System for rhinitis identification
By using sensors and cameras to collect information through a rhinitis recognition system, and combining it with a rhinitis recognition model and initial convolutional neural network training, rapid and accurate rhinitis recognition and treatment are achieved, solving the problem of time-consuming and expensive hospital visits and providing a convenient and economical solution.
Patent Information
- Application Number
- CN202510938932.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-08
- Publication Date
- 2025-11-28
AI Technical Summary
Current technologies for identifying rhinitis require a hospital visit, which is time-consuming and expensive, and is especially inconvenient for people with mild symptoms or those who are short on time.
Design a rhinitis recognition system, including a rhinitis treatment device and a recognition module. The system uses multiple sensors and cameras to collect vital signs information and nasal cavity images, and performs recognition through a pre-established rhinitis recognition model. It is then trained by combining an initial convolutional neural network model to extract and stitch features together to improve recognition accuracy.
It can quickly identify rhinitis without going to the hospital, improving convenience and reducing costs, while providing additional therapeutic effects through water ion therapy.
Smart Images
Figure CN121015166A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of rhinitis identification devices, in particular to a system for rhinitis identification. BACKGROUND
[0002] At present, rhinitis identification is only achieved by going to a hospital for treatment.
[0003] Hospital treatment involves processes such as registration, queuing, examination, and report collection. In particular, in a third-grade class-A hospital, a single visit may take several hours, which is very inconvenient for people with mild symptoms or in a hurry, such as students and office workers, and the examination fee is high. SUMMARY
[0004] The present application provides a system for rhinitis identification, which can identify rhinitis, improve convenience, and reduce costs. The specific technical solutions are as follows.
[0005] In a first aspect, the present application provides a system for rhinitis identification, comprising:
[0006] a rhinitis treatment instrument, which comprises a plurality of sensors, a head ring, a mask support arranged below the head ring, a magnetic mask arranged below the mask support and a camera arranged below the nose beam;
[0007] an information acquisition module for acquiring the sign information collected by each sensor when the user uses the rhinitis treatment instrument and the nasal cavity image captured by the camera;
[0008] an identification module for identifying the nasal cavity image and the sign information according to a pre-established rhinitis identification model to obtain an identification result;
[0009] a rhinitis identification model establishment module, comprising:
[0010] an acquisition unit for acquiring each sample data in a training set, wherein the sample data comprises the nasal cavity image and sign information of a rhinitis patient and the corresponding disease category, and the nasal cavity image and sign information of a patient with a similar disease to rhinitis and the corresponding disease category;
[0011] a training unit configured to extract first image features of the rhinitis patient's nasal cavity images and second image features of the rhinitis similar disease patient's nasal cavity images respectively by the initial convolutional neural network model, splice the first image features with the rhinitis patient's physical sign information to obtain first spliced features, splice the second image features with the rhinitis similar disease patient's physical sign information to obtain second spliced features, classify the first spliced features and the second spliced features respectively to obtain first initial disease categories and second initial disease categories, calculate a first difference value between the first initial disease categories and the rhinitis patient's corresponding disease categories and a second difference value between the second initial disease categories and the rhinitis similar disease patient's corresponding disease categories according to a loss function respectively;
[0012] a training completion unit configured to adjust parameters of the initial convolutional neural network model based on the first difference value and the second difference value, and obtain a rhinitis recognition model when a training completion condition is met.
[0013] Optionally, the training completion unit is specifically configured to:
[0014] a training completion sub-unit configured to complete the training when the number of iterations reaches a preset number, and take the current initial convolutional neural network model as a rhinitis recognition model used to associate nasal cavity images and physical sign information with corresponding disease categories and output confidence of the disease categories.
[0015] Optionally, the loss function is:
[0016] L = λ · L1 + (1 - λ) · L2 + βL3
[0017]
[0018] L3 = max(0, log f Img -log f1)
[0019] wherein, L is a loss function, λ is a weight parameter of a classification loss function, used to adjust the proportion of the classification loss function and an authentication loss function in the loss, β is a weight parameter of a spliced feature effectiveness loss function, L1 is the classification loss function, L2 is the authentication loss function, used to calculate whether two features are features of the same patient, is a network output feature of the i-th sample data of the m-th patient, is a network output feature of the j-th sample data of the m-th patient, L3 is the spliced feature effectiveness loss function, f ImgIn order to identify the image feature as the posterior probability of rhinitis through the preset network model, f1 is the posterior probability of the feature after splicing being identified as rhinitis, and the preset network model is used to correlate the nasal cavity image with rhinitis.
[0020] Optionally, the confidence of the rhinitis identification model is:
[0021] f = 1-f com
[0022]
[0023] Wherein, f is the confidence of the rhinitis identification model, f com is the confidence of the competitive category, γ is the weight parameter of the confidence, f2 is the posterior probability of the current sample data being identified as rhinitis, f3 is the posterior probability of the current sample data being identified as the largest in other disease categories, m1 is the number of sample data in the training set being identified as rhinitis, and m2 is the number of sample data in the training set being misidentified as rhinitis as the disease category corresponding to f3.
[0024] Optionally, the inside of the head ring is provided with a displacement sensor, the head ring is provided with a body temperature sensor at a position close to the forehead, the nasal beam is provided with a nasal cavity gas flow rate sensor on one side close to the nose, the magnetic mask is provided with an oral cavity gas flow rate sensor on one side close to the mouth, and the mask support is provided with a nitric oxide sensor and a gas component sensor on the two inner side walls close to the two cheeks respectively.
[0025] Optionally, the sign information at least includes displacement information, nasal cavity gas flow rate information, oral cavity gas flow rate information, body temperature information, nitric oxide information and gas component information.
[0026] Optionally, the head ring is provided with an air inlet, the inside of the head ring is provided with a water ion module, an air inlet device and an air inlet channel;
[0027] The inside of the mask support is provided with an air outlet channel connected with the air inlet channel, the nasal beam is provided with a first air outlet below, and the mask support is provided with a second air outlet on the two outer side walls close to the two cheeks.
[0028] Optionally, the head ring includes a head ring main body, a rear hatch and two side hatches;
[0029] The rear hatch covers the rear side of the head ring main body to form a rear compartment, the two side hatches cover the left and right sides of the head ring main body respectively to form two side compartments, the rear compartment and the two side compartments form the air inlet channel, and the rear hatch is provided with the air inlet;
[0030] A filter screen is arranged between the rear bin and each side bin, and the water ion module and the air inlet device are arranged inside each side bin, and the displacement sensor is arranged inside one of the two side bins.
[0031] Optionally, the rear bin is further provided with a numerical control mainboard.
[0032] The displacement sensor, the nasal cavity gas flow rate sensor, the oral cavity gas flow rate sensor, the body temperature sensor, the nitric oxide sensor, the gas component sensor and the camera are in communication connection with the numerical control mainboard.
[0033] Optionally, the rear bin is further provided with a battery module.
[0034] The displacement sensor, the nasal cavity gas flow rate sensor, the oral cavity gas flow rate sensor, the body temperature sensor, the nitric oxide sensor, the gas component sensor and the camera are in electrical connection with the battery module.
[0035] From the above, the system for rhinitis identification provided by the embodiment of the application comprises a rhinitis treatment instrument, an information acquisition module and an identification module, the rhinitis treatment instrument comprises a plurality of sensors, a head ring, a mask support arranged below the head ring, a magnetic mask arranged below a nose beam and connected to the mask support by magnetic attraction, and a camera arranged below the nose beam; the information acquisition module is used to acquire the sign information collected by each sensor and the nasal cavity image shot by the camera when the user uses the rhinitis treatment instrument; the identification module is used to identify the nasal cavity image and the sign information according to a pre-established rhinitis identification model to obtain an identification result; and the rhinitis identification model establishment module comprises: an acquisition unit used to acquire each sample data in a training set, wherein the sample data comprises the nasal cavity image and the sign information of a rhinitis patient and the corresponding disease category, and the nasal cavity image and the sign information of a patient with a similar disease to rhinitis and the corresponding disease category; a training unit used to extract a first image feature of the nasal cavity image of the rhinitis patient and a second image feature of the nasal cavity image of the patient with the similar disease to rhinitis by an initial convolutional neural network model respectively, splice the first image feature with the sign information of the rhinitis patient to obtain a first spliced feature, splice the second image feature with the sign information of the patient with the similar disease to rhinitis to obtain a second spliced feature, classify the first spliced feature and the second spliced feature respectively to obtain a first initial disease category and a second initial disease category, and calculate a first difference value between the first initial disease category and the corresponding disease category of the rhinitis patient and a second difference value between the second initial disease category and the corresponding disease category of the patient with the similar disease to rhinitis according to a loss function; and a training completion unit used to adjust the parameters of the initial convolutional neural network model based on the first difference value and the second difference value, and obtain the rhinitis identification model when a training completion condition is met. Thus, the sign information collected by each sensor and the nasal cavity image shot by the camera are obtained by using the rhinitis treatment instrument, then the nasal cavity image and the sign information are identified according to the pre-established rhinitis identification model to obtain an identification result, whether it is rhinitis can be identified, the initial convolutional neural network model is trained by feature extraction and splicing when the rhinitis identification model is established, the extraction of similar disease features by the model can be strengthened, the accuracy of the model is improved, the accuracy of identification as rhinitis is further improved, the user does not need to go to a hospital for treatment, the convenience is improved, and the cost is reduced.
[0036] The innovation points of the embodiment of the application include:
[0037] 1. By using a rhinitis treatment device to obtain vital sign information collected by various sensors and nasal cavity images captured by a camera, the nasal cavity images and vital sign information are then identified according to a pre-established rhinitis recognition model to obtain the recognition result, which can identify whether it is rhinitis. At the same time, the rhinitis recognition model is established by training the initial convolutional neural network model through feature extraction and splicing, which can enhance the model's extraction of features of similar diseases, thereby improving the accuracy of the model and further improving the accuracy of identifying rhinitis. There is no need to go to the hospital for treatment, which improves convenience and reduces costs.
[0038] 2. The magnetic mask can be changed at any time via magnetic attachment, allowing multiple people to share it and reducing usage costs.
[0039] 3. By installing a displacement sensor inside the headband, head displacement can be monitored; by installing a body temperature sensor near the forehead, body temperature can be monitored; by installing a nasal airflow velocity sensor on the side of the nose bridge near the nose, the intensity of exhaled airflow can be monitored; by installing a camera below the nose bridge, images of the nasal cavity can be captured; by installing an oral airflow velocity sensor on the side of the magnetic mask near the mouth, the intensity of exhaled gas can be monitored; and by installing a nitric oxide sensor and a gas composition sensor on the two inner walls near the cheeks of the mask holder, the presence and composition of exhaled gas can be monitored.
[0040] 4. Based on the nasal cavity images and physical signs of rhinitis patients and their corresponding disease categories, as well as the nasal cavity images and physical signs of patients with similar diseases and their corresponding disease categories, the initial convolutional neural network model is trained to obtain a rhinitis recognition model. This multi-category model training can enhance the model's extraction of features of similar diseases, thereby improving the recognition accuracy.
[0041] 5. The loss function provided in this embodiment of the invention is a multi-task loss function, including an identification task, an authentication task, and a stitching validity task. The identification task is used to identify the category of rhinitis and similar diseases. The authentication task is used to calculate whether two features belong to the same patient. By adding an authentication task, the training process no longer only focuses on the correctness of the current sample data classification, but also on whether the features belong to the same patient, thereby improving the model's generalization ability, avoiding overlearning, and further avoiding the problem of inaccurate identification caused by overlearning. The stitching validity task is used to determine whether the stitching of image features and vital sign information is effective, and provides regularization constraints to improve the model's generalization ability.
[0042] 6、The confidence of the rhinitis disease category is estimated based on the confidence of the maximum possible misjudgment of the disease category, avoiding directly predicting a confidence that makes the confidence of the rhinitis disease category greater, which has nothing to do with the competing category and cannot make a reasonable prediction by referring to the competing category. The confidence calculated based on the confidence of the competing category is more reliable, can alleviate the problem of over-learning of the model leading to over-high confidence, and can alleviate the estimation difference caused by extremely uneven distribution of category sample data by introducing the training set, a global statistical data, to calculate the confidence of the rhinitis recognition model.
[0043] 7、By setting a water ion module inside the head ring and a first air outlet below the nasal beam, and setting a second air outlet on the two outer walls of the mask support close to the two cheeks, a water ion environment is formed on the oral face through the first air outlet, and the water ions generated by the water ion module are sent to the nose through breathing, so that the rhinitis is treated by water ions, which can achieve the treatment effect. At the same time, the water ions generated by the water ion module will not be offset by charged particles in the air, improving efficiency and safety, and there is no need to go to a medical institution for treatment, reducing costs, without the need to insert the nasal cavity, direct wearing can be used, and it can be moved with the body, convenient to use.
[0044] 8、The two second air outlets of the two outer walls can release water ions to the outside, thereby forming an effective sterilization water ion environment in the vicinity of the human body, using water ion sterilization to improve the ability to inhibit allergens, formaldehyde and other harmful gases, reduce the influence of external air environment problems, and further improve the treatment effect on allergic rhinitis.
[0045] 9、By setting the rear hatch on the rear side of the head ring main body to form a rear compartment, and setting two side hatches on the left and right sides of the head ring main body to form two side compartments, the rear hatch is provided with an air inlet, so that the rear compartment and the two side compartments form an air inlet channel. By setting a filter screen between the rear compartment and each side compartment, suspended particles in the air entering from the air inlet can be filtered, and the water ion generation efficiency is improved.
[0046] 10、By setting a numerical control mainboard, the data collected by the displacement sensor, the nasal cavity gas flow rate sensor, the oral cavity gas flow rate sensor, the body temperature sensor, the nitric oxide sensor, the gas component sensor and the camera can be sent to the numerical control mainboard for processing.
[0047] 11、By setting a battery module in the rear compartment, power is supplied to the displacement sensor, the nasal cavity gas flow rate sensor, the oral cavity gas flow rate sensor, the body temperature sensor, the nitric oxide sensor, the gas component sensor and the camera.
[0048] Of course, practicing any of the products or methods of the present application does not necessarily achieve all of the above advantages. BRIEF DESCRIPTION OF DRAWINGS
[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only represent some of the embodiments of the present application. Those skilled in the art can obtain other drawings according to these drawings without any creative effort.
[0050] Figure 1 The structural schematic diagram of the system for rhinitis identification provided by the embodiment of the present application is shown in the figure.
[0051] Figure 2 The structural schematic diagram of the rhinitis treatment instrument provided by the embodiment of the present application is shown in the figure.
[0052] Figure 1 And Figure 2 In the figure, 1 is a rhinitis treatment instrument, 11 is a head ring, 111 is a head ring main body, 112 is a rear hatch cover, 113 is a side hatch cover, 12 is a mask support, 121 is an outer side wall, 13 is a nose beam, 14 is a magnetic mask, 15 is a displacement sensor, 16 is a body temperature sensor, 17 is a nasal cavity gas flow rate sensor, 18 is an oral cavity gas flow rate sensor, 19 is a nitric oxide sensor, 10 is a gas component sensor, 2 is an information acquisition module, 3 is an identification module, 4 is a rhinitis identification model establishment module, 5 is a water ion module, 6 is an air inlet device, 7 is a filter screen, 8 is a numerical control mainboard, and 9 is a battery module. DETAILED DESCRIPTION
[0053] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort fall within the scope of protection of the present application.
[0054] It should be noted that the terms “include” and “have” and any variations thereof in the embodiments of the present application and the drawings are intended to cover non-exclusive inclusion. For example, the processes, methods, systems, products or devices including a series of steps or units are not limited to the listed steps or units, but can optionally further include steps or units not listed, or can optionally further include other steps or units inherent to these processes, methods, products or devices.
[0055] The embodiment of the application discloses a system for rhinitis identification, which can identify rhinitis, improve convenience and reduce cost.
[0056] Figure 1 A structural schematic diagram of the system for rhinitis identification provided by the embodiment of the application is provided. Figure 2 A structural schematic diagram of the rhinitis treatment instrument provided by the embodiment of the application is provided.
[0057] Referring to Figure 1 The system for rhinitis identification provided by the embodiment of the application specifically comprises a rhinitis treatment instrument 1, an information acquisition module 2, an identification module 3 and a rhinitis identification model establishment module 4.
[0058] Referring to Figure 2 The rhinitis treatment instrument 1 comprises a plurality of sensors, a head ring 11, a mask support 12 arranged below the head ring 11, a magnetic mask 14 magnetically connected below a nose beam 13 of the mask support 12 and a camera arranged below the nose beam 13, wherein the head ring 11 can be a ring-shaped object with an opening at the forehead, the mask support 12 is a semi-open structure and has no obstruction above the nose, which can improve the wearing feeling of the user, and the nose beam 13 and the magnetic mask 14 can be magnetically connected through a magnetic strip. The material of the magnetic mask 14 can be pp plastic, i.e., polypropylene, the camera can be provided with illumination, which facilitates the shooting of the nasal cavity image, and the camera is detachably connected to the nose beam 13 and can be detached when not shooting.
[0059] The magnetic mask 14 can be replaced at any time through the magnetic mode, and the magnetic mask 14 can be used by multiple people together, thereby reducing the use cost.
[0060] Continuously referring to Figure 2 The inside of the head ring 11 is provided with a displacement sensor 15, the position of the head ring 11 close to the forehead is provided with a body temperature sensor 16, and the side of the nose beam 13 close to the nose is provided with a nasal cavity gas flow rate sensor 17.
[0061] The side of the magnetic mask 14 close to the mouth is provided with an oral cavity gas flow rate sensor 18, and the two inner side walls of the mask support 12 close to the two cheeks are respectively provided with a nitric oxide sensor 19 and a gas component sensor 10, wherein the arrangement mode can be fixed connection, for example, screw connection or pasting.
[0062] Thus, the head displacement is monitored by arranging the displacement sensor 15 inside the head ring 11, the body temperature is monitored by arranging the body temperature sensor 16 at the position of the head ring 11 close to the forehead, the intensity of the airflow exhaled from the nasal cavity is monitored by arranging the nasal cavity airflow speed sensor 17 at the side of the nasal bridge 13 close to the nose, the image of the nasal cavity is captured by arranging the camera below the nasal bridge 13, the intensity of the gas exhaled from the oral cavity is monitored by arranging the oral cavity airflow speed sensor 18 at the side of the magnetic mask 14 close to the mouth, and the monitoring of whether the exhaled gas contains nitric oxide and the monitoring of the composition of the exhaled gas are realized by arranging the nitric oxide sensor 19 and the gas composition sensor 10 on the two inner walls of the mask support 12 close to the two cheeks.
[0063] The information acquisition module 2 is used to acquire the sign information collected by each sensor and the image of the nasal cavity captured by the camera when the user uses the rhinitis treatment instrument 1. When the user uses the rhinitis treatment instrument 1, the camera captures the image of the nasal cavity, each sensor starts to work and can collect the sign information, wherein the sign information at least includes displacement information, nasal cavity gas flow rate information, oral cavity gas flow rate information, body temperature information, nitric oxide information and gas composition information.
[0064] The recognition module 3 is used to recognize the image of the nasal cavity and the sign information according to the pre-established rhinitis recognition model to obtain a recognition result. After the image of the nasal cavity and the sign information collected by each sensor and captured by the camera when the user uses the rhinitis treatment instrument 1 are acquired, the image of the nasal cavity and the sign information can be recognized according to the pre-established rhinitis recognition model to obtain a recognition result. The recognition result can be rhinitis or other types of diseases.
[0065] The rhinitis recognition model establishing module 4 includes an acquisition unit, a training unit and a training completion unit.
[0066] The acquisition unit is used to acquire each sample data in the training set, wherein the sample data includes the image of the nasal cavity and the sign information of the rhinitis patient and the corresponding disease category, and the image of the nasal cavity and the sign information of the patient with the disease similar to rhinitis and the corresponding disease category;
[0067] The training unit is configured to extract first image features of the nasal cavity images of the rhinitis patients and second image features of the nasal cavity images of the patients with similar diseases to rhinitis by the initial convolutional neural network model, splice the first image features with the sign information of the rhinitis patients to obtain first spliced features, splice the second image features with the sign information of the patients with similar diseases to rhinitis to obtain second spliced features, classify the first spliced features and the second spliced features to obtain first initial disease categories and second initial disease categories, and calculate a first difference value between the first initial disease categories and the disease categories corresponding to the rhinitis patients and a second difference value between the second initial disease categories and the disease categories corresponding to the patients with similar diseases to rhinitis according to a loss function.
[0068] The training completion unit is configured to adjust parameters of the initial convolutional neural network model based on the first difference value and the second difference value, and obtain the rhinitis recognition model when a training completion condition is met.
[0069] To establish the rhinitis recognition model, sample data in a training set needs to be obtained, where the sample data includes the nasal cavity images and sign information of the rhinitis patients and corresponding disease categories, and the nasal cavity images and sign information of the patients with similar diseases to rhinitis and corresponding disease categories.
[0070] Specifically, the nasal cavity images and sign information can be obtained by using a rhinitis treatment instrument. The disease category corresponding to the nasal cavity images and sign information of the rhinitis patients is rhinitis. The number of the diseases similar to rhinitis can be at least two. For example, the diseases similar to rhinitis can be cold and nasal polyps.
[0071] It can be understood that the electronic device first needs to construct an initial convolutional neural network model, then train the initial convolutional neural network model, and then obtain the rhinitis recognition model. In an implementation, a caffe tool can be used to construct an initial convolutional neural network model including at least a feature extraction layer and a fully connected layer. For example, the initial convolutional neural network model can be a CNN (Convolutional Neural Network).
[0072] After the initial network model is constructed, the sample data can be input into the initial convolutional neural network model for training to obtain the rhinitis recognition model. The rhinitis recognition model is configured to associate the nasal cavity images and sign information with corresponding disease categories and output a confidence degree corresponding to the disease categories.
[0073] Thus, based on the nasal cavity images and sign information of the rhinitis patients and corresponding disease categories, and the nasal cavity images and sign information of the patients with similar diseases of rhinitis and corresponding disease categories, the initial convolutional neural network model is trained to obtain a rhinitis recognition model. This multi-category model training can enhance the extraction of similar disease characteristics by the model, thereby improving the recognition accuracy.
[0074] During the training, first, the first image features of the nasal cavity images of the rhinitis patients can be extracted by the initial convolutional neural network model. Specifically, the first image features can be extracted by a feature extraction layer. The first image features can include nasal mucosa features and / or nasal passage features. For example, the nasal mucosa features of the rhinitis patients can be pale and edematous or light purple and edematous, and the nasal passage features can be smooth and without polypoid tissue.
[0075] Then, the extracted first image features are spliced with the sign information of the rhinitis patients to obtain first spliced features. The splicing manner can be any one of existing feature splicing manners, which is not limited in the embodiments of the present application.
[0076] After obtaining the first spliced features, the first spliced features are classified to obtain a first initial disease category. A first difference value between the first initial disease category and the corresponding disease category of the rhinitis patient is calculated according to a loss function. Specifically, the first spliced features can be classified by a full connection layer.
[0077] Similarly to the rhinitis patients, the second image features of the nasal cavity images of the patients with similar diseases of rhinitis can be extracted by the initial convolutional neural network model. Specifically, the second image features can be extracted by a feature extraction layer. The second image features can include nasal mucosa features and / or nasal passage features. For example, the nasal mucosa features of the cold patients can be nasal mucosa hyperemia, and the nasal passage features can be smooth and without polypoid tissue. The nasal mucosa features of the patients with nasal polyps can be nasal mucosa hyperemia or edema, and the nasal passage features can be one or more polypoid tissues.
[0078] Then, the extracted second image features are spliced with the sign information of the patients with similar diseases of rhinitis to obtain second spliced features. The splicing manner can be any one of existing feature splicing manners, which is not limited in the embodiments of the present application.
[0079] After obtaining the second spliced features, the second spliced features are classified to obtain a second initial disease category. A second difference value between the second initial disease category and the corresponding disease category of the patient with similar diseases of rhinitis is calculated according to a loss function. Specifically, the second spliced features can be classified by a full connection layer.
[0080] The loss function can be:
[0081] L=λ·L1+(1-λ)·L2+βL3
[0082]
[0083] L3 = max(0, log f) Img -log f1)
[0084] Where L is the loss function, λ is the weight parameter of the classification loss function, used to adjust the proportion of the classification loss function and the authentication loss function in the loss, β is the weight parameter of the concatenated feature validity loss function, L1 is the classification loss function, and L2 is the authentication loss function, used to calculate whether two features belong to the same patient. For the network output features of the i-th sample data of the m-th patient, Let f be the network output feature of the j-th sample data of the m-th patient, and L3 be the loss function for splicing features. Img f1 represents the posterior probability of identifying image features as rhinitis using a pre-defined network model. The pre-defined network model is used to associate nasal cavity images with rhinitis.
[0085] The preset network model is trained based on pure image samples, which only include nasal cavity images of rhinitis patients. The preset network model provides an auxiliary role in training the rhinitis recognition model, that is, calculating the posterior probability of the current sample data, i.e., the posterior probability of recognizing the image features as rhinitis through the preset network model.
[0086] L3 is used to calculate f Img Compared to whether f1 is improved, if it is improved, it is considered that the method of stitching together image features and vital sign information provided by the present invention is effective.
[0087] For example, L1 can be the cross-entropy loss, with β around 0.8, preferably β is 0.8.
[0088] Therefore, the loss function provided in this embodiment of the invention is a multi-task loss function, including an identification task, an authentication task, and a stitching validity task. The identification task is used to identify the category of rhinitis and similar diseases. The authentication task is used to calculate whether two features belong to the same patient. By adding the authentication task, the training process no longer only focuses on the correctness of the current sample data classification, but also on whether the features belong to the same patient, thereby improving the model's generalization ability, avoiding overlearning, and further avoiding the problem of inaccurate identification caused by overlearning. The stitching validity task is used to determine whether the stitching of image features and vital sign information is effective, providing regularization constraints to improve the model's generalization ability.
[0089] It should be noted that the process of feature extraction and splicing for the above-mentioned rhinitis patients and patients with similar rhinitis does not have a specific order. The above description is only for the convenience of description in this embodiment of the invention.
[0090] After obtaining the first and second difference values, the parameters of the initial convolutional neural network model can be adjusted based on the first and second difference values. When the training completion conditions are met, the rhinitis recognition model is obtained.
[0091] The training completion unit is specifically used for:
[0092] When the number of iterations reaches the preset number, the training is completed. The current initial convolutional neural network model is used as a rhinitis recognition model to associate nasal cavity images and vital signs with the corresponding disease categories and output the confidence level corresponding to the disease category.
[0093] The confidence level of the rhinitis recognition model can be:
[0094] f = 1 - f com
[0095]
[0096] Where f is the confidence level of the rhinitis recognition model, f com γ is the confidence score for the competing categories, f2 is the posterior probability of identifying the current sample data as rhinitis, f3 is the largest posterior probability of identifying the current sample data as other disease categories, m1 is the number of sample data in the training set that are identified as rhinitis, and m2 is the number of sample data in the training set that are misidentified as rhinitis as the disease category corresponding to f3.
[0097] In the second formula for the confidence score of the rhinitis recognition model, the part before the plus sign is the confidence score estimate of the competing classes in the current sample data, and the part after the plus sign is the confidence score estimate of the competing classes in the training set.
[0098] For example, the confidence level of the competing category is the confidence level of the disease category most likely to be misjudged, with γ being around 0.9, preferably γ being 0.9.
[0099] In the embodiment of the present application, the competition category confidence is introduced, the confidence of the rhinitis disease category is estimated directly based on the confidence of the maximum possible misjudgment disease category, and the confidence of the rhinitis disease category is avoided to be directly predicted by the model to be greater, the recognition result is irrelevant to the competition category, and there is no reference to the competition category to make a reasonable prediction. The confidence calculated based on the competition category confidence in the embodiment of the present application is more reliable, can alleviate the problem of overlearning of the model leading to overconfidence, and can alleviate the estimation difference caused by extremely uneven distribution of category sample data by introducing the training set, a global statistical data, to calculate the competition category confidence, thereby correcting the confidence of the rhinitis recognition model, and alleviating the estimation difference caused by extremely uneven distribution of category sample data.
[0100] It can be seen that the initial convolutional neural network model is trained by the above feature extraction and splicing manner, which can strengthen the extraction of similar disease characteristics by the model, thereby improving the accuracy of the model.
[0101] According to the above, the system for rhinitis identification provided by the embodiment of the application comprises a rhinitis treatment instrument 1, an information acquisition module 2 and an identification module 3. The rhinitis treatment instrument 1 comprises a plurality of sensors, a head ring 11, a mask support 12 arranged below the head ring 11, a magnetic mask 14 magnetically connected below a nose beam 13 of the mask support 12 and a camera arranged below the nose beam. The information acquisition module 2 is used to acquire the sign information collected by each sensor and the nasal cavity image shot by the camera when the user uses the rhinitis treatment instrument 1. The identification module 3 is used to identify the nasal cavity image and the sign information according to a pre-established rhinitis identification model to obtain an identification result. The rhinitis identification model establishment module comprises an acquisition unit used to acquire each sample data in a training set. The sample data comprises the nasal cavity image and the sign information of the rhinitis patient and the corresponding disease category, and the nasal cavity image and the sign information of the patient with similar diseases of rhinitis and the corresponding disease category. A training unit is used to extract the first image feature of the nasal cavity image of the rhinitis patient and the second image feature of the nasal cavity image of the patient with similar diseases of rhinitis by an initial convolutional neural network model, splice the first image feature with the sign information of the rhinitis patient to obtain first spliced features, splice the second image feature with the sign information of the patient with similar diseases of rhinitis to obtain second spliced features, classify the first spliced features and the second spliced features to obtain a first initial disease category and a second initial disease category, and calculate the first difference value between the first initial disease category and the corresponding disease category of the rhinitis patient and the second difference value between the second initial disease category and the corresponding disease category of the patient with similar diseases of rhinitis according to a loss function. A training completion unit is used to adjust the parameters of the initial convolutional neural network model based on the first difference value and the second difference value, and obtain the rhinitis identification model when a training completion condition is met. Thus, the sign information collected by each sensor and the nasal cavity image shot by the camera are obtained by using the rhinitis treatment instrument 1, and then the nasal cavity image and the sign information are identified according to the pre-established rhinitis identification model to obtain an identification result, so that whether it is rhinitis can be identified. Meanwhile, the initial convolutional neural network model is trained by feature extraction and splicing when the rhinitis identification model is established, which can strengthen the extraction of similar disease features of the model, thereby improving the accuracy of the model and further improving the accuracy of identification of rhinitis. The user does not need to go to the hospital for treatment, the convenience is improved, and the cost is reduced.
[0102] Continuing to refer to Figure 2 The head ring 11 is provided with an air inlet, the inside of the head ring 11 is provided with a water ion module 5, an air inlet device 6 and an air inlet channel, the inside of the mask support 12 is provided with an air outlet channel connected with the air inlet channel, the lower side of the nose beam 13 is provided with a first air outlet, and the two outer side walls 121 of the mask support 12 close to the two cheeks are provided with second air outlets.
[0103] The first air outlet can be one or more, and the second air outlet can be a plurality of through holes arranged in a distribution.
[0104] The first air outlet and the second air outlet can form a double environment. Specifically, the first air outlet forms a water ion environment on the oral surface, and the water ions generated by the water ion module 5 are sent to the nose through breathing, so as to perform rhinitis treatment by water ions. The two second air outlets of the two outer side walls 121 can release water ions to the outside, so as to form an effective sterilization water ion environment in the vicinity of the human body, and use water ions to sterilize and improve the ability to inhibit allergens, formaldehyde and other harmful gases. The water ion environment in the vicinity of the human body can be an area with a radius of at least one meter.
[0105] Therefore, by arranging the water ion module 5 inside the head ring 11, arranging the first air outlet below the nasal beam 13, and arranging the second air outlet on the two outer side walls 121 of the mask support 12 close to the two cheeks, a water ion environment on the oral surface is formed by the first air outlet, and the water ions generated by the water ion module 5 are sent to the nose through breathing, so as to perform rhinitis treatment by water ions. The treatment effect can be achieved, and the water ions generated by the water ion module 5 will not be offset by charged particles in the air, improving the efficiency, and also not generating excess ozone, improving the safety, reducing the cost, not needing to be inserted into the nasal cavity, and directly wearing for use, and also being portable, convenient to use.
[0106] In addition, the two second air outlets of the two outer side walls 121 can release water ions to the outside, so as to form an effective sterilization water ion environment in the vicinity of the human body, and use water ions to sterilize and improve the ability to inhibit allergens, formaldehyde and other harmful gases, reduce the influence of external air environment problems, and further improve the treatment effect on allergic rhinitis.
[0107] Continuing to refer to Figure 2 The head ring 11 includes a head ring body 111, a rear hatch 112, and two side hatches 113.
[0108] The rear hatch 112 covers the rear side of the head ring body 111 to form a rear compartment, and the two side hatches 113 cover the left and right sides of the head ring body 111 to form two side compartments, respectively. The rear compartment and the two side compartments form an air inlet channel, and the rear hatch 112 is provided with an air inlet.
[0109] A filter screen 7 is arranged between the rear compartment and each side compartment, and a water ion module 5 and an air inlet device 6 are arranged inside each side compartment. One of the two side compartments is internally provided with a displacement sensor 15. The air inlet device 6 can be a micro fan.
[0110] Thus, by covering the rear compartment cover 112 on the rear side of the head ring body 111 to form a rear compartment, covering the two side compartment covers 113 on the left and right sides of the head ring body 111 to form two side compartments, and setting the air inlet on the rear compartment cover 112, the rear compartment and the two side compartments form an air inlet channel. By setting the filter screen 7 between the rear compartment and each side compartment, suspended particles in the air entering from the air inlet can be filtered, and the water ion generation efficiency is improved.
[0111] Continuing to refer to Figure 2 , the rear compartment is also provided with a numerical control mainboard 8, and the displacement sensor 15, the nasal cavity gas flow rate sensor 17, the oral cavity gas flow rate sensor 18, the body temperature sensor 16, the nitric oxide sensor 19, the gas component sensor 10, and the camera are in communication connection with the numerical control mainboard 8.
[0112] Thus, by setting the numerical control mainboard 8, the data collected by the displacement sensor 15, the nasal cavity gas flow rate sensor 17, the oral cavity gas flow rate sensor 18, the body temperature sensor 16, the nitric oxide sensor 19, the gas component sensor 10, and the camera can be sent to the numerical control mainboard 8 for processing.
[0113] Continuing to refer to Figure 2 , the rear compartment is also provided with a battery module 9. The displacement sensor 15, the nasal cavity gas flow rate sensor 17, the oral cavity gas flow rate sensor 18, the body temperature sensor 16, the nitric oxide sensor 19, the gas component sensor 10, and the camera are in electrical connection with the battery module 9.
[0114] Thus, by setting the battery module 9 in the rear compartment, the displacement sensor 15, the nasal cavity gas flow rate sensor 17, the oral cavity gas flow rate sensor 18, the body temperature sensor 16, the nitric oxide sensor 19, the gas component sensor 10, and the camera are powered.
[0115] Those skilled in the art can understand that the drawings are only schematic diagrams of an embodiment, and the modules or flows in the drawings are not necessarily necessary for implementing the present application.
[0116] Those skilled in the art can understand that the modules in the device in the embodiment can be distributed in the device in the embodiment as described in the embodiment, or can be correspondingly changed to be located in one or more devices different from the present embodiment. The modules of the above-mentioned embodiments can be combined into one module, or can be further split into multiple sub-modules.
[0117] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the same; although the present application has been described in detail with reference to the foregoing examples, those of ordinary skill in the art should understand that the technical solutions recorded in the foregoing examples can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A system for identifying rhinitis, characterized in that, include: A rhinitis treatment device includes multiple sensors, a headband, a mask bracket located below the headband, a magnetic mask magnetically connected to the nose bridge of the mask bracket, and a camera located below the nose bridge. The information acquisition module is used to acquire vital sign information collected by each sensor and nasal cavity images captured by the camera when the user uses the rhinitis treatment device. The recognition module is used to recognize the nasal cavity image and the vital sign information according to the pre-established rhinitis recognition model, and obtain the recognition result; The rhinitis recognition model building module includes: The acquisition unit is used to acquire sample data from the training set, wherein the sample data includes nasal cavity images and signs of patients with rhinitis and their corresponding disease categories, as well as nasal cavity images and signs of patients with similar diseases to rhinitis and their corresponding disease categories. The training unit is used to extract first image features of the nasal cavity images of the rhinitis patient and second image features of the nasal cavity images of the patients with similar rhinitis through an initial convolutional neural network model. The first image features are concatenated with the physical signs of the rhinitis patient to obtain a first concatenated feature. The second image features are concatenated with the physical signs of the patients with similar rhinitis to obtain a second concatenated feature. The first concatenated feature and the second concatenated feature are classified to obtain a first initial disease category and a second initial disease category. A first difference value between the first initial disease category and the disease category corresponding to the rhinitis patient, and a second difference value between the second initial disease category and the disease category corresponding to the patients with similar rhinitis are calculated according to a loss function. The training completion unit is used to adjust the parameters of the initial convolutional neural network model based on the first difference value and the second difference value, and obtains the rhinitis recognition model when the training completion condition is met.
2. The system for rhinitis identification as described in claim 1, characterized in that, The training completion unit is specifically used for: When the number of iterations reaches the preset number, the training is completed. The current initial convolutional neural network model is used as a rhinitis recognition model to associate nasal cavity images and vital signs with the corresponding disease categories and output the confidence level corresponding to the disease category.
3. The system for rhinitis identification as described in claim 1, characterized in that, The loss function is: L=λ·L1+(1-λ)·L2+βL3 L3=max(0,log f Img -log f1) Where L is the loss function, λ is the weight parameter of the classification loss function, used to adjust the proportion of the classification loss function and the authentication loss function in the loss, β is the weight parameter of the concatenated feature validity loss function, L1 is the classification loss function, and L2 is the authentication loss function, used to calculate whether two features belong to the same patient. For the network output features of the i-th sample data of the m-th patient, Let f be the network output feature of the j-th sample data of the m-th patient, and L3 be the loss function for splicing features. Img f1 is the posterior probability of identifying image features as rhinitis through a preset network model, where f1 is the posterior probability of identifying the stitched features as rhinitis. The preset network model is used to associate nasal cavity images with rhinitis.
4. The system for rhinitis identification as described in claim 1 or 3, characterized in that, The confidence level of the rhinitis recognition model is: f=1-f com Where f is the confidence level of the rhinitis recognition model, f com γ is the confidence score for the competing categories, f2 is the posterior probability of identifying the current sample data as rhinitis, f3 is the largest posterior probability of identifying the current sample data as other disease categories, m1 is the number of sample data in the training set that are identified as rhinitis, and m2 is the number of sample data in the training set that are misidentified as rhinitis as the disease category corresponding to f3.
5. The system for rhinitis identification as described in claim 1, characterized in that, The headband is equipped with a displacement sensor inside, a body temperature sensor is located near the forehead, a nasal airflow sensor is located on the side of the nose bridge near the nose, an oral airflow sensor is located on the side of the magnetic mask near the mouth, and a nitric oxide sensor and a gas composition sensor are respectively located on the two inner walls of the mask bracket near the cheeks.
6. The system for rhinitis identification as described in claim 1, characterized in that, The vital signs information includes at least displacement information, nasal airflow rate information, oral airflow rate information, body temperature information, nitric oxide information, and gas composition information.
7. The system for rhinitis identification as described in claim 1, characterized in that, The headband is equipped with an air inlet, and the interior of the headband is equipped with a water ion module, an air intake device, and an air intake channel; The mask holder has an air outlet connected to the air inlet channel inside, a first air outlet is provided below the nose bridge, and a second air outlet is provided on the two outer walls of the mask holder near the cheeks.
8. The system for rhinitis identification as described in claim 1, characterized in that, The headband includes a headband body, a rear hatch cover, and two side hatch covers; The rear hatch covers the rear side of the head ring body to form a rear compartment, and the two side hatches cover the left and right sides of the head ring body to form two side compartments. The rear compartment and the two side compartments form the air intake channel, and the rear hatch is provided with the air intake port. A filter screen is provided between the rear compartment and each side compartment. The water ion module and the air intake device are provided inside each side compartment. The displacement sensor is provided inside one of the two side compartments.
9. The system for rhinitis identification as described in claim 8, characterized in that, The rear compartment is also equipped with a CNC motherboard; The displacement sensor, the nasal cavity gas flow rate sensor, the oral cavity gas flow rate sensor, the body temperature sensor, the nitric oxide sensor, the gas composition sensor, and the camera are all communicatively connected to the CNC motherboard.
10. The system for rhinitis identification as described in claim 8, characterized in that, The rear compartment is also equipped with a battery module; The displacement sensor, the nasal gas flow rate sensor, the oral gas flow rate sensor, the body temperature sensor, the nitric oxide sensor, the gas composition sensor, and the camera are all electrically connected to the battery module.