Part recognition method, device, apparatus and medium

Abstract

The application discloses a part recognition method and device, equipment and medium, and relates to the technical field of computers, which comprises the following steps: acquiring an image of a target object by using a preset image acquisition tool to obtain a to-be-recognized image of the target object; and recognizing the to-be-recognized image by using a target part recognition model created in advance based on a multi-label classification algorithm to obtain a target image containing a target part label. Through the above scheme, the to-be-recognized image of the target object is obtained, and when the to-be-recognized image is recognized, the target part recognition model created in advance based on the multi-label classification algorithm is used, so that the situation that only one part label can be obtained in the target image due to the use of a single-label training model is avoided, that is, each part can be accurately recognized, each part is accurately labeled with a part label, part information, lesion information and other information are reflected, and user experience is improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to methods, devices, equipment and media for identifying body parts. Background Technology

[0002] Capsule endoscopy, also known as "intelligent capsule digestive tract endoscopy system" or "medical wireless endoscopy," works by having the patient orally ingest a smart capsule containing a built-in camera and signal transmission device. The capsule moves through the digestive tract using peristalsis, capturing images of various parts and labeling them for easy observation by the doctor. Capsule endoscopy offers advantages such as convenience, non-invasiveness, no wires, no pain, no cross-infection, and no disruption to the patient's normal work. However, the gastrointestinal tract is tightly connected, and an image captured by a single capsule endoscope may contain multiple areas, such as the gastric body and gastric angle. Furthermore, capsule endoscopy is used to detect gastric lesions, which exhibit certain patterns, such as multiple polyps on the greater curvature of the gastric body and multiple ulcers in the gastric antrum. For these reasons, current technologies cannot accurately identify each area using site recognition models, resulting in low precision.

[0003] In summary, improving the accuracy of part identification is a problem that needs to be solved in this field. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for identifying body parts, which can improve the accuracy of body part identification. The specific solution is as follows:

[0005] Firstly, this application discloses a method for identifying body parts, including:

[0006] The target object is captured using a preset image acquisition tool to obtain an image of the target object to be identified;

[0007] The image to be identified is identified using a target part recognition model pre-created based on a multi-label classification algorithm to obtain a target image containing target part labels.

[0008] Optionally, before recognizing the image to be recognized using a target part recognition model pre-created based on a multi-label classification algorithm, the method further includes:

[0009] A multi-label classification algorithm is obtained by combining RNN and CNN algorithms in parallel, and a target part recognition model is created based on the label classification algorithm.

[0010] Optionally, the step of using a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be recognized, in order to obtain a target image containing target part labels, includes:

[0011] The semantic features of the image to be identified are obtained by using the CNN algorithm in the target part recognition model, and the predicted part label set of the image to be identified is obtained by using the RNN algorithm in the target part recognition model. The target part label is determined from the predicted part label set based on the semantic features, so as to obtain a target image containing the target part label.

[0012] Optionally, the step of obtaining a predicted part label set of the image to be identified using the RNN algorithm in the target part recognition model, and determining the target part label from the predicted part label set based on the semantic features, so as to obtain a target image containing the target part label, includes:

[0013] The RNN algorithm in the target part recognition model is used to obtain the predicted part label set of the image to be recognized, and one-hot encoding is used to represent the predicted part labels in the predicted part label set to obtain the target predicted part label.

[0014] The target predicted part label is input into the embedding matrix to obtain the first embedding representation of the target predicted part label, and the second embedding representation of the historical part label stored in the target part recognition model is determined.

[0015] Based on the semantic features, the first embedding representation, and the second embedding representation, the label score of the predicted part label in the predicted part label set is calculated, and the target part label is determined from the predicted part label set using the label score, so as to obtain a target image containing the target part label.

[0016] Optionally, before recognizing the image to be recognized using a target part recognition model pre-created based on a multi-label classification algorithm, the method further includes:

[0017] A multi-label classification algorithm is obtained by using ensemble learning and the Label Powerset algorithm, and a target part recognition model is created based on the label classification algorithm.

[0018] Optionally, the step of using a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be recognized, in order to obtain a target image containing target part labels, includes:

[0019] The target part recognition model, which is pre-created based on a multi-label classification algorithm, is used to recognize the image to be recognized, and the predicted part label set of the image to be recognized is obtained. The multi-class classifier in the target part recognition model is used to calculate the maximum number of votes for each predicted part label in the predicted part label set.

[0020] The predicted part labels with the maximum number of votes greater than a preset threshold in the predicted part label set are determined as the target part labels of the image to be identified, so as to obtain a target image containing the target part labels.

[0021] Optionally, the step of using a preset image acquisition tool to acquire an image of the target object to obtain an image of the target object to be identified includes:

[0022] The target capsule endoscope is identified, and the target object is acquired using the target capsule endoscope to obtain an image of the target object to be identified.

[0023] Secondly, this application discloses a part identification device, comprising:

[0024] The image acquisition module is used to acquire images of the target object using a preset image acquisition tool to obtain the image of the target object to be identified;

[0025] The image recognition module is used to recognize the image to be recognized using a target part recognition model pre-created based on a multi-label classification algorithm, so as to obtain a target image containing target part labels.

[0026] Thirdly, this application discloses an electronic device, comprising:

[0027] Memory, used to store computer programs;

[0028] A processor is configured to execute the computer program to implement the steps of the aforementioned disclosed part identification method.

[0029] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed part identification method.

[0030] As can be seen, this application utilizes a preset image acquisition tool to acquire an image of the target object to obtain an image to be identified; it then uses a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be identified, thereby obtaining a target image containing target part labels. Therefore, by acquiring the image to be identified and using a pre-created target part recognition model based on a multi-label classification algorithm, the application avoids the situation where a single-label training model results in only one part of the target image being displayed. This allows for accurate identification of various parts, precise labeling of each part, and the representation of part information, lesion information, and other information, thus improving the user experience. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0032] Figure 1 This is a flowchart of a part identification method disclosed in this application;

[0033] Figure 2 This is a schematic diagram of a specific model training disclosed in this application;

[0034] Figure 3 This is a flowchart of a specific part identification method disclosed in this application;

[0035] Figure 4 This is a schematic diagram illustrating a specific parallel use of a CNN and an RNN disclosed in this application;

[0036] Figure 5 This is a flowchart of another specific part identification method disclosed in this application;

[0037] Figure 6 This is a schematic diagram of the structure of a part identification device disclosed in this application;

[0038] Figure 7 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0039] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0040] Capsule endoscopy, also known as "intelligent capsule digestive tract endoscopy system" or "medical wireless endoscopy," works by having the patient orally ingest a smart capsule containing a built-in camera and signal transmission device. The capsule moves through the digestive tract using peristalsis, capturing images of various parts and labeling them for easy observation by the doctor. Capsule endoscopy offers advantages such as convenience, non-invasiveness, no wires, no pain, no cross-infection, and no disruption to the patient's normal work. However, the gastrointestinal tract is tightly connected, and an image captured by a single capsule endoscope may contain multiple areas, such as the gastric body and gastric angle. Furthermore, capsule endoscopy is used to detect gastric lesions, which exhibit certain patterns, such as multiple polyps on the greater curvature of the gastric body and multiple ulcers in the gastric antrum. For these reasons, current technologies cannot accurately identify each area using site recognition models, resulting in low precision.

[0041] Therefore, this application provides a part recognition scheme that can improve the accuracy of part recognition.

[0042] See Figure 1 As shown in the figure, this application discloses a method for identifying body parts, including:

[0043] Step S11: Use a preset image acquisition tool to acquire an image of the target object to obtain the image of the target object to be identified.

[0044] In this embodiment, the step of acquiring an image of the target object using a preset image acquisition tool to obtain an image to be identified includes: determining a target capsule endoscope and acquiring an image of the target object using the target capsule endoscope to obtain an image to be identified. A capsule endoscope, also known as a capsule endoscopy, is an endoscope shaped like a capsule. It is a medical instrument used to examine the human intestines. Capsule endoscopes can enter the human body to examine the health of the gastrointestinal tract and esophagus, and can help doctors diagnose diseases of the patient's digestive system.

[0045] Step S12: Use a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be recognized, so as to obtain a target image containing target part labels.

[0046] Before using a target region recognition model to identify an image, the model needs to be trained. For example... Figure 2The diagram illustrates a specific model training process. Initial training images are acquired using a capsule endoscope, and a multi-label classification algorithm is used to determine the body part labels on these images. This results in a target training image containing the body part labels, which is then input into the initial body part recognition model for training. This initial model is then trained to obtain a target body part recognition model, which is used to identify the images to be recognized. Multi-label classification algorithms can be categorized into two types based on their design principles: one is algorithm-adapted multi-label classification methods, which directly process multi-label data by modifying popular learning algorithms. This often involves modifying the output layer of a multi-classification neural network or using multiple models in parallel to adapt them to multi-label classification. This is an example of using a parallel combination of RNN (Recurrent Neural Network) and CNN (Convolutional Neural Network) algorithms. The other type is problem-transformation based multi-label classification methods, which transform the multi-label classification problem into a more sophisticated learning scenario, adapting the algorithm to the data. This is an example of using ensemble learning and Label Powerset (LP) algorithms to obtain multi-label classification algorithms.

[0047] Current methods for identifying gastric locations using capsule endoscopy images employ single-label classification. Based on the characteristics of the images captured by the capsule endoscope, they are categorized into areas such as the gastric body, antrum, and angle. However, images of different gastric locations are not entirely independent; some images may contain multiple locations, such as images captured between the gastric body and angle. Classifying such images using only a single label is inaccurate. If an image contains both the gastric body and angle, classifying it as a gastric body may lead to the model misidentifying an angle image as a gastric body, resulting in poor accuracy. Furthermore, classifying images by location only distinguishes between different gastric locations. Doctors still need to review all capsule endoscopy images when analyzing a patient's condition, leading to a poor user experience. Therefore, multi-label algorithms not only meticulously distinguish between different gastric locations but also label an image with location information, lesion information, and other surgical information. For example, an image of the greater curvature of the stomach body, when normal and without lesions, is labeled as stomach body-greater curvature; when polyps are present, it is labeled as stomach body-greater curvature-polyp; and when surgical instruments are also present, it is labeled as stomach body-greater curvature-polyp-surgical instrument. This multi-labeling algorithm is necessary to assist doctors in interpreting images.

[0048] It is understandable that single-label classification is inaccurate because it only reflects information about the stomach area in the image, without reflecting information about lesions or other details. However, this embodiment uses a multi-label classification algorithm when training the model. Multi-label classification allows the labels of an image to simultaneously reflect information about the location, lesions, and other details. The initial location recognition model is then trained using the target training images after multi-label classification. In this way, the final target location recognition model can identify all the locations contained in the image to be recognized, avoiding the inaccurate location recognition that occurs with single-label classification.

[0049] As can be seen, this application utilizes a preset image acquisition tool to acquire an image of the target object to obtain an image to be identified; it then uses a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be identified, thereby obtaining a target image containing target part labels. Therefore, by acquiring the image to be identified and using a pre-created target part recognition model based on a multi-label classification algorithm, the application avoids the situation where a single-label training model results in only one part of the target image being displayed. This allows for accurate identification of various parts, precise labeling of each part, and the representation of part information, lesion information, and other information, thus improving the user experience.

[0050] See Figure 3 As shown in the figure, this application discloses a specific method for identifying body parts, including:

[0051] Step S21: Use a preset image acquisition tool to acquire an image of the target object to obtain the image of the target object to be identified.

[0052] Step S22: Use the multi-label classification algorithm obtained by paralleling the RNN algorithm and the CNN algorithm, and create a target part recognition model based on the label classification algorithm.

[0053] Step S23: Use the CNN algorithm in the target part recognition model to obtain the semantic features of the image to be recognized, and use the RNN algorithm in the target part recognition model to obtain the predicted part label set of the image to be recognized. Based on the semantic features, determine the target part label from the predicted part label set so as to obtain the target image containing the target part label.

[0054] In this embodiment, the step of obtaining a predicted part label set of the image to be identified using the RNN algorithm in the target part recognition model, and determining the target part label from the predicted part label set based on the semantic features, in order to obtain a target image containing the target part label, includes: obtaining a predicted part label set of the image to be identified using the RNN algorithm in the target part recognition model, and representing the predicted part labels in the predicted part label set using one-hot encoding to obtain the target predicted part label; inputting the target predicted part label into an embedding matrix to obtain a first embedding representation of the target predicted part label, and determining a second embedding representation of the historical part labels stored in the target part recognition model; calculating the label score of the predicted part label in the predicted part label set based on the semantic features, the first embedding representation, and the second embedding representation, and determining the target part label from the predicted part label set using the label score, in order to obtain a target image containing the target part label.

[0055] The algorithm adaptation method uses deep learning to solve multi-label classification problems. The key idea is the fitting of the algorithm to the data, such as... Figure 4 The diagram illustrates a specific parallel use of a CNN and an RNN. In this model, the CNN part extracts semantic features of the image, while the RNN part describes the image / label relationship. Because the RNN can store contextual information through intermediate states and use it as input to influence the prediction of the next time step, it is also used to represent the dependencies between labels in this model. The currently predicted label is represented using one-hot encoding, i.e., label k is represented as follows, where the k-th position is 1 and the remaining positions are 0. The label embedding matrix is ​​shown below, with the k-th row representing the embedding of label k.

[0056] w k =U l ·e k ;

[0057] In the formula, w k The embedding representation of tag k, e k Let e ​​represent the one-hot vector of label k. k =[0,0,0...,1,...,0,0],U l This represents the label embedding matrix. Label embeddings can be derived from one-hot vectors and the label embedding matrix U. l Multiplying them together, we get U l The k-th row is the tag embedding of tag k.

[0058] The RNN part contains representations of previously predicted labels. It combines these representations with the current input label representation to model label relevance. The RNN state update is shown below:

[0059] o(t) = h0(r(t-1), w k (t));

[0060] r(t) = h r (r(t-1),w k (t))

[0061] In the formula, r(t) represents the hidden state of the RNN at step t, o(t) represents the output of the RNN at step t, h0 represents the nonlinear RNN function, and w k (t) represents the label embedding of the t-th label in the predicted path, h r This represents a nonlinear RNN function.

[0062] The memory output of the RNN part and the image semantic features extracted by the CNN part are jointly mapped to a low-dimensional feature embedding space, and and are the mapping matrices, as shown below:

[0063]

[0064] In the formula, x t Represents the mapping matrix, Indicates the output of the loop layer. Let I represent the projection matrix of the image representation, and let I represent the image representation of the convolutional neural network. and The number of columns and the label embedding matrix U l same.

[0065] The label score of the test sample is calculated by measuring the distance between each feature embedding. The probability of predicting the label can be calculated using the softmax function, as shown below:

[0066]

[0067] p = softmax(s(t));

[0068] In the formula, Represents the label embedding matrix U l transpose of x t Let p represent the mapping matrix. p is the predicted label probability, and s(t) outputs the label score of the test sample.

[0069] Using cross-entropy as the loss function of the model, as shown below:

[0070] H(p,q)=-∑p(x)logq(x);

[0071] In the formula, p(x) represents the true label distribution of the sample, q(x) represents the probability distribution of the predicted label, and H(p,q) represents the cross entropy. It can be understood that the smaller the cross entropy, the closer the two probability distributions are.

[0072] Therefore, this application modifies the output layer of the multi-class neural network or uses multiple models in parallel to make it suitable for multi-label classification, and transforms the algorithm to adapt to the data. It can accurately identify each part, accurately label each part with part labels, reflect part information, lesion information and other information, and improve the user experience.

[0073] See Figure 5 As shown in the embodiments of this application, another specific method for identifying body parts is disclosed, including:

[0074] Step S31: Use a preset image acquisition tool to acquire an image of the target object to obtain the image of the target object to be identified.

[0075] Step S32: Utilize the multi-label classification algorithm obtained by ensemble learning and Label Powerset algorithm, and create a target part recognition model based on the label classification algorithm.

[0076] Step S33: Use a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be recognized, so as to obtain a target image containing target part labels.

[0077] In this embodiment, the step of using a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be recognized in order to obtain a target image containing target part labels includes: using the target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be recognized, obtaining a predicted part label set of the image to be recognized, and using a multi-class classifier in the target part recognition model to calculate the maximum number of votes for each predicted part label in the predicted part label set; determining the predicted part labels in the predicted part label set whose maximum number of votes is greater than a preset threshold as the target part labels of the image to be recognized, so as to obtain a target image containing target part labels.

[0078] The problem transformation method converts a multi-label classification problem into a set of multi-class classification problems, where each learner in the set is trained on a random subset of y. A multi-class classifier is trained using a power-law (LP) mapping function, and the key idea is the fitting of data to the algorithm. LP is a direct method to transform a multi-label learning problem into a multi-class classification problem. Its mapping function is denoted as σ. yMapping the power set of y to natural numbers yields its inverse function. During training, LP first transforms the original multi-label training set D into a multi-class training set, treating each distinct label in D as a new class. For example:

[0079] Dataset:

[0080] X1(n1,n2,n3);

[0081] X2(n1,n2);

[0082] X3(n2,n4);

[0083] The current dataset has a total of 4 labels (n1, n2, n3, n4), so the total number of label combinations is 2. 4 =16. Transform (n1,n2,n3) into a new class m1, (n1,n2) into a new class m2, and (n2,n4) into a new class m3, resulting in 3 new classes (m1, m2, m3). This allows us to train the transformed dataset using a multi-class classifier.

[0084] Put 2 q 2 possible label pairs, mapped to 2 q A natural number. Multi-label training set D, D={(x i ,Y i 1≤i≤m)},x i ∈X, It is a d-dimensional eigenvector, y = {y1, y2, ..., y3} q} represents a tag space with q possible class tags, Y i ∈Y is related to x i The relevant label set then transforms the original dataset into D. Y D Y ={(x i ,σ y (Y i )1≤i≤m)}. Here, x i ∈X is an instance representing the attribute (feature) of an object, σ y (Y i )∈Y is the corresponding label representing the object.

[0085] A multi-classifier is trained based on a newly constructed dataset. Given a test sample, the multi-classifier outputs a natural number. The output natural number is then mapped back to the label set. This algorithm is called the LP algorithm. This algorithm has two main limitations: 1) The predicted labels are those that have already appeared in the training set and cannot be generalized to unseen label sets; 2) When the number of labels is large, the algorithm is relatively inefficient.

[0086] To maintain the simplicity of LP while overcoming its limitations, the Random k-labelsets algorithm combines ensemble learning with LP, learning from multi-label data. Its key strategy is to train an LP classifier on only a subset of y of length k, and then ensemble a large number of LP classifiers to make predictions. k Let y represent the set of all possible k-label sets in y. k (l) represents the l-th k-labelsets, This allows us to shrink the sample space, resulting in the following sample set. and tag set

[0087]

[0088] Then, a multi-class classifier is derived using the multi-class learning algorithm M.

[0089]

[0090] To create a set of classifiers with n components, Random k-Labelsets are created from n random k-label sets. k (l r The LP is called on (1≤r≤n), with each set pointing to a multi-class classifier. For an invisible instance x, calculate the following two quantities for each class label:

[0091]

[0092]

[0093] Two indicators, τ(x,y), are calculated during the prediction phase. i Calculate the maximum number of votes for label j, μ(x,y) i ) Calculate the actual number of votes. When predicting unknown samples, use a threshold of 0.5 to make predictions, and obtain the predicted label set Y:

[0094]

[0095] In other words, when the actual number of votes exceeds half of the maximum number of votes, y is considered to have won. j They are related.

[0096] Therefore, this application avoids the situation where only one part can be obtained in the target image due to the use of single-label training model. It can accurately identify each part, accurately label each part, reflect part information, lesion information and other information, and improve the user experience.

[0097] See Figure 6 As shown in the figure, this application discloses a part recognition device, including:

[0098] Image acquisition module 11 is used to acquire images of the target object using a preset image acquisition tool to obtain the image of the target object to be identified;

[0099] The image recognition module 12 is used to recognize the image to be recognized using a target part recognition model pre-created based on a multi-label classification algorithm, so as to obtain a target image containing target part labels.

[0100] As can be seen, this application utilizes a preset image acquisition tool to acquire an image of the target object to obtain an image to be identified; it then uses a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be identified, thereby obtaining a target image containing target part labels. Therefore, by acquiring the image to be identified and using a pre-created target part recognition model based on a multi-label classification algorithm, the application avoids the situation where a single-label training model results in only one part of the target image being displayed. This allows for accurate identification of various parts, precise labeling of each part, and the representation of part information, lesion information, and other information, thus improving the user experience.

[0101] Furthermore, embodiments of this application also provide an electronic device. Figure 7 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0102] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Specifically, it may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the part identification method performed by the electronic device disclosed in any of the foregoing embodiments.

[0103] In this embodiment, the power supply 23 is used to provide operating voltage for various hardware devices on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0104] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0105] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0106] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the part identification method executed by the electronic device as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0107] Furthermore, embodiments of this application also disclose a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the method steps performed during the part identification process disclosed in any of the foregoing embodiments.

[0108] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0109] The above provides a detailed description of the part identification method, apparatus, device, and medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for identifying body parts, characterized in that, include: The target object is captured using a preset image acquisition tool to obtain an image of the target object to be identified; The image to be identified is identified using a target part recognition model pre-created based on a multi-label classification algorithm to obtain a target image containing target part labels; Before using a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be recognized, the method further includes: A multi-label classification algorithm is obtained by combining RNN and CNN algorithms in parallel, and a target part recognition model is created based on the label classification algorithm; The step of using a target part recognition model pre-created based on a multi-label classification algorithm to recognize the image to be recognized, in order to obtain a target image containing target part labels, includes: The semantic features of the image to be identified are obtained using the CNN algorithm in the target part recognition model. The predicted part label set of the image to be identified is obtained using the RNN algorithm in the target part recognition model. One-hot encoding is used to represent the predicted part labels in the predicted part label set to obtain the target predicted part label. The target predicted part label is input into an embedding matrix to obtain a first embedding representation of the target predicted part label. A second embedding representation of the historical part labels stored in the target part recognition model is determined. The first embedding representation is the product of a one-hot vector and an embedding matrix. Based on the semantic features, the first embedding representation, and the second embedding representation, the label score of the predicted part label in the predicted part label set is calculated. The label score is used to determine the target part label from the predicted part label set to obtain a target image containing the target part label. The step of acquiring an image of the target object using a preset image acquisition tool to obtain an image of the target object to be identified includes: The target capsule endoscope is identified, and the target object is acquired using the target capsule endoscope to obtain an image of the target object to be identified.

2. A part identification device, characterized in that, The steps for implementing the part recognition method as described in claim 1 include: The image acquisition module is used to acquire images of the target object using a preset image acquisition tool to obtain the image of the target object to be identified; The image recognition module is used to recognize the image to be recognized using a target part recognition model pre-created based on a multi-label classification algorithm, so as to obtain a target image containing target part labels.

3. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the part recognition method as described in claim 1.

4. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the part recognition method as described in claim 1.

Images