A neuropediatric exercise risk analysis system

Abstract

The present invention discloses a neuropediatric exercise hidden danger analysis system, including a hidden danger analysis module and a data sharing processing module, obtaining an infrared imaging image and a visible light image, using an improved Gaussian difference pyramid, respectively extracting the characteristic points and characteristic values ​​of the infrared imaging image and the visible light image, performing multi-resolution decomposition and fusion on the characteristic values ​​of the infrared imaging image and the visible light image, training a YOLO model, adding DIoU matching, obtaining a YOLO recognition model combined with DIoU, and using the YOLO recognition model combined with DIoU to identify hidden dangers in the patient's exercise process. The present invention discovers and evaluates the external environment hidden dangers of children with nerve damage during exercise, the hidden dangers of the patient's own body due to nerve damage, the psychological negative hidden dangers of the patient due to illness, and the hidden dangers of poor sleep after exercise due to neurasthenia according to the sleep hidden danger analysis unit, and shares the data with parents and medical teams through the data sharing processing module.

Description

Technical Field

[0001] The present invention relates to the technical field of neuropediatrics, and in particular to a neuropediatrics exercise risk analysis system. Background Art

[0002] In the field of pediatric neurology, there may be potential risks or hidden dangers associated with exercise. These risks may involve problems with the function, development or structure of the nervous system, which may affect the child's motor skills, coordination and other aspects related to the nervous system. The rehabilitation training of children needs to be carried out in stages according to the child's own physical and psychological level. For the hidden dangers of exercise in pediatric neurology, the medical team needs to conduct comprehensive identification and evaluation, develop personalized treatment and rehabilitation plans, and maximize the quality of life and functional ability of children.

[0003] At present, the Chinese invention patent application with application number CN202410316338.1 discloses a neuropediatric exercise recovery evaluation method and evaluation system. Although the system can support the evaluation of recovery after exercise, it cannot detect the hidden dangers that may occur during exercise for children with nerve damage, including hidden dangers in the external environment during exercise, hidden dangers caused by nerve damage to the patient's body, hidden dangers of psychological negativity due to illness, hidden dangers of comorbidities caused by infection during hospital treatment, and hidden dangers of poor sleep after exercise due to neurasthenia. It also cannot share data with parents and medical teams to adjust the exercise plan and intensity for patients. Summary of the invention

[0004] The technical problem solved by the present invention is that the related technologies cannot discover the hidden dangers that may occur when children with nerve damage exercise, including hidden dangers in the external environment when the patient exercises, hidden dangers caused by nerve damage in the patient's own body, hidden dangers of psychological negativity due to illness, hidden dangers of comorbidities caused by infection during hospital treatment, and hidden dangers of poor sleep after exercise due to neurasthenia. The data cannot be shared with parents and medical teams to adjust the exercise plan and intensity for the patient.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a neuropediatric exercise hidden danger analysis system, comprising a hidden danger analysis module and a data sharing processing module;

[0006] The hidden danger analysis module is used to obtain infrared imaging images and visible light images, and use the improved Gaussian difference pyramid to extract the characteristic points and characteristic values ​​of the infrared imaging images and visible light images respectively, use the characteristic points to align the infrared imaging images and the visible light images, perform multi-resolution decomposition and fusion of the characteristic values ​​of the infrared imaging images and the visible light images, and generate a multimodal image;

[0007] Perform differential edge detection on the multimodal image to extract edge features, divide the edge features, the fused feature values ​​and the corresponding annotation information into a training set and a validation set respectively, use the training set to train the YOLO model, use DIoU as a measure of the bounding box position error combined with confidence loss for optimization, modify the loss function of the target detection head, add DIoU matching, use the validation set to evaluate the trained YOLO model, post-process the bounding box output by the trained YOLO model using a non-maximum suppression method, eliminate bounding boxes with more overlap, retain bounding boxes with confidence levels higher than a pre-set confidence threshold, obtain a YOLO recognition model combined with DIoU, and use the YOLO recognition model combined with DIoU to identify hidden dangers in patients during exercise;

[0008] The data sharing processing module uploads the identified hidden dangers of the patient during exercise to the cloud, obtains cloud data, compares it with the doctor's clinical diagnosis results and automatically improves the cloud data, and adjusts the child's rehabilitation exercise plan based on the cloud data, individual differences of the child and the patient's rehabilitation needs.

[0009] As a preferred solution of the neuropediatric exercise risk analysis system described in the present invention, the feature point and feature value extraction includes:

[0010] Performing Gaussian filtering on the infrared imaging image and the visible light image using different standard deviations, inputting the filtered infrared imaging image and the visible light image into a pooling layer, performing pooling operations of different sizes respectively, obtaining feature maps of the infrared imaging image and the visible light image, downsampling the feature maps at corresponding scales, and putting the downsampled infrared imaging image and the visible light image into the next layer as input features for repeated filtering to obtain a Gaussian pyramid, guiding the difference semantic information of each layer of the Gaussian pyramid with the difference semantic information of the previous layer, establishing a relationship between cross-scale difference feature information, adding a multi-branch pooling module, obtaining more representative difference semantic information, obtaining an improved Gaussian difference pyramid, and obtaining the position and size of the feature point according to the improved Gaussian difference pyramid;

[0011] As a preferred solution of the neuropediatric exercise risk analysis system described in the present invention, the risk analysis module specifically includes an external risk analysis unit, a muscle testing unit, an emotional risk analysis unit and a sleep risk analysis unit.

[0012] As a preferred solution of the neuropediatric exercise hazard analysis system described in the present invention, the external hazard analysis unit uses a camera with an infrared imaging function to perform an overall scan of the patient's environment, time-division multiplexing the infrared imaging function and the ordinary camera function of the camera, alternately collects infrared imaging images and visible light images, and establishes corresponding annotation files for each infrared imaging image and visible light image, respectively, the annotation files include the target position and category in the infrared imaging image and the visible light image, respectively, use a scale-invariant feature transformation algorithm to extract feature points and feature values ​​from the collected infrared imaging image and visible light image, perform registration operations on the infrared imaging image and the visible light image, find corresponding feature points in adjacent infrared imaging images and visible light images, and use a random sampling consistency algorithm to filter out error matching points;

[0013] The registration operation specifically includes:

[0014] The infrared imaging image and the visible light image are translated, rotated and scaled, and feature points are matched using a nearest neighbor matching algorithm to obtain corresponding and consistent feature points in the infrared imaging image and the visible light image after translation, rotation and scaling;

[0015] The multi-resolution decomposition and fusion of infrared imaging and visible light image feature values ​​specifically include:

[0016] The infrared image and visible light image are decomposed into feature representations of different scales according to different pixel values ​​using pyramid decomposition technology, and the features of the image at different frequencies or spatial scales are extracted. At the resolution represented by each different scale, features are extracted from the infrared image and visible light image, and feature fusion is performed using the feature concatenation method, and the dimensionality reduction operation is performed on the fused features.

[0017] The features extracted from the infrared image include color and a first shape, and the features extracted from the visible light image include a second shape and a texture;

[0018] According to the YOLO recognition model combined with DIoU, the hidden dangers of identifying patients during exercise specifically include:

[0019] The YOLO recognition model combined with DIoU is used to identify the external environmental hidden danger factors, the YOLO recognition model combined with DIoU is used to identify the human muscle vibration amplitude frequency according to the time division multiplexing frame interval duration, the YOLO recognition model combined with DIoU is used to identify the overall amplitude of human exercise movements and human body temperature changes according to the human body temperature color contained in the infrared imaging image, and the patient's emotional changes are obtained according to the recognized facial expressions. As a preferred embodiment of the neuropediatric exercise hidden danger analysis system described in the present invention, the position and size of the feature points obtained according to the improved Gaussian difference pyramid specifically include:

[0020] Query the local maximum or minimum pixel points in the Gaussian difference pyramid of each scale respectively, determine the local extreme point by comparing each pixel with its adjacent pixels at the corresponding scale, the local extreme point is the feature point, determine the exact position of the feature point by interpolation method, and estimate the scale of the feature point;

[0021] The mathematical expression for Gaussian filtering the infrared image and the visible light image is:

[0022] ;

[0023] ;

[0024] in, represents the filtered image, The scale is The Gaussian kernel of represents the source feature map, Indicates the target location in the image;

[0025] The mathematical expression of the pooling operation is:

[0026] ;

[0027] in, It indicates that the initial output after the pooling layer has a downsampling factor of 1 / n, and a indicates the input feature map. represents the number of scales in the cross-scale feature extraction module, Indicates size Pooling layer operation;

[0028] The mutual guidance mathematical expression of the difference semantic information is:

[0029] ;

[0030] ;

[0031] ;

[0032] ;

[0033] in, Indicates The initial differential features of the layer, Representation 1 1 convolution operation, Representation 3 2 convolution operations, and They represent the intermediate tensors of the cross-scale difference feature attention module, Indicates The differential feature vector after cross-scale difference feature extraction at the layer, [ ] represents the vector concatenation operator;

[0034] The multi-branch pooling module specifically includes:

[0035] The input features are downsampled at different rates to learn features of different scales, and the upsampled features of different scales are restored to their original sizes, and fused with the original features using residual connections. The mathematical expression is:

[0036] ;

[0037] in, , ... It represents the spatial features of different scales learned by each branch through up and down sampling. Up represents upsampling and down represents downsampling.

[0038] As a preferred solution of the neuro-pediatric exercise hazard analysis system described in the present invention, the muscle testing unit collects the surface electromyography information of the patient before and after exercise, the surface electromyography information includes the potential amplitude of the patient's muscles and the conduction velocity of the nerves, draws a surface electromyography information change curve, and adjusts the patient's exercise intensity according to the surface electromyography information change curve.

[0039] As a preferred solution of the neuropediatric exercise risk analysis system described in the present invention, the emotional risk analysis unit specifically includes an expression analysis part and a depression detection part:

[0040] The expression analysis part is identified by the YOLO recognition model combined with DIoU;

[0041] The depression detection part includes:

[0042] The depression test evaluates the patient's psychological state through the dimensions of anxiety and depression. The dimensions of anxiety and depression include a total of 14 different items. Each item uses the Likert 4-level scoring method to test the patient's depression. The Likert 4-level scoring method includes the first level, the second level, the third level and the fourth level. The first level corresponds to 0 points, the second level corresponds to 1 point, the third level corresponds to 2 points, and the fourth level corresponds to 3 points. The total score is 0 to 21 points. The higher the score, the higher the degree of anxiety and depression of the patient, and the worse the psychological state. If the score of any dimension of the anxiety and depression is not less than 8 points, the patient's psychological state is judged to be poor, and if the scores of the anxiety and depression dimensions are less than 8 points respectively, the patient's psychological state is judged to be good.

[0043] As a preferred solution of the neuropediatric exercise risk analysis system of the present invention, the sleep risk analysis unit is used to monitor the patient's night sleep after exercise, obtain the patient's night sleep data, analyze the patient's night sleep after exercise based on the night sleep data, and judge the effect of exercise on the patient's neurological health recovery according to the analysis results;

[0044] The sleep risk analysis unit monitors the patient's night sleep after exercise, analyzes the patient's night sleep after exercise based on the monitored night sleep data, and determines the effect of exercise on the patient's restoration of nerve health based on the analysis results;

[0045] The nighttime sleep data includes the duration of falling asleep, total sleep duration, deep sleep time at night, light sleep time at night, and wakefulness time at night;

[0046] Determining the effect of exercise on the patient's restoration of neurological health specifically includes:

[0047] The duration of falling asleep is set to negative, the duration of deep sleep at night is set to positive, the duration of light sleep at night is set to positive, and the duration of wakefulness at night is set to negative. The weighted sum is performed to obtain a total sleep value that can express the sleep quality. The threshold [e, b] is set to the healthy total sleep value interval. Within the healthy total sleep value interval, the larger the total sleep value, the better the patient's sleep quality, and the smaller the total sleep value, the worse the patient's sleep quality.

[0048] As a preferred embodiment of the neuropediatric exercise risk analysis system of the present invention, the external environmental risk factors include sharp objects, hard right angles, obstacles in the road, easily knocked over and fragile objects, easily entangled thin ropes, large particles floating in the air and objects with too high or too low temperature;

[0049] After detecting the external environmental factors that affect the patient, removing the external environmental factors;

[0050] The patient's emotional changes include happiness, calmness, sadness and anger, and the daily time proportions of the patient's happiness, calmness, sadness and anger are obtained. Based on the daily time proportions of the patient's four emotions, the patient's main emotion of the day is obtained.

[0051] As a preferred solution of the neuropediatric exercise risk analysis system described in the present invention, the data sharing processing module monitors the exercise situation and physiological state of the child in real time, and provides real-time feedback to the rehabilitation therapist, parents and doctors.

[0052] The beneficial effects of the present invention are as follows: according to the external hidden danger analysis unit, the hidden dangers of the external environment of children with nerve damage during exercise are discovered and evaluated; according to the muscle testing unit, the hidden dangers of the patient's own body caused by nerve damage are discovered and evaluated; according to the emotional hidden danger analysis unit, the hidden dangers of negative psychological state caused by illness are discovered and evaluated; according to the sleep hidden danger analysis unit, the hidden dangers of poor sleep after exercise due to neurasthenia are discovered and evaluated; and the data are shared with parents and medical teams through the data sharing processing module so as to adjust the exercise plan and intensity for the patient. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A basic flow chart of a neuropediatric exercise risk analysis system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0055] Example 1, reference Figure 1 , as one embodiment of the present invention, provides a neuropediatric exercise hidden danger analysis system, including a hidden danger analysis module and a data sharing processing module;

[0056] The hidden danger analysis module is used to obtain infrared imaging images and visible light images, and use the improved Gaussian difference pyramid to extract the characteristic points and characteristic values ​​of the infrared imaging images and visible light images respectively, use the characteristic points to align the infrared imaging images and the visible light images, perform multi-resolution decomposition and fusion of the characteristic values ​​of the infrared imaging images and the visible light images, and generate a multimodal image;

[0057] Perform differential edge detection on the multimodal image to extract edge features, divide the edge features, the fused feature values ​​and the corresponding annotation information into a training set and a validation set respectively, use the training set to train the YOLO model, use DIoU as a measure of the bounding box position error and optimize it in combination with the confidence loss when calculating the loss function during the training process, modify the loss function of the target detection head, add DIoU matching, use the validation set to evaluate the trained YOLO model, verify the degree of optimization of the DIoU matching on the detection performance, and the evaluation index is the detection accuracy. Post-process the bounding box output by the trained YOLO model using a non-maximum suppression method to eliminate bounding boxes with more overlaps, retain bounding boxes with confidence levels higher than a pre-set confidence threshold, and obtain a YOLO recognition model combined with DIoU, and use the YOLO recognition model combined with DIoU to identify hidden dangers in patients during exercise;

[0058] The data sharing processing module uploads the identified hidden dangers of the patient during exercise to the cloud, obtains cloud data, compares it with the doctor's clinical diagnosis results and automatically improves the cloud data, and adjusts the child's rehabilitation exercise plan based on the cloud data, individual differences of the child and the patient's rehabilitation needs.

[0059] Based on cloud data, individual differences of children and rehabilitation needs of patients, doctors adjust the rehabilitation exercise program for children according to professional knowledge. The rehabilitation exercise program includes exercise movements, exercise intensity and exercise duration.

[0060] The external hidden danger analysis unit discovers and evaluates the hidden dangers of the external environment of children with nerve damage during exercise; the muscle testing unit discovers and evaluates the hidden dangers of the patient's own body due to nerve damage; the emotional hidden danger analysis unit discovers and evaluates the hidden dangers of psychological negativity caused by illness; the sleep hidden danger analysis unit discovers and evaluates the hidden dangers of poor sleep after exercise due to neurasthenia in patients; and the data sharing and processing module is used to share the data with parents and medical teams so that the exercise plan and intensity can be adjusted for the patient.

[0061] The feature point and feature value extraction includes:

[0062] The infrared imaging image and the visible light image are Gaussian filtered using different standard deviations, the filtered infrared imaging image and the visible light image are input into a pooling layer, pooling operations of different sizes are performed respectively, the feature map is downsampled at a corresponding scale, the downsampled infrared imaging image and the visible light image are put into the next layer as input features for repeated filtering to obtain a Gaussian pyramid, the difference semantic information of each layer of the Gaussian pyramid is mutually guided by the difference semantic information of the previous layer, a relationship between cross-scale difference feature information is established, a multi-branch pooling module is added to obtain more representative difference semantic information, an improved Gaussian difference pyramid is obtained, and the position and size of the feature point are obtained according to the improved Gaussian difference pyramid.

[0063] The improved Gaussian difference pyramid retains the original rich semantic information while extracting more semantic information. The deep features have rich semantic information, which is very helpful for guiding the semantic information of the target. The shallow features have rich global information, which is helpful for restoring the overall contour of the target. Compared with other types of feature pyramids, it avoids the loss of feature information caused by the unified channel fusion of features of different scales, and reduces the aliasing effect caused by upsampling the features first. Different resolution channels are progressively connected to deepen the association between different scales.

[0064] The hidden danger analysis module specifically includes an external hidden danger analysis unit, a muscle testing unit, an emotional hidden danger analysis unit and a sleep hidden danger analysis unit.

[0065] The external hidden danger analysis unit uses a camera with an infrared imaging function to perform an overall scan of the patient's environment, time-division multiplexes the infrared imaging function and the ordinary camera function of the camera, alternately collects infrared imaging images and visible light images, and respectively establishes corresponding annotation files for each infrared imaging image and visible light image, the annotation files respectively include the target position and category in the infrared imaging image and the visible light image, uses a scale-invariant feature transformation algorithm to extract feature points and feature values ​​from the collected infrared imaging image and visible light image, performs a registration operation on the infrared imaging image and the visible light image, searches for corresponding feature points in adjacent infrared imaging images and visible light images, and uses a random sampling consistency algorithm to filter out error matching points;

[0066] The registration operation specifically includes:

[0067] The infrared imaging image and the visible light image are translated, rotated and scaled, and feature points are matched using a nearest neighbor matching algorithm to obtain corresponding and consistent feature points in the infrared imaging image and the visible light image after translation, rotation and scaling;

[0068] The multi-resolution decomposition and fusion of infrared imaging and visible light image feature values ​​specifically include:

[0069] The infrared image and visible light image are decomposed into feature representations of different scales according to different pixel values ​​using pyramid decomposition technology. Feature representations are extracted from the infrared image and visible light image, and feature fusion is performed using the feature concatenation method. The fused features are subjected to dimensionality reduction operations to reduce the feature dimension and improve computational efficiency.

[0070] Decomposing feature representations of different scales according to different pixel values ​​includes clustering the pixel values ​​of the infrared imaging image and the visible light image of each improved Gaussian difference pyramid layer to obtain different pixel value clusters, using different methods to extract features from different pixel value clusters, the methods including statistical features, texture features and frequency domain features, integrating or combining feature information extracted at different scales to obtain feature representations of the infrared imaging image and the visible light image at multiple scales.

[0071] The features extracted from the infrared image include color and a first shape, and the features extracted from the visible light image include a second shape and a texture;

[0072] According to the YOLO recognition model combined with DIoU, the hidden dangers of identifying patients during exercise specifically include:

[0073] The YOLO recognition model combined with DIoU is used to identify the external environmental hidden danger factors, and the YOLO recognition model combined with DIoU is used to identify the human muscle vibration amplitude frequency according to the time division multiplexing frame interval duration. The YOLO recognition model combined with DIoU is used to identify the overall amplitude of human exercise movements and human body temperature changes according to the human body temperature color contained in the infrared imaging image. When the patient's movement amplitude is too large, a warning is issued to prevent excessive exercise from aggravating the condition, and the patient's emotional changes are obtained according to the recognized facial expressions. The external hidden danger analysis unit is used to discover and evaluate the external environmental hidden dangers of children with neurological damage during exercise.

[0074] The position and size of the feature points obtained according to the improved Gaussian difference pyramid include:

[0075] Query the local maximum or minimum pixel points in the Gaussian difference pyramid of each scale respectively, determine the local extreme point by comparing each pixel with its adjacent pixels at the corresponding scale, the local extreme point is the feature point, determine the exact position of the feature point by interpolation method, and estimate the scale of the feature point;

[0076] The mathematical expression for Gaussian filtering the infrared image and the visible light image is:

[0077] ;

[0078] ;

[0079] in, represents the filtered image, The scale is The Gaussian kernel of represents the source feature map, Indicates the target location in the image;

[0080] The mathematical expression of the pooling operation is:

[0081] ;

[0082] in, It indicates that the initial output after the pooling layer has a downsampling factor of 1 / n, and a indicates the input feature map. represents the number of scales in the cross-scale feature extraction module, Indicates size Pooling layer operation;

[0083] The mutual guidance mathematical expression of the difference semantic information is:

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] in, Indicates The initial differential features of the layer, Representation 1 1 convolution operation, Representation 3 2 convolution operations, and They represent the intermediate tensors of the cross-scale difference feature attention module, Indicates The differential feature vector after cross-scale difference feature extraction at the layer, [ ] represents the vector concatenation operator;

[0089] The multi-branch pooling module specifically includes:

[0090] The input features are downsampled at different rates to learn features of different scales, and the upsampled features of different scales are restored to their original sizes, and fused with the original features using residual connections. The mathematical expression is:

[0091] ;

[0092] in, , ... It represents the spatial features of different scales learned by each branch through up and down sampling. Up represents upsampling and down represents downsampling.

[0093] The muscle testing unit collects the surface electromyography information of the patient before and after exercise, the surface electromyography information including the potential amplitude of the patient's muscles and the conduction velocity of the nerves, draws a surface electromyography information change curve, and adjusts the patient's exercise intensity according to the surface electromyography information change curve.

[0094] The emotional hidden danger analysis unit specifically includes an expression analysis part and a depression detection part:

[0095] The expression analysis part is identified by the YOLO recognition model combined with DIoU;

[0096] The depression detection part includes:

[0097] The depression test evaluates the patient's psychological state through the dimensions of anxiety and depression. The dimensions of anxiety and depression include a total of 14 different items. Each item uses the Likert 4-level scoring method to test the patient's depression. The Likert 4-level scoring method includes the first level, the second level, the third level and the fourth level. The first level corresponds to 0 points, the second level corresponds to 1 point, the third level corresponds to 2 points, and the fourth level corresponds to 3 points. The total score is 0 to 21 points. The higher the score, the higher the degree of anxiety and depression of the patient, and the worse the psychological state. If the score of any dimension of the anxiety and depression is not less than 8 points, the patient's psychological state is judged to be poor, and if the scores of the anxiety and depression dimensions are less than 8 points respectively, the patient's psychological state is judged to be good.

[0098] The depression detection part conducts a depression test on the patient after the patient finishes exercising, analyzes whether the exercise has a negative impact on the patient or whether the patient himself has negative emotions, and provides psychological counseling to the patient in a timely manner.

[0099] The sleep risk analysis unit is used to monitor the patient's nighttime sleep after exercise, obtain the patient's nighttime sleep data, analyze the patient's nighttime sleep after exercise based on the nighttime sleep data, and determine the effect of exercise on the patient's nerve recovery health according to the analysis results;

[0100] The sleep risk analysis unit monitors the patient's night sleep after exercise, analyzes the patient's night sleep after exercise based on the monitored night sleep data, and determines the effect of exercise on the patient's restoration of nerve health based on the analysis results;

[0101] The nighttime sleep data includes the duration of falling asleep, total sleep duration, deep sleep time at night, light sleep time at night, and wakefulness time at night;

[0102] Determining the effect of exercise on the patient's restoration of neurological health specifically includes:

[0103] The duration of falling asleep is set to negative, the duration of deep sleep at night is set to positive, the duration of light sleep at night is set to positive, and the duration of wakefulness at night is set to negative. The weighted sum is performed to obtain a total sleep value that can express the sleep quality. The threshold [e, b] is set to the healthy total sleep value interval. Within the healthy total sleep value interval, the larger the total sleep value, the better the patient's sleep quality, and the smaller the total sleep value, the worse the patient's sleep quality.

[0104] The external environmental hidden dangers include sharp objects, hard right angles, obstacles in the road, easily knocked over fragile objects, easily entangled thin ropes, large particles floating in the air and objects with too high or too low temperatures;

[0105] After detecting the external environmental factors that affect the patient, removing the external environmental factors;

[0106] The patient's emotional changes include happiness, calmness, sadness and anger, and the daily time proportions of the patient's happiness, calmness, sadness and anger are obtained. Based on the daily time proportions of the patient's four emotions, the patient's main emotion of the day is obtained.

[0107] The data sharing processing module is used to monitor the children's exercise status and physiological status in real time, and the data on the exercise status and physiological status are sent to the mobile phones of rehabilitation therapists, parents and doctors. Real-time feedback is provided to rehabilitation therapists, parents and doctors, which helps rehabilitation therapists and doctors to adjust the rehabilitation plan in time and helps parents to have a detailed understanding of the rehabilitation process.

[0108] The present invention uses the YOLO recognition model combined with DIoU to identify the hidden dangers of patients, discovers and evaluates the external environment hidden dangers of children with nerve damage during exercise according to the external hidden danger analysis unit in the hidden danger analysis module, discovers and evaluates the hidden dangers of the patient's own body caused by nerve damage according to the muscle testing unit, discovers and evaluates the psychological negative hidden dangers caused by illness according to the emotional hidden danger analysis unit, and discovers and evaluates the hidden dangers of poor sleep after exercise due to neurasthenia according to the sleep hidden danger analysis unit, and shares the data with parents and medical teams through the data sharing processing module to adjust the exercise plan and intensity for the patient.

[0109] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0110] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A neuropediatric exercise risk analysis system, characterized by: Including hidden danger analysis module and data sharing processing module; The hidden danger analysis module is used to obtain infrared imaging images and visible light images, and use the improved Gaussian difference pyramid to extract the characteristic points and characteristic values ​​of the infrared imaging images and visible light images respectively, use the characteristic points to align the infrared imaging images and the visible light images, perform multi-resolution decomposition and fusion of the characteristic values ​​of the infrared imaging images and the visible light images, and generate a multimodal image; Performing differential edge detection on the multimodal image to extract edge features, dividing the edge features, the fused feature values ​​and the corresponding annotation information into a training set and a validation set respectively, using the training set to train the YOLO model, optimizing the DIoU as a measure of the bounding box position error in the training in combination with the confidence loss, modifying the loss parameter and adding the DIoU matching to obtain a YOLO recognition model combined with DIoU, and using the YOLO recognition model combined with DIoU to identify hidden dangers in the patient's exercise process; The data sharing processing module uploads the identified hidden dangers of the patient during exercise to the cloud, obtains cloud data, compares it with the doctor's clinical diagnosis results and automatically improves the cloud data, and adjusts the child's rehabilitation exercise plan based on the cloud data, individual differences of the child and the patient's rehabilitation needs; The infrared imaging image and the visible light image are photographs of the patient's whole body exercise environment; The feature point and feature value extraction includes: Performing Gaussian filtering on the infrared imaging image and the visible light image using different standard deviations, inputting the filtered infrared imaging image and the visible light image into a pooling layer, performing pooling operations of different sizes respectively, obtaining feature maps of the infrared imaging image and the visible light image, downsampling the feature maps at corresponding scales, and putting the downsampled infrared imaging image and the visible light image into the next layer as input features for repeated filtering to obtain a Gaussian pyramid, guiding the difference semantic information of each layer of the Gaussian pyramid with the difference semantic information of the previous layer, establishing a relationship between cross-scale difference feature information, adding a multi-branch pooling module, obtaining more representative difference semantic information, obtaining an improved Gaussian difference pyramid, and obtaining the position and size of the feature point according to the improved Gaussian difference pyramid; The hidden danger analysis module specifically includes an external hidden danger analysis unit, a muscle testing unit, an emotional hidden danger analysis unit and a sleep hidden danger analysis unit; The position and size of the feature points obtained according to the improved Gaussian difference pyramid include: Query the local maximum or minimum pixel points in the Gaussian difference pyramid of each scale respectively, determine the local extreme point by comparing each pixel with its adjacent pixels at the corresponding scale, the local extreme point is the feature point, determine the exact position of the feature point by interpolation method, and estimate the scale of the feature point; The mathematical expression for Gaussian filtering the infrared image and the visible light image is: ； ； in, represents the filtered image, The scale is The Gaussian kernel of represents the source feature map, Indicates the target location in the image; The mathematical expression of the pooling operation is: ； in, It indicates that the initial output after the pooling layer has a downsampling factor of 1 / n, and a indicates the input feature map. represents the number of scales in the cross-scale feature extraction module, Indicates size Pooling layer operation; The mutual guidance mathematical expression of the difference semantic information is: ； ； ； ； in, Indicates The initial differential features of the layer, Representation 1 1 convolution operation, Representation 3 2 convolution operations, and They represent the intermediate tensors of the cross-scale difference feature attention module, Indicates The differential feature vector after cross-scale difference feature extraction at the layer, [ ] represents the vector concatenation operator; The multi-branch pooling module specifically includes: The input features are downsampled at different rates to learn features of different scales, and the upsampled features of different scales are restored to their original sizes, and fused with the original features using residual connections. The mathematical expression is: ； in, , ... It represents the spatial features of different scales learned by each branch through up and down sampling. Up represents upsampling and down represents downsampling.

2. The neuropediatric exercise risk analysis system according to claim 1, characterized in that: The external hidden danger analysis unit uses a camera with an infrared imaging function to perform an overall scan of the patient's environment, time-division multiplexes the infrared imaging function and the ordinary camera function of the camera, alternately collects infrared imaging images and visible light images, and respectively establishes corresponding annotation files for each infrared imaging image and visible light image, the annotation files respectively include the target position and category in the infrared imaging image and the visible light image, uses a scale-invariant feature transformation algorithm to extract feature points and feature values ​​from the collected infrared imaging image and visible light image, performs a registration operation on the infrared imaging image and the visible light image, searches for corresponding feature points in adjacent infrared imaging images and visible light images, and uses a random sampling consistency algorithm to filter out error matching points; The registration operation specifically includes: The infrared imaging image and the visible light image are translated, rotated and scaled, and feature points are matched using a nearest neighbor matching algorithm to obtain corresponding and consistent feature points in the infrared imaging image and the visible light image after translation, rotation and scaling; The multi-resolution decomposition and fusion of infrared imaging and visible light image feature values ​​specifically include: Using pyramid decomposition technology to decompose the infrared imaging image and the visible light image into feature representations of different scales according to different pixel values, extracting features of the image at different frequencies or spatial scales, extracting features from the infrared imaging image and the visible light image at the resolution represented by each different scale, using a feature concatenation method to perform feature fusion, and performing a dimensionality reduction operation on the fused features; the features extracted from the infrared imaging image include color and a first shape, and the features extracted from the visible light image include a second shape and a texture; According to the YOLO recognition model combined with DIoU, the hidden dangers of identifying patients during exercise specifically include: The YOLO recognition model combined with DIoU is used to identify the hidden dangers of the external environment. The YOLO recognition model combined with DIoU is used to identify the amplitude and frequency of human muscle vibration according to the frame interval duration of time division multiplexing. The YOLO recognition model combined with DIoU is used to identify the overall amplitude of human exercise movements and human body temperature changes according to the human body temperature color contained in the infrared imaging image. The changes in the patient's emotions are obtained based on the identified facial expressions.

3. The neuropediatric exercise risk analysis system according to claim 1, characterized in that: The muscle testing unit collects the surface electromyography information of the patient before and after exercise, the surface electromyography information including the potential amplitude of the patient's muscles and the conduction velocity of the nerves, draws a surface electromyography information change curve, and adjusts the patient's exercise intensity according to the surface electromyography information change curve.

4. The neuropediatric exercise risk analysis system according to claim 1, characterized in that: The emotional hidden danger analysis unit specifically includes an expression analysis part and a depression detection part: The expression analysis part is identified by the YOLO recognition model combined with DIoU; The depression detection part includes: The depression test evaluates the patient's psychological state through the dimensions of anxiety and depression. The dimensions of anxiety and depression include a total of 14 different items. Each item uses the Likert 4-level scoring method to test the patient's depression. The Likert 4-level scoring method includes the first level, the second level, the third level and the fourth level. The first level corresponds to 0 points, the second level corresponds to 1 point, the third level corresponds to 2 points, and the fourth level corresponds to 3 points. The total score is 0 to 21 points. The higher the score, the higher the degree of anxiety and depression of the patient, and the worse the psychological state. If the score of any dimension of the anxiety and depression is not less than 8 points, the patient's psychological state is judged to be poor, and if the scores of the anxiety and depression dimensions are less than 8 points respectively, the patient's psychological state is judged to be good.

5. The neuropediatric exercise risk analysis system according to claim 1, characterized in that: The sleep risk analysis unit is used to monitor the patient's nighttime sleep after exercise, obtain the patient's nighttime sleep data, analyze the patient's nighttime sleep after exercise based on the nighttime sleep data, and determine the effect of exercise on the patient's nerve recovery health according to the analysis results; The sleep risk analysis unit monitors the patient's night sleep after exercise, analyzes the patient's night sleep after exercise based on the monitored night sleep data, and determines the effect of exercise on the patient's restoration of nerve health based on the analysis results; The nighttime sleep data includes the duration of falling asleep, total sleep duration, deep sleep time at night, light sleep time at night, and wakefulness time at night; Determining the effect of exercise on the patient's restoration of neurological health specifically includes: The duration of falling asleep is set to negative, the duration of deep sleep at night is set to positive, the duration of light sleep at night is set to positive, and the duration of wakefulness at night is set to negative. The weighted sum is performed to obtain a total sleep value that can express the sleep quality. The threshold [e, b] is set to the healthy total sleep value interval. Within the healthy total sleep value interval, the larger the total sleep value, the better the patient's sleep quality, and the smaller the total sleep value, the worse the patient's sleep quality.

6. The neuropediatric exercise risk analysis system according to claim 2, characterized in that: The external environmental hidden dangers include sharp objects, hard right angles, obstacles in the road, easily knocked over fragile objects, easily entangled thin ropes, large particles floating in the air and objects with too high or too low temperatures; After detecting the external environmental factors that affect the patient, removing the external environmental factors; The patient's emotional changes include happiness, calmness, sadness and anger, and the daily time proportions of the patient's happiness, calmness, sadness and anger are obtained. Based on the daily time proportions of the patient's four emotions, the patient's main emotion of the day is obtained.

7. The neuropediatric exercise risk analysis system according to claim 1, characterized in that: The data sharing processing module can monitor the child's exercise status and physiological state in real time, and provide real-time feedback to the rehabilitation therapist, parents and doctors.

Images