Method and device for identifying kidney deficiency syndrome of female at childbearing age based on deep learning lingual surface fusion

Abstract

The invention discloses a deep learning lingual surface fusion-based method and device for identifying kidney-deficiency syndromes of women in the childbearing age. The method comprises the following steps: inputting collected tongue image and surface image data into a preset kidney-deficiency syndrome identification model of the women in the childbearing age; the kidney-deficiency syndrome identification model for the women in the childbearing period outputs an identification result; the kidney-deficiency syndrome identification model for the women in the childbearing period is obtained by training the determined kidney-deficiency syndrome and non-kidney-deficiency syndrome data as training data of a deep learning network. According to the method for identifying the kidney deficiency syndrome of the female at the childbearing age based on the deep learning lingual-surface fusion, the syndrome identification model of the kidney deficiency syndrome of the female at the childbearing age is constructed by utilizing two kinds of modal information of the tongue image and the surface image and utilizing a leading-edge deep learning technology, so that objective syndrome differentiation can be carried out by utilizing the collected observation information of the tongue image and the surface image, and the identification accuracy of the kidney deficiency syndrome of the female at the childbearing age is improved. Diagnosis and treatment decisions of clinicians are assisted.

Description

Technical Field

[0001] The present invention relates to deep learning technology, and in particular to a method and device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue and face fusion. Background Art

[0002] Traditional Chinese medicine plays a huge role in regulating women's reproductive health and preventing and treating female reproductive disorders. One of the important characteristics of TCM diagnosis and treatment is to distinguish "syndrome" and treat. "Syndrome" is a pathological summary of the location and nature of the disease at a certain (current) stage in the disease process. It is the state of the body's response to pathogenic factors and the conclusion of the current nature of the disease. Identifying a patient's "syndrome" is the prerequisite for TCM doctors to prescribe effective prescriptions. Therefore, how to distinguish the right syndrome and realize the standardization of syndrome differentiation is the key to improving the efficacy of TCM. However, the essential concept of "syndrome" is still vague. It has the characteristics of internal reality and external emptiness, dynamic time and space, and multi-dimensional interface, which makes the learning and objectification of syndrome differentiation very difficult.

[0003] Medical workers who have not received systematic TCM training are prone to difficulties in syndrome identification and may prescribe inappropriate TCM prescriptions, which may affect the efficacy of the treatment and even aggravate the condition. In addition, since traditional TCM syndrome differentiation relies on the subjective experience and experience of doctors, even rigorously trained TCM doctors have differences in syndrome judgment, resulting in uneven treatment effects for patients. Therefore, we urgently need to explore methods for objective syndrome differentiation.

[0004] At present, the two major research directions around this scientific issue are the development of syndrome diagnosis scales and the objectification of tongue diagnosis. In clinical research, researchers often formulate syndrome score tables based on gynecological textbooks and guidelines for efficacy evaluation. "Nanjing" says "to know by looking is called divine", and inspection is the primary link in TCM syndrome differentiation, including whole-body inspection and local inspection. Among them, the observation of the tongue is a unique and critical content in inspection. During the development of the disease, the tongue changes rapidly and obviously, which can more objectively reflect the state of the disease and is an important basis for syndrome identification. Compared with the patient's description of his own disease experience, the visual information of the patient's tongue image is more objective. However, the main body of traditional tongue image identification is the doctor, and the identification process is affected by the light source in different diagnosis and treatment scenarios. Tongue diagnosis still inevitably has the limitation of great subjectivity.

[0005] Thanks to the development of information technology, research on TCM gynecological tongue diagnosis has made certain progress in terms of acquisition equipment, light source conditions, information preprocessing, tongue image recognition and analysis, and has promoted the objectification of tongue diagnosis. In recent years, artificial intelligence has flourished, and tongue diagnosis has gained new research tools, among which the application of machine learning and deep learning strategies is the most prominent.

[0006] In previous studies on tongue diagnosis in TCM gynecology, scholars have quantitatively analyzed the characteristics of tongue images from the perspective of specific gynecological diseases, such as the lightness, chromaticity, brightness, saturation, thickness and length of tongue quality and tongue coating, described the characteristics of tongue images in different diseases, and compared the differences in tongue images between the disease group and the control group. This type of research has described the objective attributes of gynecological tongue images to a certain extent, but has not expanded to the task field of syndrome identification. In addition, TCM diagnosis emphasizes overall examination, and the four examinations (observation, auscultation, questioning and palpation) are combined to make a diagnosis based on comprehensive diagnostic information from various aspects. It is difficult to meet the needs of visual diagnosis by simply relying on tongue diagnosis information. However, previously, tongue image and facial image information were mostly separated into two types of data for research, and the fusion of multimodal information is another challenge. TCM emphasizes the four examinations. Local visual diagnosis includes observation of the head and face, five senses (eyes, ears, mouth, nose, tongue), neck, body, limbs, skin, etc., while facial diagnosis and tongue diagnosis are more important contents in visual diagnosis, and they contribute greatly to syndrome identification in clinical practice. "What is inside will be reflected outside." The tongue and face are connected to the internal organs through the meridians, and they can reflect the spirit of the human body. The color, shape, texture and other characteristics of the tongue and face can be used to understand the emptiness and fullness of the internal organs, the smoothness and obstruction of the meridians, the rise and fall of qi and blood, and the surplus and deficiency of body fluids. Inspection is the first of the four diagnoses, and it has little burden on patients. At a time when people are increasingly relying on sampling, laboratory tests, imaging examinations, and diagnostic surgery, it has irreplaceable value.

[0007] Although artificial intelligence has been used in the diagnosis and identification of many diseases, there is currently no syndrome identification model in the field of gynecology or female reproduction. Summary of the invention

[0008] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and to provide a method and device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue and face fusion, so as to assist clinicians in making diagnosis and treatment decisions.

[0009] To achieve the above object, the technical solution of the present invention is:

[0010] In a first aspect, the present invention provides a method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-surface fusion, comprising:

[0011] Input the collected tongue and facial image data into a preset kidney deficiency syndrome identification model for women of childbearing age;

[0012] The kidney deficiency syndrome identification model for women of childbearing age outputs identification results;

[0013] The kidney deficiency syndrome identification model for women of childbearing age is obtained by training the determined kidney deficiency syndrome and non-kidney deficiency syndrome data as training data for the deep learning network.

[0014] Optionally, the tongue image and facial image data are collected in the following manner:

[0015] The patient's tongue and facial image data are collected through a fixed tongue and facial image acquisition instrument, and the image data that does not meet the quality requirements are filtered out.

[0016] Optionally, the collected tongue and facial image data are pre-processed before being input into the kidney deficiency syndrome identification model for women of childbearing age, and the pre-processing includes:

[0017] The tongue area and the face area with corresponding pixel values ​​are segmented from the original image to obtain an image containing only the tongue or the face, and the image containing only the tongue or the face is adjusted to a uniform size.

[0018] Optionally, the kidney deficiency syndrome identification model for women of childbearing age outputs the identification result in the following manner:

[0019] A parallel convolutional neural network is used to extract the deep features of the tongue image and the facial image from the input tongue image and facial image data;

[0020] Use deep feature network to fuse tongue image and facial image deep features;

[0021] Kidney deficiency syndrome or non-kidney deficiency syndrome is output based on the fusion result.

[0022] Optionally, the convolutional neural network and the deep feature network form a deep learning framework, and the total loss Ltotal of the deep learning framework is:

[0023] Ltotal=Lcon+λ(Ltongue+Lface)

[0024] λ is a coefficient of balancing weights. Lcon, Ltongue, and Lface use the cross entropy loss function, which corresponds to the loss concatenated with the deep features of the tongue image, the loss of the deep features of the tongue image, and the loss of the deep features of the face image, respectively.

[0025] In a second aspect, the present invention provides a device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-surface fusion, comprising:

[0026] A model module, wherein the model module stores a pre-trained model for identifying kidney deficiency syndrome in women of childbearing age, wherein the model for identifying kidney deficiency syndrome in women of childbearing age is trained based on the determined kidney deficiency syndrome and non-kidney deficiency syndrome data as training data for a deep learning network;

[0027] An input module is used to input the collected tongue image and facial image data into the model module as input of the kidney deficiency syndrome identification model for women of childbearing age, and the kidney deficiency syndrome identification model for women of childbearing age outputs the identification result.

[0028] In a third aspect, the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0029] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion is constructed by utilizing tongue image and facial image information and cutting-edge deep learning technology to construct a syndrome recognition model for kidney deficiency syndrome in women of childbearing age. The collected tongue image and facial image information can be used for objective syndrome differentiation to assist clinicians in making diagnosis and treatment decisions. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 A flow chart of a method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion provided in an embodiment of the present application; Figure 2 It is a schematic diagram of tongue and facial image data collection; Figure 3 A flow chart for outputting recognition results for the kidney deficiency syndrome recognition model for women of childbearing age; Figure 4 Schematic diagram of the deep residual network structure; Figure 5 This is a graph showing the performance comparison of four different methods in identifying kidney deficiency syndrome in women of childbearing age; Figure 6 This is a schematic diagram of the composition of a device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue and face fusion provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0033] Example 1

[0034] In the past, syndrome recognition technology in the field of female reproduction had the following shortcomings: ① Only one modality of data was used, mostly tongue images; ② Technical research on tongue images did not rise to the task of syndrome recognition, but remained at a simple description of the attributes of tongue images; ③ Artificial intelligence has been applied in the diagnosis and identification of many diseases and achieved considerable efficiency, but there is currently no syndrome recognition model in the field of gynecology or female reproduction.

[0035] To this end, this embodiment provides a method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion, which mainly includes the following steps:

[0036] 101. Input the collected tongue image and facial image data into a preset kidney deficiency syndrome identification model for women of childbearing age;

[0037] 102. The kidney deficiency syndrome identification model for women of childbearing age outputs an identification result;

[0038] The kidney deficiency syndrome identification model for women of childbearing age is obtained by training the determined kidney deficiency syndrome and non-kidney deficiency syndrome data as training data for the deep learning network.

[0039] It can be seen that this method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue and face fusion, by utilizing the two modal information of tongue image and facial image, uses cutting-edge deep learning technology to construct a syndrome recognition model for kidney deficiency syndrome in women of childbearing age. Therefore, the collected tongue image and facial image information can be used for objective syndrome differentiation, assisting clinicians in making diagnosis and treatment decisions.

[0040] In a specific embodiment, if Figure 2 As shown, the tongue and face image data are collected in the following manner: the tongue and face images of the patient are collected in a standardized manner by a fixed tongue and face image collection instrument, and low-quality images with stained fur, tongue shaking, etc. are filtered.

[0041] In a specific embodiment, the determined kidney deficiency syndrome and non-kidney deficiency syndrome data are obtained in the following manner: experienced clinicians determine the patient's kidney deficiency syndrome and non-kidney deficiency syndrome as data for deep learning network training. The diagnosis of syndromes is based on the "Chinese Medicine Clinical Diagnosis and Treatment Terminology Part 2: Syndromes" (GB / T 16751.2-2021), "Chinese Medicine Gynecology Common Disease Diagnosis and Treatment Guidelines", "Chinese Medicine Gynecology" textbooks and "Chinese Medicine Diagnosis" textbooks. The model for identifying kidney deficiency syndrome in women of childbearing age is implemented using "Pytorch" (https: / / pytorch.org / ) and trained using a GeForce RTX 3090Ti (24GB video memory) graphics card. The Adamw optimization algorithm is used to minimize the objective function, and the torchvision built-in function is used for data enhancement. The initial learning rate is set to 2e-4, the batchsize is set to, and to alleviate the problem of overfitting, the L2 regularization technique is used, and the weight_decay is set to 1e-4. In addition, an early stopping training strategy is adopted, and it is set that if the accuracy of the validation set does not increase for 15 consecutive epochs, the training will be stopped.

[0042] In a preferred embodiment, the collected tongue and facial image data are pre-processed before being input into the kidney deficiency syndrome identification model for women of childbearing age, so as to facilitate the fusion of features of images of different scales. Specifically, the pre-processing includes:

[0043] First, the tongue area and face area with corresponding pixel values ​​outlined by labelme software (http: / / labelme.csail.mit.edu / Release3.0 / ) were segmented from the original image to obtain images containing only the tongue or face; next, these images were resized to a uniform size of 224*224; in order to eliminate the influence of different feature scales and make the model converge better, the data was normalized.

[0044] In a specific embodiment, if Figure 3 As shown, the kidney deficiency syndrome identification model for women of childbearing age outputs the identification results in the following way:

[0045] The input tongue image and facial image data are respectively extracted using a parallel convolutional neural network (CNN) to extract the deep features of the tongue image and facial image. Figure 4 As shown, the specific structure is a deep residual network (ResNet)

[0046] Use the deep feature network (DSN) to fuse the two deep features of tongue image and facial image;

[0047] Kidney deficiency syndrome or non-kidney deficiency syndrome is output based on the fusion result.

[0048] The DSN fusion framework proposed in the present invention has a cross-entropy loss function Lcoon, which corresponds to the splicing of the deep features of the tongue image and the face image for classification. In addition, the process of CNN extracting deep features from the tongue and face image for classification can also be regarded as a loss function that needs to be supervised in the network. Therefore, the total loss Ltotal of the proposed deep learning framework can be defined as Ltotal=Lcon+λ(Ltongue+Lface), where λ is a coefficient of a balancing weight, initially set to 1.0. Lcon, Ltongne, and Lface use a cross-entropy loss function, which corresponds to the loss of the splicing of the deep features of the tongue and face image, the loss of the deep features of the tongue image, and the loss of the deep features of the face image. In this way, the tongue image and the face image can be deeply fused, so that the syndrome corresponding to the tongue image and the face image can be accurately identified as kidney deficiency syndrome or non-kidney deficiency syndrome based on the input tongue image and face image data.

[0049] In a specific embodiment, the performance of the kidney deficiency syndrome identification model for women of childbearing age in identifying kidney deficiency syndrome is evaluated by the following five indicators: (1) accuracy, (2) precision, (3) recall, (4) F1 score, (5) receiver operating characteristic (ROC) and area under the curve (AUC).

[0050]

[0051] Accuracy is the ratio of the sum of correctly classified positive and negative examples in all samples. Precision is the ratio of positive examples correctly classified by the model to all positive examples predicted by the model. Recall is the ratio of positive examples correctly predicted by the model to all positive samples. The F1 score is a balanced index used to measure the accuracy of a classification model. It takes into account both the precision and recall of the model and can be regarded as the harmonic mean of the precision and recall of the model. The ROC curve combines sensitivity and specificity, and through graphical methods, it can accurately reflect the relationship between the specificity and sensitivity of the analytical method. It is a comprehensive representation of test accuracy. AUC is defined as the area under the ROC curve. As a numerical value, a classifier with a larger AUC value is better.

[0052] In order to verify the performance of the method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue and face fusion provided in this embodiment, a total of four methods were tested to identify kidney deficiency syndrome, including using tongue image features (Tongue) alone, using face image features (Face) alone, simply concatenating the deep features of tongue and face images (Concat), and classification based on the method (DSN) for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue and face fusion. Figure 5 The performance comparison results of the above four different methods in identifying kidney deficiency syndrome in women of childbearing age are shown.

[0053] The results show that the method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue and face fusion has good performance in identifying kidney deficiency syndrome in women of childbearing age.

[0054] Embodiment 2:

[0055] This embodiment provides a device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion, comprising:

[0056] A model module, wherein the model module stores a pre-trained model for identifying kidney deficiency syndrome in women of childbearing age, wherein the model for identifying kidney deficiency syndrome in women of childbearing age is trained based on the determined kidney deficiency syndrome and non-kidney deficiency syndrome data as training data for a deep learning network;

[0057] An input module is used to input the collected tongue image and facial image data into the model module as input of the kidney deficiency syndrome identification model for women of childbearing age, and the kidney deficiency syndrome identification model for women of childbearing age outputs the identification result.

[0058] It can be seen that this method and device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue and face fusion utilizes two modal information, tongue image and facial image, and uses cutting-edge deep learning technology to construct a syndrome recognition model for kidney deficiency syndrome in women of childbearing age. The collected tongue image and facial image information can then be used for objective syndrome differentiation, assisting clinicians in making diagnosis and treatment decisions.

[0059] In a specific embodiment, the kidney deficiency syndrome identification model for women of childbearing age outputs the identification result in the following manner:

[0060] The input tongue image and facial image data are respectively extracted using a parallel convolutional neural network (CNN) to extract the deep features of the tongue image and facial image.

[0061] Use the deep feature network (DSN) to fuse the two deep features of tongue image and facial image;

[0062] Kidney deficiency syndrome or non-kidney deficiency syndrome is output based on the fusion result.

[0063] The DSN fusion framework proposed in the present invention has a cross-entropy loss function Lcon, which corresponds to the splicing of the deep features of the tongue image and the face image for classification. In addition, the process of CNN extracting deep features from the tongue and face image for classification can also be regarded as a loss function that needs to be supervised in the network. Therefore, the total loss Ltotal of the proposed deep learning framework can be defined as Ltotal=Lcon+λ(Ltongue+Lface), where λ is a coefficient of balancing weights, initially set to 1.0. Lcon, Ltongne, and Lface use cross-entropy loss functions, which correspond to the loss of splicing with the deep features of the tongue and face image, the loss of the deep features of the tongue image, and the loss of the deep features of the face image.

[0064] Embodiment 3:

[0065] See also Figure 3 As shown, the electronic device provided in this embodiment includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a program for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion. When the processor executes the computer program, the steps of the above embodiment 1 are implemented, such as Figure 1 Steps shown.

[0066] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the electronic device.

[0067] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0068] The memory may be an internal storage element of the electronic device, such as a hard disk or memory of the electronic device. The memory may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the electronic device. Further, the memory may also include both an internal storage unit of the electronic device and an external storage device. The memory is used to store the computer program and other programs and data required by the electronic device. The memory may also be used to temporarily store data that has been output or is to be output.

[0069] Embodiment 4:

[0070] This embodiment provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method described in Embodiment 1 are implemented.

[0071] The computer-readable medium shown can be any device that can contain, store, communicate, propagate or transmit a program for use with an instruction execution system, device or apparatus or in conjunction with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and editable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable media on which the program can be printed, such as by optically scanning the paper or other medium, then editing, interpreting or processing in other suitable ways as necessary to obtain the program electronically, and then storing it in a computer memory.

[0072] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made based on the essence of the content of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion, It is characterized in that include: Input the collected tongue and facial image data into a preset kidney deficiency syndrome identification model for women of childbearing age; The kidney deficiency syndrome identification model for women of childbearing age outputs identification results; The kidney deficiency syndrome identification model for women of childbearing age is obtained by training the determined kidney deficiency syndrome and non-kidney deficiency syndrome data as training data for the deep learning network.

2. The method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion as claimed in claim 1, wherein the tongue image and face image data are collected in the following manner: The patient's tongue and facial image data are collected through a fixed tongue and facial image acquisition instrument, and the image data that does not meet the quality requirements are filtered out.

3. The method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion as claimed in claim 1, Its special feature is that The collected tongue and facial image data are pre-processed before being input into the kidney deficiency syndrome identification model for women of childbearing age. The pre-processing includes: The tongue area and the face area with corresponding pixel values ​​are segmented from the original image to obtain an image containing only the tongue or the face, and the image containing only the tongue or the face is adjusted to a uniform size.

4. The method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-surface fusion as claimed in claim 1, wherein the kidney deficiency syndrome identification model for women of childbearing age outputs the recognition result in the following manner: A parallel convolutional neural network is used to extract the deep features of the tongue image and the facial image from the input tongue image and facial image data; Use deep feature network to fuse tongue image and facial image deep features; Kidney deficiency syndrome or non-kidney deficiency syndrome is output based on the fusion result.

5. The method for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion as claimed in claim 4, It is characterized in that The convolutional neural network and the deep feature network form a deep learning framework, and the total loss Ltotal of the deep learning framework is: Ltotal=Lcon+λ(Ltongue+Lface) λ is a coefficient of balancing weights. Lcon, Ltongue, and Lface use the cross entropy loss function, which corresponds to the loss concatenated with the deep features of the tongue image, the loss of the deep features of the tongue image, and the loss of the deep features of the face image, respectively.

6. A device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion, It is characterized in that include: A model module, wherein the model module stores a pre-trained model for identifying kidney deficiency syndrome in women of childbearing age, wherein the model for identifying kidney deficiency syndrome in women of childbearing age is trained based on the determined kidney deficiency syndrome and non-kidney deficiency syndrome data as training data for a deep learning network; An input module is used to input the collected tongue image and facial image data into the model module as input of the kidney deficiency syndrome identification model for women of childbearing age, and the kidney deficiency syndrome identification model for women of childbearing age outputs the identification result.

7. The device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion as claimed in claim 6, It is characterized in that The kidney deficiency syndrome identification model for women of childbearing age outputs the identification results in the following manner: A parallel convolutional neural network is used to extract the deep features of the tongue image and the facial image from the input tongue image and facial image data; Use deep feature network to fuse tongue image and facial image deep features; Kidney deficiency syndrome or non-kidney deficiency syndrome is output based on the fusion result.

8. The device for identifying kidney deficiency syndrome in women of childbearing age based on deep learning tongue-face fusion as claimed in claim 7, It is characterized in that The convolutional neural network and the deep feature network form a deep learning framework, and the total loss Ltotal of the deep learning framework is: Ltotal=Lcon+λ(Ltongue+Lface) λ is a coefficient of balancing weights. Lcon, Ltongue, and Lface use the cross entropy loss function, which corresponds to the loss concatenated with the deep features of the tongue image, the loss of the deep features of the tongue image, and the loss of the deep features of the face image, respectively.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images