Model training methods, image recognition methods and devices

Abstract

This application discloses a model training method and apparatus, relating to the field of image processing technology in the medical and health fields. One specific implementation of the method includes: acquiring a training sample set; constructing an initial recognition model based on at least two different feature extraction networks and a preset ensemble learning classifier; and training the initial recognition model using the training sample set to obtain a recognition model. This implementation effectively improves the recognition accuracy of the trained recognition model for identifying disease image feature categories in first-class images.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to the field of image processing technology in the medical and health field, and particularly to a model training method, an image recognition method, and an apparatus. Background Technology

[0002] Current technologies for automatic feature classification of pediatric lung disease images based on deep learning are still in their initial stages. Existing methods mainly fall into two categories: 1) Designing a corresponding deep neural network and directly training it on a small-scale pediatric lung disease dataset to determine the feature category of lung disease images; 2) Employing transfer learning, i.e., pre-training a classic neural network on the public ImageNet dataset and then training it on chest X-rays to achieve the final prediction. However, neural networks in method 1) are prone to overfitting, poor generalization ability, and low accuracy; while the images in ImageNet in method 2) are all real-world images, which can lead to lower accuracy in identifying lung disease image feature categories using the trained neural network. Summary of the Invention

[0003] This application provides a model training method, apparatus, device, and storage medium.

[0004] According to the first aspect, embodiments of this application provide a model training method, which includes: acquiring a training sample set, constructing an initial recognition model based on at least two different feature extraction networks and a preset ensemble learning classifier, and training the initial recognition model using the training sample set to obtain the recognition model.

[0005] According to a second aspect, embodiments of this application provide an image recognition method, the method comprising: acquiring a first type of image including body parts; inputting the first type of image into a recognition model to obtain a disease image feature category corresponding to the body parts, wherein the recognition model is a recognition model obtained by the method described in any implementation of the first aspect above.

[0006] According to a third aspect, embodiments of this application provide a model training apparatus, the apparatus comprising: an acquisition module configured to acquire a training sample set; a construction module configured to construct an initial recognition model based on at least two different feature extraction networks and a preset ensemble learning classifier; and a training module configured to train the initial recognition model using the training sample set to obtain a recognition model.

[0007] According to a fourth aspect, embodiments of this application provide an image recognition device, the device comprising: an acquisition module configured to acquire a first type of image including body parts; and a recognition module configured to input the first type of image into a recognition model to obtain a disease image feature category corresponding to the body parts, wherein the recognition model is a recognition model obtained by the method described in any implementation of the first aspect above.

[0008] According to a fifth aspect, embodiments of this application provide an electronic device including one or more processors; a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first or second aspect.

[0009] According to a sixth aspect, embodiments of this application provide a computer-readable medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any of the implementations of the first or second aspect.

[0010] This application acquires a training sample set, constructs an initial recognition model based on at least two different feature extraction networks and a pre-defined ensemble learning classifier, and trains the initial recognition model using the training sample set to obtain a trained recognition model. Specifically, it trains a disease image feature category recognition network with strong generalization ability and good robustness based on multiple classic convolutional neural networks and a second-class image dataset. Based on this recognition network, a feature extraction network is determined for feature transfer learning of first-class images. This allows the pathological knowledge learned in the disease image feature category recognition network to be generalized to the disease image feature category recognition of first-class images, thereby improving the recognition accuracy of disease image feature categories of first-class images without the need for a large amount of first-class image data labeled with disease image feature categories.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0013] Figure 2a This is a flowchart of an embodiment of the model training method according to this application;

[0014] Figure 2b This is a schematic diagram of an embodiment of the model training method according to this application;

[0015] Figure 3This is a schematic diagram of an application scenario of the model training method according to this application;

[0016] Figure 4 This is a flowchart of yet another embodiment of the image recognition method according to this application;

[0017] Figure 5 This is a schematic diagram of one embodiment of the model training apparatus according to this application;

[0018] Figure 6 This is a schematic diagram of an embodiment of the image recognition device according to this application;

[0019] Figure 7 This is a schematic diagram of the structure of a computer system suitable for implementing the server embodiments of this application. Detailed Implementation

[0020] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] Figure 1 An exemplary system architecture 100 is shown, in which the model training method of this application can be applied.

[0023] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0024] Terminal devices 101, 102, and 103 interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications, such as image recognition applications and communication applications, can be installed on terminal devices 101, 102, and 103.

[0025] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays, including but not limited to mobile phones and laptops. When terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software programs or software modules (e.g., for providing model training services) or as a single software program or software module. No specific limitations are made here.

[0026] Server 105 can be a server that provides various services, such as acquiring a training sample set, constructing an initial recognition model based on at least two different feature extraction networks and a preset ensemble learning classifier, and training the initial recognition model using the training sample set to obtain the recognition model.

[0027] It should be noted that server 105 can be either hardware or software. When server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules (e.g., used to provide services for training models), or as a single software program or software module. No specific limitations are made here.

[0028] It should be noted that the model training method provided in the embodiments of this disclosure can be executed by server 105, by terminal devices 101, 102, and 103, or by server 105 and terminal devices 101, 102, and 103 in cooperation with each other. Accordingly, the various parts (e.g., various units, sub-units, modules, and sub-modules) of the model training apparatus can all be set in server 105, all in terminal devices 101, 102, and 103, or they can be set in server 105 and terminal devices 101, 102, and 103 respectively.

[0029] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0030] Figure 2a A flowchart 200 illustrates an embodiment of the model training method applicable to this application. In this embodiment, the model training method includes the following steps:

[0031] Step 201: Obtain the training sample set.

[0032] In this embodiment, the execution subject (such as...) Figure 1The server 105 or terminal devices 101, 102, 103 shown can obtain the training sample set from a local or remote server that stores the training sample set.

[0033] The training sample set can be a publicly available set of medical images, and the training samples are first-class images labeled with disease image feature categories.

[0034] Here, the first type of image may include body parts, such as lungs, heart, brain, etc., which are not limited in this application.

[0035] Specifically, the training samples are images of children's lungs labeled with lung disease image feature categories, which can include category A (no disease manifestations), category B, and category C, etc.

[0036] In some alternative methods, obtaining the training sample set includes: obtaining an initial training sample set; performing preprocessing operations on the initial training sample set; and determining the initial training sample set after performing the preprocessing operations as the training sample set.

[0037] In this implementation, the executing entity can first obtain the initial training sample set and divide the initial training sample set into a validation set, a test set, and a training set, that is, perform preprocessing operations according to different sets.

[0038] Furthermore, preprocessing operations are performed on each set of the initial training sample set, and the initial training sample set after the preprocessing operations is determined as the training sample set.

[0039] Here, preprocessing operations can include data augmentation and data standardization. Data augmentation mainly involves using common data augmentation techniques such as horizontal flipping, rotation, cropping, and adding noise to expand the training samples in the initial training sample set, thereby increasing the diversity of the training data and preventing model overfitting.

[0040] Data standardization is mainly aimed at reducing the impact of noise and grayscale variations in training samples on feature extraction and learning.

[0041] For example, the grayscale values ​​of each channel of the data-augmented training sample can be normalized to [-1, 1]. For each data-augmented training sample I, its normalized expression can be represented by the following formula:

[0042] I = (I - 127.5) / 127.5

[0043] This implementation method obtains an initial training sample set; performs preprocessing operations on the initial training sample set; determines the initial training sample set after preprocessing operations as the training sample set; and then uses the training sample set to train the initial recognition model to obtain the recognition model, which helps to further improve the recognition accuracy of the trained recognition model.

[0044] Step 202: Construct an initial recognition model based on at least two different feature extraction networks and a pre-defined ensemble learning classifier.

[0045] In this embodiment, the executing entity can construct an initial recognition model based on at least two different feature extraction networks and a preset ensemble learning classifier.

[0046] The preset ensemble learning classifier can be a multi-classifier system in existing or future technologies, such as Stacking ensemble classifier, Bagging ensemble classifier, etc., and this application does not limit it.

[0047] Stacking ensemble classifier is a hierarchical model ensemble framework, specifically consisting of two layers. The first layer consists of multiple base classifiers, whose input is the original training set. The second layer consists of meta-classifiers. The output of the base classifiers in the first layer is added to the training set as features to retrain the meta-classifiers in the second layer, resulting in the trained Stacking ensemble classifier.

[0048] Here, the number of at least two different feature extraction networks is the same as the number of base classifiers in the preset ensemble learning classifier. One feature extraction network corresponds to one base classifier, and the output of the feature extraction network is the input of the corresponding base classifier.

[0049] The feature extraction network is determined based on the recognition network, which can be trained on second-class images labeled with disease image feature categories in a publicly available dataset. The body parts included in the first-class and second-class images are the same.

[0050] Here, body parts can be lungs, heart, brain, etc., and this application does not limit them.

[0051] It should be noted that the first type of image and the second type of image are different images. The second type of image is easier to obtain than the first type of image, and the amount of data in the second type of image is usually much larger than that in the first type of image.

[0052] For example, the first type of image is an X-ray image of a child's lungs, and the second type of image is an X-ray image of an adult's lungs.

[0053] Here, the recognition network can be an image recognition network in existing or future technologies, such as VGG16, Resent50, Inception-v3, etc.

[0054] It should be noted that different feature extraction networks correspond to different recognition networks. A feature extraction network can be a recognition network, meaning the feature extraction network and the recognition network are equivalent, or it can be the part of the recognition network used to extract features.

[0055] Specifically, the execution entity constructs an initial recognition model based on two different feature extraction networks, such as a first feature extraction network and a second feature extraction network, and a stacking ensemble classifier. The first feature extraction network is based on a first recognition network, such as VGG16 (which includes 13 convolutional layers and 3 fully connected layers). Specifically, the convolutional layers and the first two fully connected layers of VGG16 are used as the first feature extraction network, meaning the features output by the second fully connected layer are used as the extracted features. The second feature extraction network is based on a second recognition network, such as ResNet50. Specifically, the entire ResNet50 can be used as the second feature extraction network.

[0056] In some optional approaches, the default ensemble learning classifier is the Stacking ensemble classifier.

[0057] In this implementation, the Stacking ensemble classifier is a hierarchical model ensemble framework, specifically consisting of two layers. The first layer is composed of multiple base classifiers, and the second layer is composed of meta-classifiers. It can effectively combat overfitting and does not require much parameter tuning and feature selection, resulting in good learning performance.

[0058] In some alternative approaches, an initial recognition model is constructed based on at least two different feature extraction networks and a pre-defined ensemble learning classifier, including: constructing an initial recognition model based on at least three different feature extraction networks and a pre-defined ensemble learning classifier.

[0059] In this implementation, the executing entity can construct an initial recognition model based on at least three different feature extraction networks and a pre-defined ensemble learning classifier.

[0060] Specifically, the execution entity constructs an initial recognition model based on three different feature extraction networks, such as a first feature extraction network, a second feature extraction network, and a third feature extraction network, and a pre-defined ensemble learning classifier. The first feature extraction network is determined based on a first recognition network, such as VGG16. The second feature extraction network is determined based on a second recognition network, such as ResNet50. The third feature extraction network is determined based on a third recognition network, such as Inception-v3.

[0061] This implementation method constructs an initial recognition model based on at least three different feature extraction networks and a pre-set ensemble learning classifier, and then trains the initial recognition model to obtain a trained recognition model, which helps to further improve the recognition accuracy and robustness of the trained recognition model.

[0062] Step 203: Train the initial recognition model using the training sample set to obtain the recognition model.

[0063] In this embodiment, after determining the initial recognition model, the executing entity can train the initial recognition model using a training sample set. If the training termination condition is met, the training is terminated, and the recognition model is obtained.

[0064] The conditions for ending training may include, but are not limited to, at least one of the following: the training time exceeds the preset duration; the number of training iterations exceeds the preset number of iterations; or the loss value calculated using the target loss function is less than the preset loss value threshold.

[0065] Specifically, such as Figure 2b As shown, the initial recognition model includes two feature extraction networks, namely a first feature extraction network and a second feature extraction network, and a Stacking ensemble classifier. The two different feature extraction networks are used to extract features from training samples, such as X-ray images of children's lungs labeled with lung disease image feature categories, to construct a transfer feature training set T = {T1, T2}, where T... k ={(X k，1 y k，1 ), (X k，2 y k，2 )…(X k，N y k，N )},(X k，i y k，i ) represents the features extracted by feature extraction network k from training sample i and the corresponding disease image feature category information, k∈{1,2}, y k,i ∈{1,2,…,M}, where M is the number of disease categories and N is the total number of training samples.

[0066] The Stacking ensemble classifier consists of two base classifiers H1 and H2 and a meta-classifier C. The base classifiers are composed of a Logistic regression model. The Logistic regression model is trained using two features from the transfer feature training set, i.e., T1 trains H1 and T2 trains H2.

[0067] For any data X k,i ∈T k Using base classifier H k Make predictions and obtain the probability distribution. in, This indicates that training sample i has passed through base classifier H. k Predict the probability that it belongs to class j, and then train a meta-classifier C consisting of an SVM based on the dataset constructed from the probability distribution.

[0068] In addition, the training sample set can be divided into a validation set, a test set, and a training set, and the initial recognition model is trained using the validation set, the test set, and the training set.

[0069] Specifically, the training sample set includes 5856 training samples. The training samples are X-ray images of children's lungs labeled with lung disease image feature categories. There are three categories of lung disease image features: category A, category B, and category C. For each category of training samples, 10% is randomly selected as the validation set, 10% as the test set, and the remaining 80% as the training set.

[0070] Here, the validation set is used to evaluate the convergence of the model and whether the model is overfitting during the model training process. That is, after the model is trained on the training set a certain number of times, it will be evaluated on the validation set. If the performance on the training set continues to rise, but the performance on the validation set does not rise but falls, it indicates overfitting. When the performance of the model on the training set and the validation set is consistent and tends to stabilize, it indicates that the model training has converged.

[0071] In some alternative approaches, the initialization parameters during the training process of the recognition network are determined by using model parameters trained on a pre-defined image dataset as the initialization parameters of the recognition network.

[0072] In this implementation, the executing entity can use the model parameters trained on a preset image dataset as the initialization parameters of the recognition network.

[0073] The preset image dataset can be any existing image dataset, such as a medical image dataset, the ImageNet dataset, etc., and this application does not limit it.

[0074] The ImageNet dataset is a public dataset, and the images in the ImageNet dataset are all natural images from the real world.

[0075] This implementation method uses model parameters trained on a pre-set image dataset as initialization parameters for the recognition network, which helps the recognition network converge to the global optimum more quickly.

[0076] See also Figure 3 , Figure 3 This is a schematic diagram illustrating an application scenario of the model training method according to this embodiment.

[0077] exist Figure 3In the application scenario, the execution entity 301 can obtain a training sample set 302 from a local or remote server that stores the training sample set. The training samples are first-class images labeled with disease image feature categories, such as X-ray images of children's lungs labeled with lung disease image feature categories. Based on at least two different feature extraction networks 303 and a Stacking ensemble classifier 304, an initial recognition model 305 is constructed. The number of at least two different feature extraction networks is the same as the number of base classifiers in the Stacking ensemble classifier. One feature extraction network corresponds to one base classifier, and the output of the feature extraction network is the input of the corresponding base classifier. The feature extraction network is determined based on the recognition network, such as VGG16 and ResNet50. The recognition network is trained based on second-class images labeled with disease image feature categories, such as X-ray images of adult lungs labeled with lung disease image feature categories. The initial recognition model 305 is trained using the training sample set 302 to obtain the recognition model 306.

[0078] The model training method disclosed herein obtains a training sample set, constructs an initial recognition model based on at least two different feature extraction networks and a Stacking ensemble classifier, and trains the initial recognition model using the training sample set to obtain a recognition model, which effectively improves the recognition accuracy of the trained recognition model in identifying disease types of children's body parts.

[0079] Continue to refer to Figure 4 The diagram illustrates a flow 400 of an embodiment of an image recognition method according to this application. The image recognition method includes the following steps:

[0080] Step 401: Obtain a first type of image including body parts.

[0081] In this embodiment, the executing entity can acquire a first type of image containing body parts via a wired or wireless connection.

[0082] Here, body parts can include the lungs, heart, brain, etc.

[0083] The wireless connection methods may include, but are not limited to, 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other currently known or future wireless connection methods.

[0084] Step 402: Input the first type of image into the recognition model to obtain the disease image feature category corresponding to the body part.

[0085] In this embodiment, after acquiring a first type of image including body parts, the executing entity can input the first type of image into a recognition model to obtain the disease image feature category corresponding to the body parts. The recognition model is the recognition model obtained by the method described in the embodiment corresponding to Figure 2, and will not be elaborated further here.

[0086] Specifically, the implementing entity can acquire images of children's lungs using wired or wireless methods, input the images into the recognition model, and obtain the disease image feature categories corresponding to the children's lungs.

[0087] The image recognition method provided in this disclosure improves the recognition accuracy of disease image feature categories for images of children's body parts by acquiring an image to be recognized that includes a child's body parts and inputting the image to be recognized into a recognition model.

[0088] Further reference Figure 5 As an implementation of the methods shown in the above figures, this application provides an embodiment of a model training device, which corresponds to the method embodiment shown in Figure 2, and the device can be applied to various electronic devices.

[0089] like Figure 5 As shown, the model training device 500 in this embodiment includes: an acquisition module 501, a construction module 502, and a training module 503.

[0090] The acquisition module 501 can be configured to acquire a training sample set.

[0091] Module 502 can be configured to build an initial recognition model based on at least two different feature extraction networks and a preset ensemble learning classifier.

[0092] The training module 503 can be configured to train the initial recognition model using the training sample set to obtain the recognition model.

[0093] In some alternative embodiments of this example, the building module is further configured to: build an initial recognition model based on at least three different feature extraction networks and a preset ensemble learning classifier.

[0094] In some optional embodiments of this example, the acquisition module is further configured to acquire an initial training sample set; perform preprocessing operations on the initial training sample set; and determine the initial training sample set after performing the preprocessing operations as the training sample set.

[0095] In some optional embodiments of this example, the initialization parameters of the recognition network during training are determined by using model parameters trained on a preset image dataset as the initialization parameters of the recognition network.

[0096] In some optional embodiments of this example, the preset ensemble learning classifier is the Stacking ensemble classifier.

[0097] Further reference Figure 6 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of an image recognition device, which is similar to... Figure 4 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0098] like Figure 6 As shown, the image recognition device 600 of this embodiment includes: an acquisition module 601 and a recognition module 602.

[0099] The acquisition module 601 can be configured to acquire a first type of image including body parts.

[0100] The recognition module 602 can be configured to input the first type of image into the recognition model to obtain the disease image feature category corresponding to the body part.

[0101] According to embodiments of this application, this application also provides an electronic device and a readable storage medium.

[0102] like Figure 7 The diagram shown is a block diagram of an electronic device for a model training method according to an embodiment of this application.

[0103] 700 is a block diagram of an electronic device for a model training method according to embodiments of this application. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present application described and / or claimed herein.

[0104] like Figure 7As shown, the electronic device includes one or more processors 701, a memory 702, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components are interconnected via different buses and can be mounted on a common motherboard or otherwise as required. The processors can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple electronic devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 7 Take the 701 processor as an example.

[0105] The memory 702 is the non-transitory computer-readable storage medium provided in this application. The memory stores instructions executable by at least one processor to cause the at least one processor to perform the model training method provided in this application. The non-transitory computer-readable storage medium of this application stores computer instructions for causing a computer to perform the model training method provided in this application.

[0106] Memory 702, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the model training method in the embodiments of this application (e.g., appendix). Figure 5 The acquisition module 501, construction module 502, and training module 503 are shown. The processor 701 executes various functional applications and data processing of the server by running non-transient software programs, instructions, and modules stored in the memory 702, thereby implementing the model training method in the above method embodiments.

[0107] Memory 702 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the use of the electronic device for model training. Furthermore, memory 702 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory 702 may optionally include memory remotely located relative to processor 701, and these remote memories can be connected to the electronic device for model training via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0108] The electronic device for the model training method may further include an input device 703 and an output device 704. The processor 701, memory 702, input device 703, and output device 704 can be connected via a bus or other means. Figure 7 Taking the example of a connection between China and Israel via a bus.

[0109] Input device 703 can receive input numerical or character information, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 704 may include a display device, auxiliary lighting device (e.g., LED), and haptic feedback device (e.g., vibration motor). The display device may include, but is not limited to, a liquid crystal display (LCD), a light-emitting diode (LED) display, and a plasma display. In some embodiments, the display device may be a touchscreen.

[0110] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, application-specific integrated circuits (ASICs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0111] These computational programs (also referred to as programs, software, software applications, or code) include machine instructions for a programmable processor and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms “machine-readable medium” and “computer-readable medium” refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) used to provide machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term “machine-readable signal” refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0112] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0113] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0114] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0115] According to the technical solution of the embodiments of this application, the recognition accuracy of the trained recognition model for the first type of image body part disease image feature category is effectively improved.

[0116] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, and this is not limited herein.

[0117] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A model training method, the method comprising: Obtain a training sample set, in which the training samples are first-class images labeled with disease image feature categories; An initial recognition model is constructed based on at least two different feature extraction networks and a preset ensemble learning classifier. The number of the at least two different feature extraction networks is the same as the number of base classifiers in the preset ensemble learning classifier. One feature extraction network corresponds to one base classifier, and the output of the feature extraction network is the input of its corresponding base classifier. The feature extraction network is determined based on a recognition network, which is trained on a second type of images labeled with disease image feature categories. The first and second types of images include the same body parts; the first type of images are images of children's body parts, and the second type of images are images of adults' body parts. The number of images in the second type is greater than the number of images in the first type. The preset ensemble learning classifier also includes a meta-classifier, whose input is the output of multiple base classifiers. The initialization parameters of the recognition network are model parameters trained using a medical image dataset. The initial recognition model is trained using the training sample set to obtain the recognition model.

2. The method according to claim 1, wherein, The initial recognition model is constructed based on at least two different feature extraction networks and a pre-defined ensemble learning classifier, including: An initial recognition model is constructed based on at least three different feature extraction networks and a pre-defined ensemble learning classifier.

3. The method according to claim 1, wherein, The acquisition of the training sample set includes: Obtain the initial training sample set; Preprocessing operations are performed on the initial training sample set, and the initial training sample set after preprocessing operations is determined as the training sample set. The preprocessing operations include data augmentation and data standardization.

4. The method according to any one of claims 1-3, wherein, The preset ensemble learning classifier is the Stacking ensemble classifier.

5. An image recognition method, the method comprising: Acquire first-class images including body parts; The first type of image is input into the recognition model to obtain the disease image feature category corresponding to the body part, wherein the recognition model is the recognition model obtained by the method described in any one of claims 1-4.

6. A model training device, comprising: The acquisition module is configured to acquire a training sample set, wherein the training samples are first-class images labeled with disease image feature categories; The building module is configured to construct an initial recognition model based on at least two different feature extraction networks and a preset ensemble learning classifier. The number of the at least two different feature extraction networks is the same as the number of base classifiers in the preset ensemble learning classifier. One feature extraction network corresponds to one base classifier, and the output of the feature extraction network is the input of its corresponding base classifier. The feature extraction network is determined based on a recognition network, which is trained on a second type of images labeled with disease image feature categories. The first and second types of images include the same body parts; the first type of images are images of children's body parts, and the second type of images are images of adults' body parts. The number of images in the second type is greater than the number of images in the first type. The preset ensemble learning classifier also includes a meta-classifier, whose input is the output of multiple base classifiers. The initialization parameters of the recognition network are model parameters trained using a medical image dataset. The training module is configured to train the initial recognition model using the training sample set to obtain the recognition model.

7. The apparatus according to claim 6, wherein, The building module is further configured to: An initial recognition model is constructed based on at least three different feature extraction networks and a pre-defined ensemble learning classifier.

8. The apparatus according to claim 6, wherein, The acquisition module is further configured to: Obtain the initial training sample set; Preprocessing operations are performed on the initial training sample set, and the initial training sample set after preprocessing operations is determined as the training sample set. The preprocessing operations include data augmentation and data standardization.

9. The apparatus according to any one of claims 6-8, wherein, The preset ensemble learning classifier is the Stacking ensemble classifier.

10. An image recognition device, comprising: The acquisition module is configured to acquire a first-class image including body parts; The recognition module is configured to input the first type of image into the recognition model to obtain the disease image feature category corresponding to the body part, wherein the recognition model is the recognition model obtained by the method as described in any one of claims 1-4.

11. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores information that can be executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-5.

12. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

Images