module, a sample medical image acquisition module and a sample original label, wherein the sample original data acquisition module is used for acquiring a sample medical image and a sample original label corresponding to the sample medical image;
the sample medical image updating module is used for performing super-resolution processing on the sample medical image with the resolution lower than the preset resolution and updating the sample medical image according to the sample medical image after the super-resolution processing;
the sub-sample data set splitting module is used for determining at least two target categories and determining a sub-sample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image;
the to-be-processed sample data set updating module is used for determining the to-be-processed sample data set in the sub-sample data set, and updating the to-be-processed sample data set by means of two-stage classification model training and testing aiming at each to-be-processed sample data set;
and the sample data set determining module is used for determining the sample data set based on the updated sub-sample data set.
In a third aspect, an embodiment of the present invention further provides an electronic device, where the electronic device includes:
one or more processors;
a storage device for storing one or more programs,
when the one or more programs are executed by the one or more processors, the one or more processors implement the sample data processing method according to any one of the embodiments of the present invention.
In a fourth aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the sample data processing method according to any one of the embodiments of the present invention.
The technical scheme of the embodiment of the invention comprises the steps of performing super-resolution processing on a sample medical image with a resolution lower than a preset resolution by acquiring the sample medical image and a sample original label corresponding to the sample medical image, updating the sample medical image according to the sample medical image after the super-resolution processing to improve the quality of the sample medical image, determining at least two target classes, determining a sub-sample data set corresponding to each target class according to the sample medical image corresponding to each target class and the sample original label corresponding to each sample medical image, determining a sample data set to be processed in the sub-sample data set, updating the sample data set to be processed according to each sample data set to be processed through two-stage classification model training and testing to improve the accuracy of the sample original label, and based on the updated sub-sample data set, the sample data set is determined, the problem that the sample data cannot be used for model training due to poor quality of the sample medical image and inaccuracy of the sample original label is solved, and the effects of improving the image quality of the sample medical image and the accuracy of the sample original label are achieved, so that model training can be better performed in the subsequent process.
Drawings
In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, a brief description is given below of the drawings used in describing the embodiments. It should be clear that the described figures are only views of some of the embodiments of the invention to be described, not all, and that for a person skilled in the art, other figures can be derived from these figures without inventive effort.
Fig. 1 is a schematic flowchart illustrating a sample data processing method according to an embodiment of the present invention;
fig. 2 is a schematic flowchart illustrating a sample data processing method according to a second embodiment of the present invention;
fig. 3 is a schematic diagram of an SRGAN according to a second embodiment of the present invention;
FIG. 4 is a schematic flowchart of a genetic disease facial recognition method according to a third embodiment of the present invention;
fig. 5 is a schematic structural diagram of a sample data processing apparatus according to a fourth embodiment of the present invention;
fig. 6 is a schematic structural diagram of an electronic device according to a fifth embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some structures related to the present invention are shown in the drawings, not all of them.
Example one
Fig. 1 is a schematic flow diagram of a sample data processing method according to an embodiment of the present invention, where the present embodiment is applicable to a situation where quality of sample data is improved before model training, and the method may be executed by a sample data processing apparatus, and the apparatus may be implemented in a form of software and/or hardware, where the hardware may be an electronic device, and optionally, the electronic device may be a mobile terminal, a PC terminal, a server, and the like.
As shown in fig. 1, the method of this embodiment specifically includes the following steps:
and S110, acquiring a sample medical image and a sample original label corresponding to the sample medical image.
Wherein the sample medical image may be an image containing a face region for medical classification according to the face region. The sample original label may be a label corresponding to the sample medical image, i.e., a label labeling a category to which the sample medical image belongs.
Specifically, the sample medical image and the sample original label corresponding to the sample medical image may be acquired through a public database, or an image that has been confirmed in daily diagnosis may be used as the sample medical image, and a diagnosis result for the sample medical image may be determined as the sample original label corresponding to the sample original image.
And S120, performing super-resolution processing on the sample medical image with the resolution lower than the preset resolution, and updating the sample medical image according to the sample medical image after the super-resolution processing.
Wherein the preset resolution may be a resolution required for the sample medical image determined according to the requirements of the subsequent model. The super-resolution processing may be a way of processing a low-resolution image into a high-resolution image.
It should be noted that the low resolution of the medical image of the sample affects the image quality, resulting in less obvious features in the image. The subsequent model is trained by utilizing the sample medical image with low resolution, so that the model is difficult to learn the characteristics of the sample medical image, and the problem of poor classification effect of the model is caused.
Specifically, for each sample medical image, the image resolution of the sample medical image may be identified and compared with a preset resolution. If the image resolution is greater than or equal to the preset resolution, the image quality of the sample medical image can meet the requirement of a subsequent model to be trained, and processing is not needed; if the image resolution is smaller than the preset resolution, the image quality of the sample medical image can be considered to be unable to meet the requirements of the subsequent model to be trained, so the sample medical image is subjected to super-resolution processing to improve the resolution of the sample medical image, and the processed high-resolution image is used as a new sample medical image to update the sample medical image.
S130, determining at least two target categories, and determining a sub-sample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image.
The target category may be a category to which each sample medical image belongs, that is, a category corresponding to the original label of the sample. Each object class corresponds to a sample original label, and each sample original label has an object class corresponding to the sample original label. The subsample dataset may be a sample dataset corresponding to the target category, i.e. one category for each subsample dataset.
Specifically, all the target categories are determined, for each target category, a sample medical image corresponding to the target category and a sample original label corresponding to each sample medical image may be acquired, and the acquired sample medical image and the sample original label are combined to determine a sub-sample data set corresponding to the target category.
S140, determining a sample data set to be processed in the sub-sample data set, and updating the sample data set to be processed through two-stage classification model training and testing aiming at each sample data set to be processed.
The sample data set to be processed may be one or more sub-sample data sets to be subjected to tag reconfirmation, for example: the sample original label effect detection method can be used for detecting the sample original label effect of the sample, and can be used for detecting the sample original label effect of the sample. The two-stage classification model can be a model which is classified through two-time data division and three-time model training and is used for determining whether the original label of the sample is correct or not.
Specifically, one or more sample data sets to be processed are determined from the sub-sample data sets to reconfirm the original tags of the samples. And processing the sample data set to be processed by the two-stage classification model aiming at each sample data set to be processed, re-determining the original sample label in the sample data set to be processed, and updating the sample data set to be processed according to the re-determined original sample label.
And S150, determining a sample data set based on the updated sub-sample data set.
Specifically, each sample data set to be processed is updated, that is, the sub-sample data set is updated, so the updated sub-sample data set may include the updated sample data set to be processed and the sub-sample data set that is not determined as the sample data set to be processed. Further, the updated combination of sub-sample data sets is determined as the sample data set.
On the basis of the above embodiments, optionally, the model training may be performed by using a sample data set:
and training the initial image recognition model according to the sample medical images in the sample data set and the sample original labels corresponding to the sample medical images, and taking the trained initial image recognition model as a target image recognition model.
The initial image recognition model is constructed based on a model scaling method which gives consideration to both speed and precision.
Specifically, the initial image recognition model is trained according to the sample medical images in the sample data set and the sample original labels corresponding to the sample medical images, so that appropriate parameters and/or hyper-parameters are obtained through training. For example, the training is stopped when the convergence condition is reached, and the target image recognition model is determined.
It should be noted that the model scaling method considering both speed and precision may be an EfficientNet, which amplifies the network from three dimensions of network depth, network width, and image resolution by a composite coefficient, thereby increasing the number of layers of the network, and acquiring more information and extracting more features.
Optionally, before model training using the sample data set, image preprocessing may be performed on sample medical images in the sample data set.
Wherein the image pre-processing includes at least one of data cleansing and image cropping.
Specifically, image preprocessing is performed on the sample medical images in the sample data set to improve the sample quality, so that the accuracy of model training is improved subsequently. The data cleansing may be to remove smaller size, image-obscuring medical images of the sample. Image cropping may be cropping of a facial region in a sample medical image to exclude interference of redundant information.
It should be noted that the size of the graph after image preprocessing needs to be adjusted to a size that fits the initial image recognition model.
Optionally, before model training is performed using the sample data set, image enhancement processing may be performed on sample medical images in the sample data set.
Wherein the image enhancement processing includes at least one of rotation, mirroring, sharpening, translation, color enhancement, brightness, saturation, and scaling.
Specifically, image enhancement is performed on the sample medical image in the sample data set to perform data amplification to increase the sample size in the sample data set.
It should be noted that, the sample number difference of each target class in the sample data set is large, for example: the small number is only dozens, and the large number is one thousand. The sample imbalance causes a problem that the accuracy of some target classes is low. In addition, there are great differences in resolution, size, color and human face pose of the image data. Therefore, the sample imbalance problem can be solved by using more than ten image processing methods of rotation, mirroring, sharpening, translation, color enhancement, brightness, saturation, and the like.
For example, the image enhancement methods can be used more for sample medical images with small sample size, and the image enhancement methods can be used less for sample medical images with large sample size, so that the sample size of each target category is the same or the difference amount is within a preset range, for example, 2500 ± 50 samples. By the image enhancement method, on one hand, the sample size of the sample data set can be increased, on the other hand, the overfitting degree of the model can be reduced, and the generalization capability of the model is enhanced.
Optionally, when the model training is performed by using the sample data set, the training mode is as follows: and training the initial image recognition model based on a training warm-up method and a preset decline mode of the learning rate.
Specifically, a warmup strategy is adopted for training the learning rate. The training is carried out at a small learning rate when the training is started so that the network model is familiar with data, the learning rate gradually increases along with the training, the training is carried out at a set initial learning rate to a certain extent, and then the learning rate gradually decreases. In addition, a preset learning rate reduction mode is selected for training, illustratively, cosine attenuation (cosine _ decay) can be used, the learning rate reduction mode does not need to adjust hyper-parameters, the robustness is high, and the convergence effect is good.
Optionally, when the sample data set is used for model training, the batch size value of the model training and the learning rate are in a linear relationship.
Specifically, the setting of the Batch size value (Batch size) and the value of the learning rate are in a linear relationship, so that the convergence accuracy during model training is prevented from being affected.
Optionally, when the sample data set is used for model training, the model optimizer used is Adam.
Specifically, Adam can be used as a model optimizer in the model optimization process, so that the loss function is as small as possible, and the convergence speed is relatively faster. The Adam model optimizer has the advantages of high computational efficiency, low memory requirements, suitability for problems with large data and/or parameters, suitability for problems with very noisy and/or sparse gradients, and the hyper-parameters have an intuitive interpretation, usually requiring little adjustment.
Optionally, before performing model training using the sample data set, the initial image recognition model may be pre-trained, which may be: and pre-training the initial image recognition model based on the face data set, and updating the initial image recognition model based on the pre-trained initial image recognition model.
The face data set may be a data set used for pre-training a model, and may include face images of various angles, lighting conditions, and the like.
Specifically, the face data set is used for training the initial image recognition model for pre-training to obtain a pre-training model, and the initial image recognition model is updated based on the pre-trained initial image recognition model, so that the convergence rate of subsequent model training is increased, the model precision is improved, and the generalization effect of the model can be improved.
Optionally, when the model training is performed by using the sample data set, the label smoothing is applied to correct the loss function, so as to generate a better calibration network, thereby improving the generalization capability, and generating an accurate prediction for the invisible data.
On the basis of the above embodiments, the target image recognition model is constructed based on EfficientNet or determined based on weighted fusion of at least two models of ResNeSt269, EfficientNet-B7 and EfficientNet-B8.
Specifically, the target image recognition model may be constructed based on EfficientNet, or the models resenestt 269, EfficientNet-B7, and EfficientNet-B8 may be selected to perform training for multiple times, respectively, and finally, a preset number of models (e.g., 3 models, etc.) with the highest accuracy are selected, and model fusion is performed by a weighted average method to obtain the target image recognition model.
It should be noted that the preset number of models selected when the target image recognition model is determined by using the weighted fusion method may be the same type of model or different types of models, and the determination needs to be specifically determined according to the model accuracy.
It should be noted that ResNeSt269, EfficientNet-B7 and EfficientNet-B8 are only exemplary, and other types of deep learning models can be used for fusion according to the usage requirement.
On the basis of the foregoing embodiments, optionally, the target image recognition model may be used for recognition, and may be:
acquiring an image to be identified; wherein the image to be recognized comprises a face region;
and inputting the image to be recognized into the target image recognition model to obtain the recognition category corresponding to the image to be recognized.
The image to be recognized may be an image obtained by shooting a target object based on a shooting device, the shooting device may be a device such as a camera, a video camera, etc., the target object may be a person to be subjected to image recognition, and the image to be processed includes a face area, i.e., a face area of the target object.
Specifically, the image to be processed is acquired and input into the target image recognition model, and the target image recognition model processes the image to be processed to obtain a processing result, that is, a recognition category.
For example, if the application scene of the target image recognition model is genetic disease recognition, the target image recognition model may process the image to be processed to determine the genetic disease category corresponding to the image to be processed. Accordingly, super-resolution processing can be performed on each sample medical image, and label reconfirmation can also be performed. For example: the target category can be the turner syndrome, and the re-confirmation of the label can be performed aiming at the sample data set to be processed corresponding to the turner syndrome so as to improve the identification effect of the turner syndrome.
The technical scheme of the embodiment of the invention comprises the steps of performing super-resolution processing on a sample medical image with a resolution lower than a preset resolution by acquiring the sample medical image and a sample original label corresponding to the sample medical image, updating the sample medical image according to the sample medical image after the super-resolution processing to improve the quality of the sample medical image, determining at least two target classes, determining a sub-sample data set corresponding to each target class according to the sample medical image corresponding to each target class and the sample original label corresponding to each sample medical image, determining a sample data set to be processed in the sub-sample data set, updating the sample data set to be processed according to each sample data set to be processed through two-stage classification model training and testing to improve the accuracy of the sample original label, and based on the updated sub-sample data set, the sample data set is determined, the problem that the sample data cannot be used for model training due to poor quality of the sample medical image and inaccuracy of the sample original label is solved, and the effects of improving the image quality of the sample medical image and the accuracy of the sample original label are achieved, so that model training can be better performed in the subsequent process.
Example two
Fig. 2 is a schematic flow chart of a sample data processing method according to a second embodiment of the present invention, and this embodiment may refer to the technical solution of this embodiment specifically for a super-resolution processing method of a sample medical image and a method for updating a sample data set to be processed through a two-stage classification model based on the above embodiments. The explanations of the same or corresponding terms as those in the above embodiments are omitted.
As shown in fig. 2, the method of this embodiment specifically includes the following steps:
s201, obtaining a sample medical image and a sample original label corresponding to the sample medical image.
S202, determining the sample medical image with the resolution lower than the preset resolution as a sample to-be-processed image, processing the sample to-be-processed image through an image super-resolution generation countermeasure network to obtain a sample processed image corresponding to the sample to-be-processed image, and updating the sample medical image based on the sample processed image.
The sample to-be-processed image is a sample medical image with a resolution lower than a preset resolution, and the sample to-be-processed image is processed in a subsequent processing process. The image Super-Resolution generation countermeasure network may be an SRGAN (Super-Resolution generation adaptive Networks). Briefly, as shown in the schematic diagram of SRGAN in fig. 3, SRGAN is a distribution space that uses a low resolution picture, i.e. a sample to-be-processed image, as an input of noise and fits the probability distribution space of the noise to the distribution space of real data as much as possible through the transformation of a generator (generation model). The sample processed image may be a sample to-be-processed image after the super-resolution processing.
Specifically, a sample to-be-processed image with a resolution lower than a preset resolution is determined from the sample medical image, the sample to-be-processed image is processed through an image super-resolution generation countermeasure network, the processed high-resolution image is used as a sample processed image, and the sample processed image is used for replacing the sample to-be-processed image, so that the sample medical image is updated.
S203, determining at least two target categories, and determining a sub-sample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image.
And S204, determining a sample data set to be processed in the sub-sample data set.
S205, aiming at each sample data set to be processed, dividing the sample data set to be processed into a first training set and a first testing set.
The first training set is a part of sample medical images in the sample data set to be processed. The first test set is sample medical images in the sample data set to be processed except the first training set.
Specifically, for each sample data set to be processed, half of the sample medical images may be determined as sample images in the first training set, and the remaining half of the sample medical images may be determined as sample images in the first test set.
It should be noted that the sample data set to be processed may be divided according to a ratio of 1:1, or may be divided according to other ratios, which is not specifically limited in this embodiment.
S206, training the initial classification model based on the first training set to obtain a first classification model, and training the initial classification model based on the first testing set to obtain a second classification model.
The initial classification model may be a binary classification model, and is used to determine whether the sample medical image belongs to a target class corresponding to the sample data set to be processed.
Specifically, the first training set is used as training data of the initial classification model, and the model obtained through training is used as the first classification model. And taking the first test set as training data of the initial classification model, and taking the model obtained by training as a second classification model.
S207, determining a second training set and a second testing set based on the first training set, the first testing set, the first classification model and the second classification model.
And the second training set and the second testing set are training sets and testing sets for retraining the initial classification model subsequently.
Specifically, the first test set is input into the first classification model to obtain a first classification result, and the first training set is input into the second classification model to obtain a second classification result. And integrating the sample medical images of which the first classification results are the target classes and the sample medical images of which the second classification results are the target classes into a second training set, and determining the sample medical images in the sample data set to be processed except the second training set as a second test set.
Optionally, the second training set and the second test set may be determined by:
step one, inputting a first test set into a first classification model, and determining that a sample medical image of a target class corresponding to a sample set to be processed, which is an output result in the first test set, is a first part in a second training set.
And step two, inputting the first training set into a second classification model, and determining the sample medical image of which the output result in the first training set is the target class as a second part in the second training set.
And step three, determining a second training set according to the first part and the second part, and determining a second test set according to the parts of the sample data set to be processed except the second training set.
Illustratively, the sample data set to be processed contains 200 sample medical images, 100 of the sample data sets are used as a first training set to train an initial classification model to a first classification model, and the remaining 100 sample data sets are used as a first testing set to train the initial classification model to a second classification model. And testing the first test set by using the first classification model to obtain 90 sample medical images with output results of the target category, namely the first part in the second training set, and obtain 10 sample medical images with output results of the non-target category, namely the first part in the second test set. And testing the first training set by using a second classification model to obtain 85 sample medical images with output results of target categories, namely a second part in the second training set, and obtain 15 sample medical images with output results of non-target categories and a second part in the second testing set. Thus, it may be determined that there are 90+ 85-175 sample medical images in the second training set, 10+ 15-25 sample medical images in the second testing set, or 200-175-25 sample medical images in the second testing set.
S208, training the initial classification model based on the second training set to obtain a third classification model, inputting the second test set into the third classification model, and determining an output result in the second test set.
Specifically, the second training set is used as training data of the initial classification model, and the model obtained through training is used as a third classification model. And inputting the second test set into the third classification model to obtain an output result in the second test set.
Based on the above example, the sample medical images in the second training set are 175 in total and the sample medical images in the second testing set are 25 in total. And training the initial classification model to a third classification model by using the second training set, and testing the second test set by using the third classification model to obtain 20 sample medical images of which the output results in the second test set are in the target category and 5 sample medical images of which the output results in the second test set are in the non-target category.
And S209, updating the sample data set to be processed according to the output results in the second training set and the second testing set.
Specifically, the original sample image of the target category, which is not the output result in the second test set and corresponds to the sample data set to be processed, may be removed from the sample data set to be processed, so as to update the sample data set to be processed.
Optionally, the sample data set to be processed may be updated through the following steps:
step one, updating the sample medical images in the sample data set to be processed based on the sample medical images in the second training set and the sample medical images with the output results in the second testing set as the target categories.
Specifically, the sample medical images in the second training set are used as the first part of the sample medical images in the sample data set to be processed. And taking the sample medical image of the target class corresponding to the sample data set to be processed as the output result in the second test set as the second part of the sample medical image in the sample data set to be processed. And updating the sample medical image in the sample data set to be processed based on the first part and the second part.
And secondly, updating the sample data set to be processed based on the updated sample medical image and the target category.
Specifically, the original sample category of each sample medical image in the updated sample data set to be processed is determined according to the target category, and then the sample data set to be processed is updated.
Based on the above example, for example, 175 sample medical images in the second training set and 20 sample medical images with the output result in the second testing set as the target category may be combined to obtain 195 sample medical images, that is, the updated sample medical images in the sample data set to be processed, and the sample medical images may be given sample original labels corresponding to the target category. Or 5 sample medical images with the output result of the second test set as a non-target category are removed from 200 sample medical images corresponding to the sample data to be processed, the residual 195 sample medical images are used as the sample medical images in the updated sample data set to be processed, and sample original labels corresponding to the target category are given to the sample medical images.
And S210, determining a sample data set based on the updated sub-sample data set.
The technical scheme of the embodiment of the invention comprises the steps of determining a sample medical image with a resolution lower than a preset resolution as a sample to-be-processed image by acquiring the sample medical image and a sample original label corresponding to the sample medical image, processing the sample to-be-processed image through an image super-resolution generation countermeasure network to obtain a sample processed image corresponding to the sample to-be-processed image, updating the sample medical image based on the sample processed image to improve the quality of the sample medical image, determining at least two target categories, determining a subsample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image, determining a sample data set to be processed in the subsample data set, dividing the sample data set to be processed into a first training set and a first testing set aiming at each sample data set to be processed, training an initial classification model based on a first training set to obtain a first classification model, training the initial classification model based on a first test set to obtain a second classification model, determining a second training set and a second test set based on the first training set, the first test set, the first classification model and the second classification model, training the initial classification model based on the second training set to obtain a third classification model, inputting the second test set into the third classification model, determining an output result in the second test set, updating a sample data set to be processed according to the output results in the second training set and the second test set to improve the accuracy of an original label of the sample, determining the sample data set based on the updated sub-sample data set, and solving the problem that the sample data set cannot be used for model training due to poor medical image quality of the sample and inaccurate original label of the sample, the method and the device have the advantages that the effects of improving the image quality of the sample medical image and the accuracy of the original label of the sample are achieved, and therefore model training can be better carried out subsequently.
EXAMPLE III
Based on the above embodiments, taking genetic disease facial recognition as an example, fig. 4 is a schematic flow chart of a genetic disease facial recognition method provided by a third embodiment of the present invention.
As shown in fig. 4, the genetically diseased facial method of the present embodiment includes:
(1) image pre-processing
The influence of the training data (a training sample set) on the model is decisive, the collected data (a sample original image) is cleaned, the data with smaller size and fuzzy image are removed, then face cropping is carried out, the interference of redundant information is eliminated, and meanwhile, the size of the image after cropping is consistent with the size required by the model (such as EfitientNet).
(2) Image super-resolution
The low-Resolution image is reconstructed into a corresponding high-Resolution image through SRGAN (Super-Resolution general adaptive network), a Generated Adaptive Network (GAN) for Super-Resolution (SR) of the image.
(3) Validating tags by model
Due to the fact that partial image labels (sample original labels) are inaccurate in the collection process of images (sample medical images) corresponding to certain genetic diseases, model accuracy is low. Thus, the tag needs to be validated. The label can be confirmed in the same way for each genetic disease, taking a turner syndrome as an example, firstly, taking half of image data as a training set (a first training set), taking the other half of the image data as a testing set (a first testing set), performing first model training, then, reversing the original training set and the original testing set, performing second model training, taking the turner syndrome tested by two models as a training data set (a second training set), taking the error in testing as a testing data set (a second testing set), performing third model training, and adding the turner syndrome obtained by third model testing to the training data set of the third model training, namely, the data set with the correct label theoretically.
(4) Image enhancement
Because the difference of the number of various samples in the training sample set is large, only dozens of samples are used, and more samples are used, so that the accuracy of partial classes is low. Therefore, more than ten image processing methods of rotation, mirroring, sharpening, translation, color enhancement, brightness, saturation, and the like are used to solve the data imbalance problem. In principle, the image processing methods are used for a large amount of data, and the image processing methods are used for a large amount of data, so that the number of each genetic disease reaches about 2500. The image enhancement method enhances the training data volume, reduces the degree of network overfitting and enhances the generalization capability of the model.
(5) Model selection
The convolutional neural network EfficientNet, namely a simple and efficient composite coefficient, is used for amplifying the network from three dimensions of network depth, network width and image resolution, so that the number of layers of the network is increased, more information is acquired, and more features are extracted. The convolution neural network EfficientNet does not scale the dimensionality of the network at will like a traditional deep learning method, but can obtain an optimal set of parameters, namely a composite coefficient, based on a neural structure search technology.
(6) Root of Yunnan Manchurian wildginger
And a warp strategy (a training warm-up method) is adopted, the selected learning rate reduction mode is cosine _ decay, the hyper-parameters do not need to be adjusted, the robustness is high, and the convergence effect is good. The setting of Batch size and the value of learning rate are in a linear relationship, avoiding the convergence accuracy from being affected. And (3) selecting an optimizer Adam with a self-adaptive learning rate to make a loss function as small as possible and make the convergence speed relatively faster so as to find out appropriate parameters to complete the classification recognition task.
(7) Training model
Before training a multi-classification model (a target genetic disease identification model), an EfficientNet pre-training model (an initial genetic disease identification model after pre-training) is trained by using a mass human face data set (a human face data set), the convergence speed during training the multi-classification model is increased later, the precision of the model is improved, and a better generalization effect is obtained. During the training process, label smoothing is applied to correct the loss function, so that a better calibration network is generated, and more accurate prediction is generated on invisible data through better generalization. Further, the trained multi-classification model is determined as a prediction model (target genetic disease recognition model).
Optionally, the ResNeSt269, the efficientNet-B7 and the efficientNet-B8 with good classification and recognition effects can be selected for training for multiple times, and finally, three (preset number) models with the highest accuracy are selected for model fusion by a weighted average method to obtain the prediction model.
In the case of turner syndrome as an example, the recognition accuracy of turner syndrome was 53.2% when the image super-resolution processing and the tag confirmation were not performed, and the recognition accuracy of turner syndrome was 88.6% when the image super-resolution processing and the tag confirmation were performed.
According to the technical scheme of the embodiment of the invention, through preprocessing the data corresponding to the target category, image super-resolution and image label confirmation, and continuous training of the model, the parameters are subjected to targeted adjustment, label smoothing is reasonably applied, the model is pre-trained, and the model is selected and fused, so that the accuracy of the target category identification and the accuracy of the multi-classification model are improved, the problems of over-fitting and under-fitting in the model training process are avoided, and the generalization capability of the model is enhanced.
Example four
Fig. 5 is a schematic structural diagram of a sample data processing apparatus according to a fourth embodiment of the present invention, where the apparatus includes: the system comprises a sample original data acquisition module 310, a sample medical image updating module 320, a subsample data set splitting module 330, a to-be-processed sample data set updating module 340 and a sample data set determining module 350.
The system comprises a sample original data acquisition module 310, a sample medical image acquisition module, a sample original label acquisition module and a sample original label acquisition module, wherein the sample original data acquisition module is used for acquiring a sample medical image and a sample original label corresponding to the sample medical image; a sample medical image updating module 320, configured to perform super-resolution processing on the sample medical image with a resolution lower than a preset resolution, and update the sample medical image according to the super-resolution processed sample medical image; a sub-sample data set splitting module 330, configured to determine at least two target categories, and determine a sub-sample data set corresponding to each target category according to a sample medical image corresponding to each target category and a sample original label corresponding to each sample medical image; a to-be-processed sample data set updating module 340, configured to determine a to-be-processed sample data set in the sub-sample data set, and update the to-be-processed sample data set through two-stage classification model training and testing for each to-be-processed sample data set; a sample data set determining module 350, configured to determine a sample data set based on the updated sub-sample data set.
Optionally, the sample medical image updating module 320 is specifically configured to determine a sample medical image with a resolution lower than a preset resolution as a sample image to be processed, and process the sample image to be processed through an image super-resolution generation countermeasure network to obtain a sample processed image corresponding to the sample image to be processed; updating the sample medical image based on the sample processed image.
Optionally, the to-be-processed sample data set updating module 340 is specifically configured to, for each to-be-processed sample data set, divide the to-be-processed sample data set into a first training set and a first test set; training an initial classification model based on the first training set to obtain a first classification model, and training the initial classification model based on the first test set to obtain a second classification model; determining a second training set and a second test set based on the first training set, the first test set, the first classification model, and the second classification model; training the initial classification model based on the second training set to obtain a third classification model, inputting the second test set into the third classification model, and determining an output result in the second test set; and updating the sample data set to be processed according to the output results in the second training set and the second testing set.
Optionally, the to-be-processed sample data set updating module 340 is specifically configured to input the first test set into the first classification model, and determine that a sample medical image of a target class corresponding to the to-be-processed sample data set, which is an output result in the first test set, is a first part in a second training set; inputting the first training set into the second classification model, and determining that the sample medical images with the output results of the target category in the first training set are the second part in the second training set; and determining a second training set according to the first part and the second part, and determining a second test set according to the parts of the sample data set to be processed except the second training set.
Optionally, the to-be-processed sample data set updating module 340 is specifically configured to update the sample medical images in the to-be-processed sample data set based on the sample medical images in the second training set and the sample medical images in the target category of the output result in the second test set; and updating the sample data set to be processed based on the updated sample medical image and the target category.
Optionally, the apparatus further comprises: and the model training module is used for training an initial image recognition model according to the sample medical images in the sample data set and the sample original labels corresponding to the sample medical images, and taking the trained initial image recognition model as a target image recognition model, wherein the initial image recognition model is constructed based on a model scaling method which gives consideration to both speed and precision.
Optionally, the apparatus further comprises: the image recognition module is used for acquiring an image to be recognized; wherein the image to be recognized includes a face region; and inputting the image to be recognized into the target image recognition model to obtain a recognition category corresponding to the image to be recognized.
The technical scheme of the embodiment of the invention comprises the steps of performing super-resolution processing on a sample medical image with a resolution lower than a preset resolution by acquiring the sample medical image and a sample original label corresponding to the sample medical image, updating the sample medical image according to the sample medical image after the super-resolution processing to improve the quality of the sample medical image, determining at least two target classes, determining a sub-sample data set corresponding to each target class according to the sample medical image corresponding to each target class and the sample original label corresponding to each sample medical image, determining a sample data set to be processed in the sub-sample data set, updating the sample data set to be processed according to each sample data set to be processed through two-stage classification model training and testing to improve the accuracy of the sample original label, and based on the updated sub-sample data set, the sample data set is determined, the problem that the sample data cannot be used for model training due to poor quality of the sample medical image and inaccuracy of the sample original label is solved, and the effects of improving the image quality of the sample medical image and the accuracy of the sample original label are achieved, so that model training can be better performed in the subsequent process.
The sample data processing device provided by the embodiment of the invention can execute the sample data processing method provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.
It should be noted that, the units and modules included in the apparatus are merely divided according to functional logic, but are not limited to the above division as long as the corresponding functions can be implemented; in addition, specific names of the functional units are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the embodiment of the invention.
EXAMPLE five
Fig. 6 is a schematic structural diagram of an electronic device according to a fifth embodiment of the present invention. FIG. 6 illustrates a block diagram of an exemplary electronic device 40 suitable for use in implementing embodiments of the present invention. The electronic device 40 shown in fig. 6 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiment of the present invention.
As shown in fig. 6, electronic device 40 is embodied in the form of a general purpose computing device. The components of electronic device 40 may include, but are not limited to: one or more processors or processing units 401, a system memory 402, and a bus 403 that couples the various system components (including the system memory 402 and the processing unit 401).
The system memory 402 may include computer system readable media in the form of volatile memory, such as Random Access Memory (RAM)404 and/or cache 405. The electronic device 40 may further include other removable/non-removable, volatile/nonvolatile computer system storage media. By way of example only, storage system 406 may be used to read from and write to non-removable, nonvolatile magnetic media (not shown in FIG. 6, commonly referred to as a "hard drive"). Although not shown in FIG. 6, a magnetic disk drive for reading from and writing to a removable, nonvolatile magnetic disk (e.g., a "floppy disk") and an optical disk drive for reading from or writing to a removable, nonvolatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 403 by one or more data media interfaces. System memory 402 may include at least one program product having a set (e.g., at least one) of program modules that are configured to carry out the functions of embodiments of the invention.
A program/utility 408 having a set (at least one) of program modules 407 may be stored, for example, in system memory 402, such program modules 407 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which examples or some combination thereof may comprise an implementation of a network environment. Program modules 407 generally perform the functions and/or methodologies of embodiments of the invention as described.
The electronic device 40 may also communicate with one or more external devices 409 (e.g., keyboard, pointing device, display 410, etc.), with one or more devices that enable a user to interact with the electronic device 40, and/or with any devices (e.g., network card, modem, etc.) that enable the electronic device 40 to communicate with one or more other computing devices. Such communication may be performed through an I/O interface (input/output interface) 411. Also, the electronic device 40 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the Internet) via the network adapter 412. As shown, the network adapter 412 communicates with the other modules of the electronic device 40 over the bus 403. It should be understood that although not shown in FIG. 6, other hardware and/or software modules may be used in conjunction with electronic device 40, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.
The processing unit 401 executes various functional applications and data processing by running programs stored in the system memory 402, for example, to implement the sample data processing method provided by the embodiment of the present invention.
Example six
An embodiment of the present invention further provides a storage medium containing computer-executable instructions, which when executed by a computer processor, are configured to perform a sample data processing method, including:
acquiring a sample medical image and a sample original label corresponding to the sample medical image;
performing super-resolution processing on the sample medical image with the resolution lower than the preset resolution, and updating the sample medical image according to the sample medical image after the super-resolution processing;
determining at least two target categories, and determining a sub-sample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image;
determining a sample data set to be processed in the sub-sample data set, and updating the sample data set to be processed through two-stage classification model training and testing aiming at each sample data set to be processed;
based on the updated subsample dataset, a sample dataset is determined.
Computer storage media for embodiments of the invention may employ any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Computer program code for carrying out operations for embodiments of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).
It is to be noted that the foregoing description is only exemplary of the invention and that the principles of the technology may be employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims (10)
1. A sample data processing method is characterized by comprising the following steps:
acquiring a sample medical image and a sample original label corresponding to the sample medical image;
performing super-resolution processing on the sample medical image with the resolution lower than a preset resolution, and updating the sample medical image according to the sample medical image after the super-resolution processing;
determining at least two target categories, and determining a sub-sample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image;
determining a sample data set to be processed in the sub-sample data set, and updating the sample data set to be processed through two-stage classification model training and testing aiming at each sample data set to be processed;
based on the updated subsample dataset, a sample dataset is determined.
2. The method according to claim 1, wherein the super-resolution processing is performed on the sample medical image with a resolution lower than a preset resolution, and the sample medical image is updated according to the super-resolution processed sample medical image, and the method comprises:
determining a sample medical image with a resolution lower than a preset resolution as a sample to-be-processed image, and processing the sample to-be-processed image through an image super-resolution generation countermeasure network to obtain a sample processed image corresponding to the sample to-be-processed image;
updating the sample medical image based on the sample processed image.
3. The method according to claim 1, wherein updating the sample data set to be processed by two-stage classification model training and testing for each sample data set to be processed comprises:
for each sample data set to be processed, dividing the sample data set to be processed into a first training set and a first testing set;
training an initial classification model based on the first training set to obtain a first classification model, and training the initial classification model based on the first test set to obtain a second classification model;
determining a second training set and a second test set based on the first training set, the first test set, the first classification model, and the second classification model;
training the initial classification model based on the second training set to obtain a third classification model, inputting the second test set into the third classification model, and determining an output result in the second test set;
and updating the sample data set to be processed according to the output results in the second training set and the second testing set.
4. The method of claim 3, wherein determining a second training set and a second test set based on the first training set, the first test set, the first classification model, and the second classification model comprises:
inputting the first test set into the first classification model, and determining that a sample medical image of a target class corresponding to the sample data set to be processed as an output result in the first test set is a first part in a second training set;
inputting the first training set into the second classification model, and determining that the sample medical image of which the output result is the target class in the first training set is a second part in the second training set;
and determining a second training set according to the first part and the second part, and determining a second test set according to the parts of the sample data set to be processed except the second training set.
5. The method of claim 3, wherein updating the set of sample data to be processed according to the output results in the second training set and the second test set comprises:
updating the sample medical images in the sample data set to be processed based on the sample medical images in the second training set and the output result in the second testing set as the sample medical images in the target category;
and updating the sample data set to be processed based on the updated sample medical image and the target class.
6. The method of claim 1, further comprising:
training an initial image recognition model according to the sample medical images in the sample data set and sample original labels corresponding to the sample medical images, and taking the trained initial image recognition model as a target image recognition model, wherein the initial image recognition model is constructed based on a model scaling method which gives consideration to both speed and precision.
7. The method of claim 6, further comprising:
acquiring an image to be identified; wherein the image to be recognized includes a face region;
and inputting the image to be recognized into the target image recognition model to obtain a recognition category corresponding to the image to be recognized.
8. A sample data processing apparatus, comprising:
the system comprises a sample original data acquisition module, a sample original data acquisition module and a sample original label, wherein the sample original data acquisition module is used for acquiring a sample medical image and a sample original label corresponding to the sample medical image;
the sample medical image updating module is used for performing super-resolution processing on the sample medical image with the resolution lower than the preset resolution and updating the sample medical image according to the sample medical image after the super-resolution processing;
the sub-sample data set splitting module is used for determining at least two target categories and determining a sub-sample data set corresponding to each target category according to the sample medical image corresponding to each target category and the sample original label corresponding to each sample medical image;
the to-be-processed sample data set updating module is used for determining the to-be-processed sample data set in the sub-sample data set, and updating the to-be-processed sample data set by means of two-stage classification model training and testing aiming at each to-be-processed sample data set;
and the sample data set determining module is used for determining the sample data set based on the updated sub-sample data set.
9. An electronic device, characterized in that the electronic device comprises:
one or more processors;
a storage device for storing one or more programs,
when executed by said one or more processors, cause said one or more processors to implement a sample data processing method as claimed in any one of claims 1 to 7.
10. A computer readable storage medium having stored thereon a computer program, characterized in that the program, when executed by a processor, implements the sample data processing method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210637331.0A CN115049546A (en) | 2022-06-07 | 2022-06-07 | Sample data processing method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210637331.0A CN115049546A (en) | 2022-06-07 | 2022-06-07 | Sample data processing method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115049546A true CN115049546A (en) | 2022-09-13 |
Family
ID=83160598
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210637331.0A Pending CN115049546A (en) | 2022-06-07 | 2022-06-07 | Sample data processing method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115049546A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117115568A (en) * | 2023-10-24 | 2023-11-24 | 浙江啄云智能科技有限公司 | Data screening method, device, equipment and storage medium |
-
2022
- 2022-06-07 CN CN202210637331.0A patent/CN115049546A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117115568A (en) * | 2023-10-24 | 2023-11-24 | 浙江啄云智能科技有限公司 | Data screening method, device, equipment and storage medium |
| CN117115568B (en) * | 2023-10-24 | 2024-01-16 | 浙江啄云智能科技有限公司 | Data screening method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| RU2699687C1 (en) | Detecting text fields using neural networks | |
| RU2695489C1 (en) | Identification of fields on an image using artificial intelligence | |
| CN113313022B (en) | Training method of character recognition model and method for recognizing characters in image | |
| CN111738169B (en) | Handwriting formula recognition method based on end-to-end network model | |
| US9208374B2 (en) | Information processing apparatus, control method therefor, and electronic device | |
| CN113254654B (en) | Model training method, text recognition method, device, equipment and medium | |
| CN115861462B (en) | Training method and device for image generation model, electronic equipment and storage medium | |
| CN111046771A (en) | Training method of network model for recovering writing track | |
| CN113780326A (en) | Image processing method and device, storage medium and electronic equipment | |
| CN110796108B (en) | Method, device and equipment for detecting face quality and storage medium | |
| CN115601820A (en) | Face fake image detection method, device, terminal and storage medium | |
| CN115049546A (en) | Sample data processing method and device, electronic equipment and storage medium | |
| CN114299366A (en) | Image detection method and device, electronic equipment and storage medium | |
| EP4060526A1 (en) | Text processing method and device | |
| CN117437423A (en) | Weak supervision medical image segmentation method and device based on SAM collaborative learning and cross-layer feature aggregation enhancement | |
| CN112307769B (en) | Natural language model generation method and computer equipment | |
| CN115565186B (en) | Training method and device for character recognition model, electronic equipment and storage medium | |
| CN111753736A (en) | Human body posture recognition method, device, equipment and medium based on packet convolution | |
| CN114241411B (en) | Counting model processing method and device based on target detection and computer equipment | |
| CN116047418A (en) | Multi-mode radar active deception jamming identification method based on small sample | |
| CN112801960B (en) | Image processing method and device, storage medium and electronic equipment | |
| CN116543246A (en) | Training method of image denoising model, image denoising method, device and equipment | |
| Zhang et al. | Style classification of media painting images by integrating ResNet and attention mechanism | |
| CN117876751A (en) | Image processing method, image processing system, and computer readable medium | |
| CN116166920A (en) | Business object analysis method, device, equipment and storage medium |
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 |