A method for enhancing the discrimination of root canal ROI identification model

Abstract

The present invention provides a method for enhancing the discrimination of a root canal ROI recognition model. A sample set is prepared based on clinical data images. At the same time, in order to improve the generalization ability of the training model, the sample set is amplified by data enhancement. In order to highlight the difference between root canal-positive teeth (with a history of root canal treatment) and root canal-negative teeth (without a history of root canal treatment), the present invention proposes a method for ROI feature enhancement for root canal treatment history judgment, which is achieved by enhancing the brightness of the protruding parts of root canal-positive teeth. All samples after ROI feature enhancement are divided into training sets and test sets, and input into a convolutional neural network for training and testing. The obtained model can be used to diagnose whether the root canal of a tooth has a treatment history, and the model can be saved for further training and optimization.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a method for enhancing the discrimination of a root canal ROI recognition model. Background Art

[0002] X-ray image-assisted diagnosis based on artificial intelligence is a hot topic in clinical medicine research. It has great academic, economic and social value in breaking through the bottleneck of doctors' subjective diagnosis, developing remote smart medical care, and improving the medical level in remote and difficult areas.

[0003] Artificial intelligence-assisted diagnosis of oral X-ray images is still in its infancy. The present invention mainly performs artificial intelligence-assisted diagnosis on the history of root canal treatment. By collecting clinical oral tooth periapical X-ray image data, the root canal-positive teeth and negative teeth in the periapical X-ray images are manually annotated to establish the sample set required for machine learning. In the identification of root canal treatment history, it is divided into: positive root canal treatment, the tooth has a history of root canal treatment (the tooth has undergone root canal treatment); negative root canal treatment, the tooth does not have a history of root canal treatment (the tooth has not undergone root canal treatment).

[0004] China's invention patent disclosure CN112037913A discloses an intelligent periodontitis detection method and system based on a convolutional neural network. The disclosed samples are panoramic films and periapical films. The periapical films are preprocessed by cutting the edges and flipping the teeth on the periapical films to construct a sample set, and the transfer learning method is used for training and testing.

[0005] Chinese invention patent publication CN112837279A discloses a tooth deformity detection and pre-diagnosis system based on artificial intelligence, specifically disclosing the steps of obtaining an RGB image and a depth image of a detection person: segmenting the tooth area image in the RGB image and analyzing the tooth area to select standard teeth; analyzing each tooth based on the standard teeth to obtain the degree of dental arch curvature, the left-right offset of the tooth, and the front-back offset of the tooth; and obtaining the tooth deformity grade through classifier processing based on the three features obtained.

[0006] Verma D et al. proposed a CNN-SVM method to perform binary classification of normal and diseased teeth. Diseased teeth include caries and periapical infection, etc. CNN is used as a feature extractor and SVM is used as a classifier. The present invention uses VGG16 as a feature extractor to better extract root canal features. The preprocessing process of this paper only performs OSTU threshold segmentation, and then the binary image is multiplied with the original image, but the preprocessing method is not reasonable enough.

[0007] Mahmoud YE et al. proposed two methods, FNN (feedforward neural network) and K-NN (nearest neighbor algorithm), to predict periapical lesions, using median filtering and average filtering for denoising, histogram equalization for image enhancement, and two-dimensional discrete wavelet transform algorithm for feature extraction. Summary of the invention

[0008] The embodiment of the present invention provides a method for enhancing the discrimination of a root canal ROI recognition model, so as to solve the problems existing in the prior art.

[0009] In order to achieve the above object, the present invention adopts the following technical scheme.

[0010] A method for enhancing the discrimination of a root canal ROI recognition model, comprising:

[0011] S1 constructs a sample set based on the collected root slice root canal original image, and performs amplification processing on the sample set;

[0012] S2 performs ROI feature enhancement operation on the feature map in the amplified sample set by using the color gamut enhancement method to obtain a feature enhancement set;

[0013] S3 obtains a training set and a test set by partitioning the feature enhancement set;

[0014] S4 inputs the training set and the test set into the convolutional neural network for training and testing;

[0015] S5 randomly obtains the original image of the root canal of the root slice and inputs it into the convolutional neural network after executing step S4, and optimizes the convolutional neural network.

[0016] Preferably, step S1 comprises:

[0017] The sample set was augmented by inversion, contrast enhancement, and brightness increase;

[0018] The root slice tooth root canal original image includes the root slice tooth root canal negative image and the root slice tooth root canal positive image.

[0019] Preferably, step S2 comprises:

[0020] S21 performs grayscale conversion on the original image of the root canal of the root slice in the sample set to obtain a grayscale image, and selects a root canal ROI area image from the grayscale image;

[0021] S22, based on all pixel points in the root canal ROI area of ​​the grayscale image, calculate the pixel mean of the root canal ROI area of ​​the grayscale image, and then calculate the difference between the pixel value of each root canal ROI pixel point in the grayscale image and the pixel mean of the root canal ROI area of ​​the grayscale image;

[0022] S23 compares the difference with a preset threshold value, and if the difference exceeds the preset threshold value, increases the brightness of the pixel points in the root canal ROI area of ​​the grayscale image to obtain a feature enhancement set.

[0023] Preferably, the length of the root canal ROI region image in the grayscale image is two-thirds of the length of the grayscale image, and the width of the root canal ROI region image in the grayscale image is four-fifths of the width of the grayscale image.

[0024] Preferably, step S3 comprises:

[0025] The sample set is shuffled, and 80% of the grayscale images are randomly obtained to form the training set, and 20% of the grayscale images are used to form the test set.

[0026] Preferably, the convolutional neural network is a convolutional neural network based on the VGG16 architecture, and step S4 includes:

[0027] S41 selects the Adam optimizer as the optimizer in the convolutional neural network, and selects the Softmax classifier as the classifier in the convolutional neural network;

[0028] S42 General

[0029]

[0030] As the cross entropy loss function in the convolutional neural network; where m represents the number of samples, n is the number of label categories, q(xij) represents the probability that the i-th sample is judged as the j-th label by the model, and p(xij) represents the probability that the i-th sample is actually the j-th label;

[0031] S43 sequentially inputs the training set and the test set into the convolutional neural network for training and testing;

[0032] S44 repeats step S43 multiple times to obtain a test result, and obtains a tested convolutional neural network based on the test result.

[0033] It can be seen from the technical solutions provided by the above-mentioned embodiments of the present invention that the present invention provides a method for enhancing the discrimination of the root canal ROI recognition model, and a sample set is prepared based on clinical data images. At the same time, in order to improve the generalization ability of the training model, the sample set is amplified by data enhancement. In order to highlight the difference between root canal-positive teeth and root canal-negative teeth, the present invention proposes a method for ROI feature enhancement for root canal treatment history judgment, which is achieved by enhancing the brightness of the protruding parts of root canal-positive teeth. All samples after ROI feature enhancement are divided into training sets and test sets, and input into a convolutional neural network for training and testing. The obtained model can be used to diagnose whether the root canal of a tooth has a treatment history, and the model can be saved for further training and optimization. The method provided by the present invention performs ROI (region of interest) regional enhancement on the tooth apex film in the preprocessing stage. The root canals of teeth with a history of root canal treatment will have bright colors. In view of this feature, the bright colors are further enhanced to better distinguish teeth with negative root canals; this fills the gap in artificial intelligence-assisted diagnosis of root canal treatment history, uses convolutional neural networks to train models, and the network architecture used in the present invention is VGG16, which has strong feature extraction capabilities. The Softmax used in the fully connected layer is used as a classifier, which can prepare output prediction results.

[0034] Additional aspects and advantages of the present invention will be given in part in the following description, which will become obvious from the following description, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0036] Figure 1 A processing flow chart of a method for enhancing the discrimination of a root canal ROI identification model provided by the present invention;

[0037] Figure 2 A flowchart of a preferred embodiment of a method for enhancing the discrimination of a root canal ROI identification model provided by the present invention;

[0038] Figure 3 A schematic diagram of the structure of a convolutional neural network in a method for enhancing the discrimination of a root canal ROI recognition model provided by the present invention;

[0039] Figure 4 A flow chart of a test embodiment of a method for enhancing the discrimination of a root canal ROI recognition model provided by the present invention;

[0040] Figure 5 A schematic diagram of the effect of ROI enhancement in a test embodiment of a method for enhancing the discrimination of a root canal ROI recognition model provided by the present invention;

[0041] Figure 6 A comparison diagram of processing results of a test embodiment of a method for enhancing the discrimination of a root canal ROI identification model provided by the present invention. DETAILED DESCRIPTION

[0042] The embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention.

[0043] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items.

[0044] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0045] To facilitate understanding of the embodiments of the present invention, several specific embodiments will be further explained below with reference to the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present invention.

[0046] The present invention provides a method for enhancing the accuracy of a root canal ROI (region of interest) identification model, which is used to solve the following problems in the prior art:

[0047] At present, there are not many studies on artificial intelligence-assisted diagnosis of oral X-ray images. At the same time, the few studies that have been conducted have focused on caries, periodontitis and wisdom teeth, and no artificial intelligence-assisted diagnosis scheme for root canal treatment history has been proposed. Even though the status after root canal treatment is more obvious in X-rays and easy to judge, it is also an indispensable part of artificial intelligence-assisted diagnosis in clinical medicine;

[0048] In the existing technology, whether it is caries, periodontitis, wisdom teeth or other dental diseases, the pre-processing methods of samples are too simple, which may result in the key features of the corresponding dental diseases not being highlighted well, which will have a great impact on the accuracy of the final diagnosis. If the features of the corresponding dental diseases are highlighted and then used as training samples for training, the accuracy will be greatly improved;

[0049] Dental lesions and treatment status are often reflected in subtle details in periapical films. If the ROI area enhancement operation is not performed on the samples before training, the trained model will be insensitive to the ROI area and will have a great impact on the robustness of the model, resulting in low accuracy of the final prediction.

[0050] See also Figure 1 The present invention provides a method for enhancing the discrimination of a root canal ROI recognition model, comprising the following steps:

[0051] S1 constructs a sample set based on the collected root slice root canal original image, and performs amplification processing on the sample set;

[0052] S2 performs ROI feature enhancement operation on the feature map in the amplified sample set by using the color gamut enhancement method to obtain a feature enhancement set;

[0053] S3 is based on the feature enhancement set and simultaneously obtains a training set and a test set by partitioning;

[0054] S4 inputs the training set and the test set into a convolutional neural network for training and testing;

[0055] S5 randomly obtains the original image of the root canal of the root slice and inputs it into the convolutional neural network after executing step S4, saves the convolutional neural network model parameters, and continuously optimizes the convolutional neural network.

[0056] The optimized convolutional neural network is used to identify and process root canal ROI images.

[0057] Furthermore, in a preferred embodiment of the present invention, step S1 includes:

[0058] Amplifying the sample set by inverting, enhancing contrast and increasing brightness;

[0059] The root slice tooth root canal original image includes a root slice tooth root canal negative image and a root slice tooth root canal positive image.

[0060] Further, in a preferred embodiment provided by the present invention, step S2 includes:

[0061] S21 performs grayscale conversion on the original image of the root canal of the root slice in the sample set to obtain a grayscale image, and selects a root canal ROI area image from the grayscale image;

[0062] For all samples, the X-ray film may cause inconsistent light and dark intensities due to exposure and other reasons when shooting, so it is necessary to obtain the pixel mean of each image to be processed as a reference standard for the sample image; therefore, the execution process of sub-step S22 is to calculate the pixel mean of the root canal ROI area of ​​the grayscale image based on all pixel points of the root canal ROI area of ​​the grayscale image, and then calculate the difference between the root canal ROI pixel point value of each grayscale image and the pixel mean of the root canal ROI area of ​​the grayscale image;

[0063] S23 compares the difference with a preset threshold value, and if the difference exceeds the preset threshold value, increases the brightness of the pixel points in the root canal ROI area of ​​the grayscale image to obtain a feature enhancement set.

[0064] The root canals are roughly distributed in the center of the image. In order to effectively eliminate the influence of image boundary noise or other factors, in an improved method, the length of the root canal ROI area image in the grayscale image is two-thirds of the length of the grayscale image, and the width of the root canal ROI area image in the grayscale image is four-fifths of the width of the grayscale image.

[0065] Furthermore, step S3 includes:

[0066] The sample set is shuffled, and 80% of the grayscale images are randomly obtained to form the training set, and 20% of the grayscale images are used to form the test set.

[0067] In the preferred embodiment provided by the present invention, the convolutional neural network adopts a convolutional neural network based on the VGG16 architecture. The VGG16 algorithm structure consists of a convolutional layer, a pooling layer, and a fully connected layer, and the activation function of all hidden layers adopts the ReLU activation function. The VGG16 algorithm uses multiple convolutional layers with smaller convolutional kernels (3×3) to replace a convolutional layer with a larger convolutional kernel. On the one hand, it can reduce parameters, and on the other hand, it is equivalent to performing more nonlinear mappings, which can increase the fitting and expression capabilities of the network. The specific process of the VGG16 neural network is as follows: Figure 3 as shown.

[0068] Four optimizers are commonly used in deep learning: SGD, RMSProp, and Adam. SGD introduces noise while randomly selecting gradients, so that the direction of weight update may not be correct. In addition, SGD has not been able to overcome the problem of local optimal solutions alone. The RMSProp algorithm has been empirically proven to be an effective and practical deep neural network optimization algorithm. It is currently one of the optimization methods frequently used by deep learning practitioners. However, in actual operation, Adam is more of a default algorithm and is generally better than RMSProp. Therefore, in this embodiment, the Adam optimizer is selected as the optimizer in the convolutional neural network.

[0069] In this embodiment, the classifier belongs to the fully connected layer of the neural network. Its specific function is to give the probability of judging the sample. The Softmax used as the classifier will simultaneously output the probability of judging the sample as a negative sample and a positive sample, and the sum of the probabilities of the two is 1. The loss function calculates the value of the loss function by cross entropy based on the probability judgment result output by the classifier and the true result. Each time the classifier is trained, the classifier will change once, so the loss function is also constantly changing with the number of training times. The value of the loss function will also decrease with the increase of the number of training times.

[0070] In this embodiment, the cross entropy loss function (CrossEntropy Loss) in the convolutional neural network is a loss function often used in classification problems. In particular, when the neural network is used for classification problems, cross entropy is often used as a loss function. Cross entropy involves calculating the probability of each category, so cross entropy appears almost every time with the Softmax function. The loss of a training number is shown in the following formula, where m represents the number of samples, n is the number of label categories, x represents a sample, p(x) is often used to represent the true distribution of the sample, q(x) is used to represent the distribution predicted by the model, i represents the sample number, j represents the label number, q(xij) represents the probability that the i-th sample is judged by the model as the j-th label, and p(xij) represents the probability that the i-th sample is actually the j-th label (only 0 and 1).

[0071]

[0072] Step S4 includes:

[0073] S41 selects an Adam optimizer as the optimizer in the convolutional neural network, and selects a Softmax classifier as the classifier in the convolutional neural network;

[0074] S42 General

[0075]

[0076] As the cross entropy loss function in the convolutional neural network; where m represents the number of samples and n is the number of label categories;

[0077] S43 sequentially inputs the training set and the test set into the convolutional neural network for training and testing;

[0078] S44 repeats steps S42 and S43 multiple times to obtain a test result, and obtains the tested convolutional neural network based on the test result.

[0079] The trained and tested convolutional neural network can be used to identify the root canal ROI.

[0080] Those skilled in the art should understand that the above-mentioned convolutional neural network application types are only examples, and other existing or future convolutional neural network application types such as VGG19 and AlexNet may be applicable to the embodiments of the present invention, should also be included in the scope of protection of the present invention, and are included here by reference.

[0081] The present invention also provides an embodiment to show the effect of implementing the method. The processing flow is as follows:

[0082] Collect clinical data images to construct a sample set, and amplify the sample set by inverting, enhancing contrast, and increasing brightness; the sample set includes root slice root canal negative images and root slice root canal positive images;

[0083] The original root canal images of the root slices in the sample set are converted into grayscale to obtain grayscale images;

[0084] Determine the root canal ROI area: The root canal is roughly distributed in the center of the image, so the left and right two-thirds of the center point of the image and the upper and lower four-fifths are selected as the root canal ROI area;

[0085] All pixels in the root canal ROI area: All pixels in the root canal ROI area are compared with the average value of the image. The threshold is determined after multiple tests. When the difference between the pixel point and the pixel average value of the sample image is greater than the threshold, the pixel intensity is enhanced to make the feature more obvious. The effect is as follows: Figure 5 As shown: It can be seen that the feature ROI effect of the positive sample after processing is further enhanced, while the negative sample has no significant change. After processing, the distinction between positive and negative samples can be made greater. After feature extraction by the convolutional neural network, the difference is more obvious and easier to distinguish;

[0086] The sample set is shuffled, and 80% of the grayscale images are randomly obtained to form the training set, and 20% of the grayscale images are used to form the test set;

[0087] Selecting Adam optimizer as the optimizer in the convolutional neural network, and selecting Softmax classifier as the classifier in the convolutional neural network;

[0088] General

[0089]

[0090] As the cross entropy loss function in the convolutional neural network; where m represents the number of samples and n is the number of label categories;

[0091] Inputting the training set and the test set into the convolutional neural network for training and testing in sequence;

[0092] The above training and testing process is repeated for 200 iterations to obtain a test result, and the tested convolutional neural network is obtained based on the test result.

[0093] The original root canal image of the root slice is randomly obtained and input into the convolutional neural network after executing step S4, the convolutional neural network model parameters are saved, and the convolutional neural network is continuously optimized.

[0094] The test results of the model are as follows Figure 6 As shown in the confusion matrix, Figure a is after ROI enhancement, and Figure b is not after ROI enhancement. A total of 184 samples were used, half of which were positive and half were negative. It can be seen that the prediction effect of the model after ROI enhancement is significantly better than that of the model without ROI enhancement.

[0095] In summary, the present invention provides a method for enhancing the discrimination of a root canal ROI recognition model. A sample set is prepared based on clinical data images. At the same time, in order to improve the generalization ability of the training model, the sample set is amplified by data enhancement. In order to highlight the difference between root canal-positive teeth (with a history of root canal treatment) and root canal-negative teeth (without a history of root canal treatment), the present invention proposes a method for ROI feature enhancement for root canal treatment history judgment, which is achieved by enhancing the brightness of the protruding parts of root canal-positive teeth. All samples after ROI feature enhancement are divided into training sets and test sets, and input into a convolutional neural network for training and testing. The obtained model can be used to diagnose whether the root canal of a tooth has a history of treatment, and the model can be saved for further training and optimization. The method provided by the present invention performs ROI (region of interest) regional enhancement on the tooth apex film in the preprocessing stage. The root canals of teeth with a history of root canal treatment will have bright colors. In view of this feature, the bright colors are further enhanced to better distinguish teeth with negative root canals; this fills the gap in artificial intelligence-assisted diagnosis of root canal treatment history, uses convolutional neural networks to train models, and the network architecture used in the present invention is VGG16, which has strong feature extraction capabilities. The Softmax used in the fully connected layer is used as a classifier, which can prepare output prediction results.

[0096] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of an embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present invention.

[0097] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0098] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0099] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for enhancing the discrimination of a root canal ROI identification model, characterized in that: include: S1 constructs a sample set based on the collected root slice root canal original image, and performs amplification processing on the sample set; S2 performs ROI feature enhancement operation on the feature map in the amplified sample set by using the color gamut enhancement method to obtain a feature enhancement set; specifically, it includes: S21 performs grayscale conversion on the original image of the root canal of the root slice in the sample set to obtain a grayscale image, and selects a root canal ROI area image from the grayscale image; S22, based on all pixel points in the root canal ROI area of ​​the grayscale image, calculating the pixel mean of the root canal ROI area of ​​the grayscale image, and then calculating the difference between the pixel value of each root canal ROI pixel point in the grayscale image and the pixel mean of the root canal ROI area of ​​the grayscale image; S23 compares the difference with a preset threshold, and if the difference exceeds the preset threshold, increases the brightness of the pixel points in the root canal ROI area of ​​the grayscale image to obtain the feature enhancement set; S3 obtains a training set and a test set by partitioning the feature enhancement set; S4 inputs the training set and the test set into a convolutional neural network for training and testing; S5 randomly obtains the original image of the root canal of the root slice and inputs it into the convolutional neural network after executing step S4, and optimizes the convolutional neural network.

2. The method according to claim 1, characterized in that Step S1 includes: Amplifying the sample set by inverting, enhancing contrast and increasing brightness; The root slice tooth root canal original image includes a root slice tooth root canal negative image and a root slice tooth root canal positive image.

3. The method according to claim 1, characterized in that The length of the root canal ROI region image in the grayscale image is two-thirds of the length of the grayscale image, and the width of the root canal ROI region image in the grayscale image is four-fifths of the width of the grayscale image.

4. The method according to claim 1, characterized in that: Step S3 includes: The sample set is shuffled, and 80% of the grayscale images are randomly obtained to form the training set, and 20% of the grayscale images are used to form the test set.

5. The method according to claim 4, characterized in that The convolutional neural network is a convolutional neural network based on the VGG16 architecture, and step S4 includes: S41 selects an Adam optimizer as the optimizer in the convolutional neural network, and selects a Softmax classifier as the classifier in the convolutional neural network; S42 General As the cross entropy loss function in the convolutional neural network; where m represents the number of samples, n is the number of label categories, q(xij) represents the probability that the i-th sample is judged by the model to be the j-th label, and p(xij) represents the probability that the i-th sample is actually the j-th label; S43 sequentially inputs the training set and the test set into the convolutional neural network for training and testing; S44 repeats step S43 multiple times to obtain a test result, and obtains the tested convolutional neural network based on the test result.

Images