CN113505784A

CN113505784A - Automatic nail annotation analysis method and device, electronic equipment and storage medium

Info

Publication number: CN113505784A
Application number: CN202110655958.4A
Authority: CN
Inventors: 刘知远; 孙茂松; 邱可玥; 韩旭; 李永威; 肖光烜; 吕天
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2021-06-11
Filing date: 2021-06-11
Publication date: 2021-10-15
Anticipated expiration: 2041-06-11
Also published as: CN113505784B

Abstract

The invention provides a method and a device for automatically marking and analyzing a nail plate, wherein the method comprises the following steps: obtaining an original A picture and removing background noise of the original A picture to obtain a preprocessed A picture; inputting the first image to a target detection model to obtain a target detection result output by the target detection model; the target detection result comprises an oracle area and a corresponding category, and the target detection model is obtained based on forged oracle samples. According to the invention, the interference of background noise on the detection of the oracle target is reduced by introducing the preprocessing of the oracle image, and the forged oracle is used for training the target detection model, so that the application effect of the target detection model in a real scene is ensured, and the accurate identification of the oracle in a complex oracle is realized.

Description

Automatic nail annotation analysis method and device, electronic equipment and storage medium

Technical Field

The invention relates to the technical field of nail annotation analysis, in particular to a nail automatic annotation analysis method and device, electronic equipment and a storage medium.

Background

Since in the actual situation the oracle is basically located on the unearthed nail and often there are many oracle on one nail, the model of single-word recognition is difficult to be used directly in real-world applications. The model with more practical application significance should have the capability of directly finding single carapace characters from complex whole carapaces and identifying the single carapaces, namely the capability of automatically labeling and analyzing the carapaces.

Disclosure of Invention

The invention provides a method and a device for automatically marking and analyzing a nail, which are used for solving the defect that a single character recognition model in the prior art is difficult to be used for a complex nail containing a plurality of oracle characters, and realizing the automatic frame selection and recognition of the oracle characters in the complex nail.

The invention provides an automatic nail annotation analysis method, which comprises the following steps:

obtaining an original A picture and removing background noise of the original A picture to obtain a preprocessed A picture;

inputting the first image to a target detection model to obtain a target detection result output by the target detection model;

the target detection result comprises an oracle area and a corresponding category, and the target detection model is obtained based on forged oracle samples.

According to the automatic nail annotation analysis method provided by the invention, the step of generating the forged nail sample comprises the following steps:

simulating the arrangement mode of the individual oracle characters in the individual oracle character data set to randomly place the individual oracle characters based on a preset individual oracle character data set and the arrangement mode of the oracle characters on the oracle, and adding random noise to generate a forged oracle data set;

and sampling the forged nail plate data set to obtain a forged nail plate sample.

According to the automatic nail annotation analysis method provided by the invention, the removing of the background noise of the original nail image comprises the following steps:

and removing the background noise of the original nail image based on image binarization and image painting.

According to the automatic nail annotation analysis method provided by the invention, the step of inputting the nail image into a target detection model to obtain a target detection result output by the target detection model comprises the following steps:

selecting a candidate region from the first image to obtain a candidate region set;

extracting the features of each candidate region based on a convolutional neural network to obtain corresponding feature vector representation;

inputting the feature vector representation into a classifier to obtain a corresponding category and a confidence level;

sequentially calculating the overlapping degree between every two candidate regions;

if the overlapping degree is larger than a preset threshold value, deleting the candidate region corresponding to the characteristic vector representation with low confidence coefficient from the candidate region set to obtain a candidate region subset;

adjusting the frame of the candidate area in the candidate area subset to obtain a corrected oracle area;

and taking the oracle region and the corresponding category as a target detection result.

According to the method for automatically labeling and analyzing the nail plate, provided by the invention, the forged nail plate data set is sampled to obtain a forged nail plate sample, and the method comprises the following steps:

dividing the sampling interval of the forged A-plate data set to obtain a plurality of sampling subintervals;

and selecting the corresponding to-be-selected positive sample or to-be-selected negative sample from the sampling subinterval as a forged nail sample.

According to the automatic nail annotation analysis method provided by the invention, the convolutional neural network-based feature extraction is performed on each candidate region to obtain a corresponding feature vector representation, and the method comprises the following steps:

extracting features of each candidate region based on the convolutional neural network to obtain original level features corresponding to different neural network layers;

fusing the original level features to obtain fused semantic features;

processing the original level features based on the fused semantic features to obtain corresponding enhanced semantic features;

and taking the enhanced semantic features of the output layer of the convolutional neural network as feature vectors corresponding to the candidate regions for representation.

According to the automatic nail annotation analysis method provided by the invention, the target detection model is trained by adopting the following loss function:

wherein x is the sample loss, α, γ, b and C are parameters, α, γ and b satisfy the following constraints:

αln(b+1)＝γ。

the invention also provides an automatic nail plate labeling and analyzing device, which comprises:

the device comprises a preprocessing unit, a background noise generation unit and a background noise generation unit, wherein the preprocessing unit is used for acquiring an original A picture and removing the background noise of the original A picture to obtain a preprocessed A picture;

the target detection unit is used for inputting the first image to a target detection model to obtain a target detection result output by the target detection model;

According to the automatic nail annotation analysis device provided by the invention, the device further comprises a data sample generation unit, which is used for:

According to the automatic nail plate labeling and analyzing device provided by the invention, the preprocessing unit is further used for:

According to the automatic nail annotation analysis device provided by the invention, the target detection unit is further configured to:

According to the automatic nail annotation analysis device provided by the invention, the data sample generation unit is further configured to:

fusing the original level features to obtain fused semantic features;

According to the automatic nail annotation analysis device provided by the invention, the device further comprises a model training unit, which is used for:

training the target detection model using the following loss function:

αln(b+1)＝γ。

the invention also provides an electronic device, which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein when the processor executes the program, the steps of the automatic nail annotation analysis method can be realized.

The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method for automatic annotation resolution of a nail file as described in any one of the above.

According to the method and the device for automatically marking and analyzing the oracle, the interference of background noise on the detection of the oracle target is reduced by introducing the preprocessing of the oracle image, and the forged oracle is used for training the target detection model, so that the application effect of the target detection model in a real scene is ensured, and the accurate identification of the oracle in a complex oracle is realized.

Drawings

In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.

Fig. 1 is a schematic flow chart of a method for automatically labeling and analyzing a nail plate according to an embodiment of the present invention;

FIG. 2 is a second schematic flow chart of the method for automatically labeling and analyzing a nail plate according to the embodiment of the present invention;

FIG. 3 is a schematic diagram of an automatically generated training sample provided by an embodiment of the present invention;

FIG. 4 is a schematic illustration of images of a nail before and after pre-processing provided by an embodiment of the invention;

fig. 5 is a third schematic flow chart of an automatic nail annotation analysis method according to an embodiment of the present invention;

FIG. 6 is a fourth schematic flowchart of an automatic nail annotation analyzing method according to an embodiment of the present invention;

fig. 7 is a fifth flowchart illustrating an automatic nail annotation analyzing method according to an embodiment of the present invention;

FIG. 8 is a schematic diagram illustrating the detection effect of oracle-bone inscription according to an embodiment of the present invention;

fig. 9 is a schematic structural diagram of an automatic nail annotation analyzing device according to an embodiment of the present invention;

fig. 10 is a schematic structural diagram of an electronic device provided by the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The technical solutions provided by the embodiments of the present invention are described below with reference to fig. 1 to 10.

Fig. 1 is a schematic flow chart of a method for automatically labeling and analyzing a nail plate according to an embodiment of the present invention, where the method includes:

and step 110, obtaining an original image of the nail plate and removing background noise of the original image of the nail plate to obtain a preprocessed image of the nail plate.

One of the difficulties of the automatic nail annotation and analysis task is that the noise greatly reduces the accuracy of target detection in the case of the non-uniform background color, nail fouling and excessive cracks in the whole nail image, and therefore the problem of the noise of the original nail image needs to be solved in nail data preprocessing.

And 120, inputting the image of the nail into the target detection model to obtain a target detection result output by the target detection model.

It should be noted that the automatic nail annotation analysis is to find and identify individual nails from complex whole nails, and such a task is very similar to a target detection task in the field of computer vision, and the target detection task is to select and classify a target object frame from a picture with a plurality of objects, so that a model with the capability of automatically selecting and identifying the oracle characters by using a relevant method of target detection can be constructed.

However, since the well-labeled data is very rare, the training data is deficient, and the target detection model cannot be directly applied to the automatic nail labeling analysis task, the embodiment provides a 'fake nail' training method, a real nail is simulated by fake nail, a large number of training samples with fine labels can be conveniently generated, and the difference between the training samples and the real scene is not large, so that the application effect of the model trained by the method in the real scene can be ensured.

According to the automatic nail annotation analysis method provided by the invention, the interference of background noise on the detection of the oracle target is reduced by introducing the preprocessing of the nail image, and the forged nail is used for training the target detection model, so that the application effect of the target detection model in a real scene is ensured, and the accurate identification of the oracle in a complex nail is realized.

Further, in an embodiment of the present invention, a counterfeit nail sample is generated by:

step 210, randomly placing individual oracle characters in the individual oracle character data set in a simulated arrangement mode based on a preset individual oracle character data set and an arrangement mode of the oracle characters on the oracle, and adding random noise to generate a forged oracle data set.

Specifically, in order to simulate a preprocessed nail image, samples of a plurality of individual oracle characters are randomly selected from an individual oracle character data set and placed on a black background picture, and the placed positions are recorded as correct marks of object selection frames, so that the distribution of the oracle characters on the nail is simulated; meanwhile, in order to reduce overfitting and enhance the robustness of the system, random white noise points are added into the picture, noise points which are difficult to remove in the picture scanned by the A-slice are simulated, and the model is adapted to and learns to ignore the noise. Fig. 3 shows the training sample automatically generated by the method, and it can be seen that the difference between the training sample and the real scene is not large, so that the application effect of the model trained by the method in the real scene can be ensured. By using the method, a large number of fake nails which are equivalent to those with fine labels can be easily generated, so that a large enough data set is provided for training the target detection model.

And step 220, sampling the forged nail plate data set to obtain a forged nail plate sample.

Further, in an embodiment of the present invention, the background noise of the original nail image is removed based on image binarization and image painting.

Specifically, firstly, the nail image is subjected to binarization processing, namely each pixel is changed into pure black or pure white according to a set threshold value, so that strokes of the oracle characters can be highlighted, the light and shade problems caused by scanning are removed, and meanwhile, the next step of processing is facilitated. After the binarization processing, the background except the nail plate needs to be blackened together, that is, the edge information of the nail plate needs to be removed, so as to reduce the influence caused by unnecessary boundaries and contours. Therefore, in this embodiment, the Floodfill method is used to blacken the outer ring background, so as to obtain a picture with black background and white characters. Thus, a large amount of unnecessary noise generated by the nail contour, scanning and the like is removed, and the performance of the target detection model can be greatly improved. Fig. 4 shows the nail image before and after the nail image preprocessing is performed by the method, it can be seen that after the data preprocessing, the background noise of the nail image is removed, meanwhile, the overall image contrast is improved, and the processed image can be directly processed by using the target detection model.

Further, the method of the Region-CNN (R-CNN) series has good effect on a plurality of target detection tasks, so the invention also selects the R-CNN as a basic method for target detection of the nail image. In an embodiment of the present invention, step 120 specifically includes:

and step 510, selecting candidate areas of the nail image to obtain a candidate area set.

It should be noted that the candidate regions generated by the candidate region selection algorithm are different in size, and the shapes and sizes of the candidate regions are ignored for compatibility with the convolutional neural network, and are scaled to a uniform size for subsequent processing.

And 520, extracting the features of each candidate region based on the convolutional neural network to obtain corresponding feature vector representation.

Step 530, inputting the feature vector representation to a classifier to obtain a corresponding category and confidence.

And 540, sequentially calculating the overlapping degree between every two candidate areas.

And step 550, if the overlapping degree is greater than a preset threshold, deleting the candidate region corresponding to the feature vector representation with low confidence from the candidate region set to obtain a candidate region subset.

It should be noted that, as described in

steps

530, 540, and 550, the present embodiment adopts a non-maximum inhibition method to select the final oracle character selection box. Specifically, for two different nail image candidate regions, IoU indices of the two are calculated. Wherein IoU of two different candidate regions D1, D2 is defined as:

at the same time, the target detection model learns a threshold over which the regions with lower confidence are rejected if IoU for two different candidate regions exceed, i.e., is a non-maximum suppression method. By the method, a large number of candidate areas with high overlapping rates can be screened out by the target detection model, and then the best selection frame containing the single oracle characters is found.

And step 560, adjusting the frame of the candidate area in the candidate area subset to obtain the corrected oracle area.

And step 570, taking the oracle region and the corresponding category as a target detection result.

It should be noted that Libra R-CNN, as one of the latest achievements of R-CNN series methods, can achieve the best effect on multiple tasks of target detection, and the balance learning scheme proposed by the invention has better adaptability to a task with higher requirements on frame selection positions, such as a nail automatic labeling analysis task, so the invention further selects the Libra R-CNN method as a target detection model in the nail automatic labeling analysis method.

Further, in an embodiment of the present invention, the method based on Libra R-CNN collects the counterfeit nail sample as follows:

and 610, dividing the sampling interval of the forged A-plate data set to obtain a plurality of sampling subintervals.

And step 620, selecting the corresponding to-be-selected positive sample or to-be-selected negative sample from the sampling subinterval as the forged nail sample.

It should be noted that in order to solve the problem of unbalanced sampling in the R-CNN method, Libra R-CNN proposes loU balanced sampling. Specifically, sampling imbalance means that when a fake nail sample is collected, fewer samples which are larger in appearance and difficult to represent for a target detection model are submerged in a large number of simple samples, so that gradient updating caused by the difficult samples is more difficult to guide the model. Therefore, in the embodiment, a Libra R-CNN method is adopted, a first-order image sampling is performed through an OHME algorithm, a sampling interval is divided into K, so that samples to be selected are placed in K buckets, all n negative cases to be selected (Libra R-CNN mainly concerns negative cases, but the method can be also used for positive cases) can be uniformly distributed in the buckets, and then uniform sampling is performed in each bucket, so that sampling is more balanced as much as possible.

Further, in an embodiment of the present invention, the image features are extracted by using the Libra R-CNN-based method as follows:

and 710, extracting the features of each candidate region based on the convolutional neural network to obtain original level features corresponding to different neural network layers.

And 720, fusing the original level features to obtain fused semantic features.

And step 730, processing the original level features based on the fused semantic features to obtain corresponding enhanced semantic features.

And 740, representing the enhanced semantic features of the output layer of the convolutional neural network as feature vectors corresponding to the candidate regions.

It should be noted that, in order to solve the problem of feature imbalance in the R-CNN method, Libra R-CNN proposes a method of fusing balanced semantic features. Specifically, feature imbalance refers to deep networks where higher layers have more semantic information and shallow layers have more detailed information. The features of the high and shallow layers may complement each other to improve the effect of object detection, and how they are used to integrate pyramid characterization determines the performance of the detector. Studies have shown that the integrated features should be able to handle imbalance information in each resolution. The prior art sequential integration approaches all focus on features at adjacent resolutions and do not focus on features at other layers. Therefore, in the information flow, the semantic information of the non-adjacent layer is diluted once every time the semantic information is integrated. Therefore, this embodiment uses the Libra R-CNN method, using a new SSD-based efficient pyramid structure that integrates these oracle image features in a highly nonlinear but efficient manner, fusing balanced semantic features to enhance the original features, each resolution in the feature pyramid can obtain the same information from the other resolutions, thereby balancing the information flow and making the features more discriminative.

Further, in an embodiment of the present invention, the target detection model is trained by using the following loss function based on the Libra R-CNN method:

αln(b+1)＝γ。

it should be noted that, in order to solve the target imbalance problem existing in the R-CNN method, Libra R-CNN introduces a balance L1 loss function. Target imbalance refers to two targets in target detection network training-if localization and classification are not well balanced, compromise of one target can compromise the high performance of the other target, thereby degrading the overall performance of the model. The same is true for the samples involved in the training process, since the goal of regression is not bounded, directly increasing the weight of the regression loss will make the model more sensitive to peripheral samples (sample loss greater than 1.0). The peripheral samples can be regarded as difficult samples, the difficult samples can generate large gradients, the training is not facilitated, and the inner peripheral samples (the sample loss is less than 1.0) can be simple samples. If not well balanced, the small gradients produced by a simple sample may be overwhelmed by the large gradients produced by a mishandled sample, limiting further refinement, thus requiring re-balancing of the involved tasks and samples to achieve better convergence. Therefore, in the embodiment, the Libra R-CNN method is adopted to introduce the balance L1 loss function, so that the balance between the task of positioning and classifying the single oracle characters, the peripheral forged nail sample and the inner forged nail sample can be realized, and the training effect of the target detection model can be improved.

In addition, in the aspect of experiments, the invention adopts a Libra R-CNN target detection method, wherein the identification module selects ResNet 50, and meanwhile, the method of forging nail chips is utilized to generate the nail data and is divided into a training set and a verification set. The target detection model is trained on the training set, so that a good identification effect is obtained, and the mAP index on the verification set reaches 0.995. FIG. 8 is a schematic diagram of the detection effect of oracle-bone inscription, in which (a) and (c) are original images of the nail, and (b) and (d) are images after automatic labeling and analysis, so that the target detection model trained by using the method can obtain a good detection effect in a real scene in cooperation with an image preprocessing method,

the above experiments show that the automatic nail annotation analysis method provided by the invention has good detection and identification capabilities of oracle characters, and can be used as an auxiliary tool in the annotation work of newly unearthed nails to help researchers to label more efficiently; meanwhile, the corresponding modern Chinese character sequence can be generated by utilizing the automatic labeling result, so that the conversion from the picture to the character sequence is realized, and the oracle identification system has the potential.

The automatic nail annotation analyzing device provided by the invention is described below, and the automatic nail annotation analyzing device described below and the automatic nail annotation analyzing method described above can be referred to in a corresponding manner.

Fig. 9 is a schematic structural diagram of an automatic nail annotation analyzing device according to an embodiment of the present invention, as shown in fig. 9, the device includes:

the preprocessing unit 910 is configured to obtain an original nail image and remove background noise of the original nail image to obtain a preprocessed nail image.

And the target detection unit 920 is configured to input the nail image to the target detection model to obtain a target detection result output by the target detection model.

The target detection result comprises an oracle area and a corresponding category, and the target detection model is obtained based on the training of a forged oracle sample.

According to the automatic nail annotation analysis device provided by the invention, the interference of background noise on the detection of the oracle target is reduced by introducing the preprocessing of the nail image, and the forged nail is used for training the target detection model, so that the application effect of the target detection model in a real scene is ensured, and the accurate identification of the oracle in a complex nail is realized.

Further, in an embodiment of the present invention, the automatic nail annotation analyzing apparatus further includes a data sample generating unit, configured to:

based on a preset single-character oracle data set and an arrangement mode of the oracle on the nail, randomly placing the single-character oracle in the single-character oracle data set in a simulated arrangement mode, and adding random noise to generate a forged nail data set.

Further, in an embodiment of the present invention, the preprocessing unit 910 is further configured to:

Further, in an embodiment of the present invention, the target detecting unit 920 is further configured to:

and selecting candidate regions of the first image to obtain a candidate region set.

And extracting the features of each candidate region based on the convolutional neural network to obtain corresponding feature vector representation.

And inputting the feature vector representation into a classifier to obtain a corresponding category and confidence.

And sequentially calculating the overlapping degree between every two candidate regions.

And if the overlapping degree is greater than a preset threshold value, deleting the candidate region corresponding to the characteristic vector representation with low confidence coefficient from the candidate region set to obtain a candidate region subset.

And adjusting the frame of the candidate area in the candidate area subset to obtain the corrected oracle area.

Further, in an embodiment of the present invention, the data sample generating unit is further configured to:

and dividing the sampling interval of the forged A-plate data set to obtain a plurality of sampling subintervals.

And selecting the corresponding positive sample to be selected or the negative sample to be selected from the sampling subinterval as the forged nail sample.

and extracting the characteristics of each candidate region based on the convolutional neural network to obtain the original level characteristics corresponding to different neural network layers.

And fusing the original level features to obtain fused semantic features.

And processing the original level features based on the fused semantic features to obtain corresponding enhanced semantic features.

Further, in an embodiment of the present invention, the automatic nail annotation analyzing apparatus further includes a model training unit, configured to:

training the target detection model using the following loss function:

αln(b+1)＝γ。

fig. 10 illustrates a physical structure diagram of an electronic device, and as shown in fig. 10, the electronic device may include: a processor (processor)1010, a communication Interface (Communications Interface)1020, a memory (memory)1030, and a communication bus 1040, wherein the processor 1010, the communication Interface 1020, and the memory 1030 communicate with each other via the communication bus 1040. Processor 1010 may invoke logic instructions in memory 1030 to perform a method of automatic annotation resolution for a nail, the method comprising: obtaining an original A picture and removing background noise of the original A picture to obtain a preprocessed A picture; inputting the first image to a target detection model to obtain a target detection result output by the target detection model; the target detection result comprises an oracle area and a corresponding category, and the target detection model is obtained based on forged oracle samples.

Furthermore, the logic instructions in the memory 1030 can be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

In another aspect, the present invention further provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, when the program instructions are executed by a computer, the computer being capable of executing the method for automatic nail annotation parsing provided in the foregoing embodiments, the method including: obtaining an original A picture and removing background noise of the original A picture to obtain a preprocessed A picture; inputting the first image to a target detection model to obtain a target detection result output by the target detection model; the target detection result comprises an oracle area and a corresponding category, and the target detection model is obtained based on forged oracle samples.

In yet another aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, the computer program being implemented by a processor to perform the method for automatic nail annotation parsing provided in the foregoing embodiments, the method including: obtaining an original A picture and removing background noise of the original A picture to obtain a preprocessed A picture; inputting the first image to a target detection model to obtain a target detection result output by the target detection model; the target detection result comprises an oracle area and a corresponding category, and the target detection model is obtained based on forged oracle samples.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. An automatic nail plate labeling and analyzing method is characterized by comprising the following steps:

2. The method for automatically labeling and analyzing nail plates according to claim 1, wherein the step of generating the forged nail plate sample comprises the following steps:

3. The method for automatic nail annotation analysis according to claim 1, wherein the removing the background noise of the original nail image comprises:

4. The method for automatically labeling and analyzing the nail plate according to claim 1, wherein the step of inputting the nail plate image into a target detection model to obtain a target detection result output by the target detection model comprises the steps of:

5. The method for automatically labeling and analyzing the nail plate according to claim 2, wherein the step of sampling the forged nail plate data set to obtain a forged nail plate sample comprises the following steps:

6. The method for automatically labeling and analyzing the nail plate according to claim 4, wherein the extracting the features of each candidate region based on the convolutional neural network to obtain the corresponding feature vector representation comprises:

fusing the original level features to obtain fused semantic features;

7. The method for automatically labeling and analyzing the nail according to claim 1, wherein the target detection model is trained by using the following loss function:

αln(b+1)＝γ。

8. the utility model provides an automatic mark analytical equipment of first piece which characterized in that includes:

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method for automatic nail annotation resolution according to any one of claims 1 to 7 when executing the program.

10. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the steps of the method for automatic annotation resolution of a nail plate according to any one of claims 1 to 7.