Ankle pump training monitoring and recognition system of deep learning model based on electromyographic signals

Abstract

According to the ankle pump training monitoring and recognition system of the deep learning model based on the electromyographic signals, based on a target ankle pump passive training scheme, a first driving assembly is controlled to be started, an arc-shaped rotating seat is driven to rotate, an ankle and sole bearing seat is driven to rotate accordingly, and the ankle of a target bedridden patient swings left and right accordingly for training; a second driving assembly is controlled to be started, an ankle and sole bearing seat is driven to move front and back, and then the ankles of the target bedridden patient are subjected to stretching and bending training; the method comprises the following steps: receiving an electromyographic signal, measured by an electromyographic sensor, of a shank of a target bedridden patient within a set time at the beginning of training, preprocessing the electromyographic signal, performing feature extraction on the preprocessed electromyographic signal to obtain an electromyographic time domain feature and an electromyographic frequency domain feature, and inputting a formed one-dimensional feature into a target deep venous thrombosis risk identification model for training. Whether the target bedridden patient has the lower limb deep vein thrombosis risk or not is recognized, the ankle stretching and bending training intensity and the ankle left-right swinging training intensity are increased sometimes, and the original intensity is kept if not.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an ankle pump training monitoring and recognition system based on a deep learning model of electromyographic signals. Background Art

[0002] At present, ankle pump exercise is one of the most important measures for bedridden patients to prevent deep vein thrombosis, and it is also a basic preventive measure. In clinical practice, it is found that many bedridden patients are unwilling to take the initiative to do ankle pump exercise or cannot persist in doing ankle pump exercise due to various reasons, such as physical reasons, cognitive reasons, pain after surgery, etc.

[0003] There is currently a patent for invention with application number CN202111015256.6, entitled "An ankle assistive device for preventing venous thrombosis". Although this patent can realize passive ankle pump training, it can only realize forward and backward flexion and extension movements, and cannot realize left and right swinging movements. There is also a patent for invention with application number CN202010528883.9, entitled "An auxiliary therapeutic device for lower limb and ankle activities for bedridden patients". Although this patent can realize passive ankle pump training, it can only realize forward and backward flexion and extension movements, and cannot realize left and right swinging movements.

[0004] Based on this, an ankle pump training monitoring and identification system based on a deep learning model of electromyographic signals was designed, which enables bedridden patients to perform passive ankle pump training. It can not only realize the forward and backward flexion and extension movement of the ankle of bedridden patients, but also the left and right swinging movement. During the ankle pump training, it can monitor whether bedridden patients have the risk of deep vein thrombosis in the lower limbs. If so, further intensive training will be implemented for bedridden patients to prevent the occurrence of deep vein thrombosis in the lower limbs. Summary of the invention

[0005] In view of the problems and shortcomings of the prior art, the present invention provides an ankle pump training monitoring and identification system based on a deep learning model of electromyographic signals.

[0006] The present invention solves the above technical problems through the following technical solutions:

[0007] The present invention provides an ankle pump training monitoring and identification system based on a deep learning model of electromyographic signals, which is characterized in that it comprises a fixed bracket, a clamping bracket for clamping on a footboard is installed on the vertical part of the fixed bracket, an arc-shaped installation cavity with a mounting protrusion fixedly extended at the front end is fixed on the horizontal part of the fixed bracket, a first driving component is fixed to the bottom of the inner cavity of the arc-shaped installation cavity, a calf support arc base matching the shape of the calf of a human body is fixed at the rear end of the arc-shaped installation cavity, an arc-shaped rotating seat with a rotating protrusion fixedly extended at the front end and a rotating shell fixed on the top outer side of the arc-shaped rotating seat are arranged in the arc-shaped installation cavity, the first driving component is connected to the bottom of the arc-shaped rotating seat, a first rotating shaft is fixed to the rear end outer side of the rotating protrusion, the first rotating shaft is penetrated by the mounting protrusion, an ankle sole bearing seat is arranged in the arc-shaped rotating seat, and the ankle sole bearing seat is rotationally connected to the arc-shaped rotating seat, a second driving component is installed in the rotating shell, the second driving component is connected to the ankle sole bearing seat, and an angle sensor is installed on the ankle sole bearing seat;

[0008] The system also includes a controller and an electromyographic sensor for attaching to the calf of a bedridden patient;

[0009] The controller is used to pre-store the identity information and ankle pump passive training program of each bedridden patient, match the input information with the pre-stored identity information to match the target bedridden patient, and obtain the ankle pump passive training program of the target bedridden patient as the target ankle pump passive training program, the target ankle pump passive training program includes ankle flexion and extension training and its training time and ankle left-right swing training and its training time, the ankle flexion and extension training includes the ankle flexion and extension speed and angle, and the ankle left-right swing training includes the ankle swing speed and angle;

[0010] The controller is also used to control the first drive component to start based on the target ankle pump passive training program, drive the arc-shaped rotating seat to rotate, drive the ankle and foot support seat to rotate accordingly, and the ankle of the target bedridden patient on the ankle and foot support seat is swung left and right accordingly, and the training speed and training angle fed back by the angle sensor are received; control the second drive component to start, drive the ankle and foot support seat to move forward and backward, and the ankle of the target bedridden patient on the ankle and foot support seat is flexed and extended accordingly, and the training speed and training angle fed back by the angle sensor are received;

[0011] The controller is also used to receive the electromyographic signal of the calf of the target bedridden patient detected by the electromyographic sensor within a set time at the beginning of training, preprocess the electromyographic signal, perform feature extraction on the preprocessed electromyographic signal, obtain electromyographic time domain features and electromyographic frequency domain features, form one-dimensional features, and input the one-dimensional features into a target deep vein thrombosis risk identification model constructed based on the deep learning model for training to identify whether the target bedridden patient has a risk of deep vein thrombosis in the lower limbs, increase the intensity of ankle flexion and extension training and the intensity of ankle left and right swing training sometimes, and maintain the target ankle pump passive training program for training when not in use.

[0012] The positive and progressive effects of the present invention are:

[0013] Compared with the existing ankle pump training structure, the ankle pump training structure of the present invention is more compact and flexible in structure design and occupies less space. Such a small ankle pump training structure can not only realize the forward and backward extension and flexion movement of the ankle, but also realize the left and right swinging movement of the ankle.

[0014] The present invention can be hung on the footboard of a bed, and the ankles and soles of the bedridden patient are placed on the ankle and sole supporting seats, and the calves are placed on the calf supporting arc base. Under the driving action, the ankle and sole supporting seats can move, so that the ankles of the bedridden patient can perform passive ankle pump training, which can not only realize the forward and backward extension and flexion movement of the ankle, but also the left and right swinging movement, thereby contracting the calf muscle groups of the bedridden patient, so that the blood in the deep veins of the lower limbs can circulate faster, promote venous return, and prevent the occurrence of venous thrombosis.

[0015] Moreover, during the ankle pump training, it is necessary to monitor whether bedridden patients are at risk of deep vein thrombosis in the lower limbs. If so, further intensive training is implemented on bedridden patients to prevent the actual occurrence of the risk of deep vein thrombosis in the lower limbs. If not, the original training is implemented on bedridden patients to prevent the occurrence of the risk of deep vein thrombosis in the lower limbs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1-7 Schematic diagram of the structure of the ankle pump training monitoring and identification system according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] like Figure 1-7As shown, this embodiment provides an ankle pump training monitoring and identification system based on a deep learning model of electromyographic signals, which includes a fixed bracket 10, a clamping bracket 20 for clamping on the footboard is installed on the vertical portion of the fixed bracket 10, and an arc-shaped installation cavity 30 is fixed on the horizontal portion of the fixed bracket 10. Specifically, the fixed bracket 10 includes support columns 11 located at four corner positions, a first connecting rod 12 is fixed between each two adjacent support columns of the four support columns 11, and the two support columns 11 located at the front end are respectively extended upward and fixed with vertical tubes 13, and the arc-shaped installation cavity 30 is fixed on the two opposite first connecting rods 12 on the left and right. The clip-on frame 20 includes a first clip-on frame body 21 and a second clip-on frame body 22. The first clip-on frame body 21 and the second clip-on frame body 22 are both formed by sequentially splicing a long vertical rod extending upward, a transverse rod extending forward and a short vertical rod extending downward into one piece. The long vertical rods of the first clip-on frame body 21 and the second clip-on frame body 22 are both inserted into the corresponding vertical tubes 13 and the bottom ends of the long vertical rods are clamped in the corresponding vertical tubes 13 to form a stretchable structure. A second connecting rod 23 is fixed between the first clip-on frame body 21 and the second clip-on frame body 22.

[0019] A mounting protrusion 31 is fixedly extended at the front end of the arc-shaped mounting cavity 30, a mounting seat 32 is fixed at the bottom of the inner cavity of the arc-shaped mounting cavity, a first drive motor 33 is fixed in the mounting seat 32, a first gear 34 is fixed after the output end of the first drive motor 33 passes through the mounting seat 32, a calf support arc base 40 matching the shape of the human calf is fixed at the rear end of the arc-shaped mounting cavity 30, an arc-shaped rotating seat 50 and a rotating shell 51 fixed on the top outer side of the arc-shaped rotating seat 50 are arranged in the arc-shaped mounting cavity 30, a rotating protrusion 52 is fixedly extended at the front end of the arc-shaped rotating seat 50, an arc-shaped rotating block 53 is fixed at the bottom of the arc-shaped rotating seat 50, an arc-shaped rack 54 matching the first gear 34 is provided at the bottom of the arc-shaped rotating block 53, a first rotating shaft 55 is fixed to the outer side of the rear end of the rotating protrusion 52, and the first rotating shaft 55 passes through the mounting protrusion 31.

[0020] A second drive motor 56 and a small box 57 are installed in the rotating shell 51, and a second gear is fixed to the end of the output shaft of the second drive motor 56. A second rotating shaft is connected to the inner shaft of the small box, and a third gear is fixed on the second rotating shaft. The second gear is meshed with the third gear. One end of the second rotating shaft extends out of the small box 57, and a fourth gear 58 is fixed to the extended end of the second rotating shaft. The fourth gear 58 is meshed with the fifth gear 59. The fifth gear 59 is fixed to the outside of one side of the ankle and foot support seat 60 through a third rotating shaft 61 that passes through one side of the arc-shaped rotating seat 50, and a fourth rotating shaft 62 is fixed to the outside of the other side of the ankle and foot support seat 60, and the fourth rotating shaft 62 passes through the other side of the arc-shaped rotating seat 50. The ankle and foot support seat 60 is placed in the arc-shaped rotating seat 50, and an angle sensor is installed on the ankle and foot support seat 60.

[0021] The system also includes a controller and an electromyographic sensor (not shown) for bedridden patients to attach to the calf. In this embodiment, there are multiple electromyographic sensors, which are attached to different positions of the calf to form a multi-channel electromyographic sensor.

[0022] When in use, the clamping frame 20 is clamped on the footboard of the bed, and the long vertical rods of the first clamping frame body 21 and the second clamping frame body 22 extend relative to the corresponding vertical tube 13. The lengths of the long vertical rods of the first clamping frame body 21 and the second clamping frame body 22 can be set so that when the maximum stretching length is reached, the clamping frame 20 is hung on the footboard of the bed or the support column 11 is just supported on the bed. The ankles and soles of the bedridden patient are placed on the ankle and sole bearing seat 60, and the calves are placed on the calf support arc base 40.

[0023] The controller is used to pre-store the identity information and ankle pump passive training plan of each bedridden patient, match the input information with the pre-stored identity information to match the target bedridden patient, and obtain the ankle pump passive training plan of the target bedridden patient as the target ankle pump passive training plan. The target ankle pump passive training plan includes ankle flexion and extension training and its training time and ankle left-right swing training and its training time. The ankle flexion and extension training includes ankle flexion and extension speed and angle, and the ankle left-right swing training includes ankle swing speed and angle.

[0024] The controller is also used to control the first drive motor 33 to start and drive the first gear 34 to rotate based on the target ankle pump passive training plan, so that the arc rack 54 meshing with the first gear 34 drives the arc rotating seat 50 to rotate, and then drives the ankle and foot support seat 60 to rotate accordingly, so that the ankle of the target bedridden patient on the ankle and foot support seat 60 is swung left and right for training, and the training speed and training angle fed back by the angle sensor are received during the left and right swinging training; control the second drive motor 56 to start and drive the second gear to rotate, and the third gear meshing with the second gear and the second rotating shaft rotate accordingly, and the fourth gear 58 on the second rotating shaft is linked, so that the fifth gear 59 meshing with the fourth gear 58 rotates, driving the ankle and foot support seat 60 to move forward and backward, so that the ankle of the target bedridden patient on the ankle and foot support seat 60 is flexed and extended accordingly, and the training speed and training angle fed back by the angle sensor are received during the flexion and extension training.

[0025] The controller is also used to receive multi-channel electromyographic signals of the calf of the target bedridden patient detected by the multi-channel electromyographic sensor within the set time at the beginning of training, perform noise reduction filtering and artifact removal preprocessing on each channel electromyographic signal, perform feature extraction on each channel electromyographic signal after preprocessing, obtain electromyographic time domain features and electromyographic frequency domain features of each channel, form multidimensional features corresponding to the number of channels, input the multidimensional features into the target deep vein thrombosis risk identification model constructed based on the deep learning model for training, so as to identify whether the target bedridden patient has the risk of lower limb deep vein thrombosis, sometimes increase the ankle flexion and extension training intensity (such as increasing the ankle flexion and extension speed and / or increasing the ankle flexion and extension angle) and its training time and the ankle left and right swing training intensity (such as increasing the ankle left and right swing speed and / or increasing the ankle left and right swing angle) and its training time, and the increased ankle flexion and extension training time is greater than the increased ankle left and right swing training time, and when there is no risk, maintain the target ankle pump passive training plan for training.

[0026] In this program, increasing the intensity of ankle flexion and extension training can further promote the contraction of the calf muscle groups of bedridden patients, so that the blood in the deep veins of the lower limbs can circulate faster and promote venous return, thereby preventing the actual occurrence of deep vein thrombosis in the lower limbs.

[0027] Specifically, for the preprocessed electromyographic signals of each channel, the preprocessed electromyographic signals of the channel are divided according to the angle fed back by the angle sensor, and the preprocessed electromyographic signals corresponding to an ankle flexion and extension movement or an ankle left-right swinging movement are taken as a signal segment to be processed, and it is determined whether the amplitude of any signal segment to be processed is within a preset amplitude range. If yes, the signal segment to be processed is retained, and if no, the signal segment to be processed is discarded. The signal segments to be processed retained by the channel are spliced ​​in time sequence to obtain the spliced ​​electromyographic signals of the channel, thereby obtaining the spliced ​​electromyographic signals of each channel, and feature extraction is performed on the spliced ​​electromyographic signals of each channel to obtain the electromyographic time domain features and electromyographic frequency domain features of each channel, forming a multidimensional feature corresponding to the number of channels.

[0028] A more optimized implementation method is: determine whether the amplitude of any signal segment to be processed is within the preset amplitude range, retain the signal segment to be processed if it is, and use it as a signal segment to be discarded if it is not, count the number of reserved segments of the signal segments to be processed reserved by the channel, and when the number of reserved segments reaches the preset number of segments, discard the signal segment to be discarded corresponding to the channel; when the number of reserved segments does not reach the preset number of segments, for the signal segment to be discarded that is both the reserved signal segment to be processed and the discarded signal segment, use the average amplitude value of the signal segment to be processed that is reserved before and after as the amplitude of the signal segment to be discarded, and change the signal segment to be discarded to the reserved signal segment. For the remaining signal segments to be processed, continue to count the number of reserved signal segments to be processed of the channel. When the number of reserved segments reaches the preset number, discard the remaining signal segments to be discarded corresponding to the channel. When the number of reserved segments does not reach the preset number, for the consecutive signal segments to be discarded, retain the signal segment to be discarded with the smallest amplitude and discard the remaining signal segments to be discarded, and use the average amplitude of the signal segments to be processed retained before and after the consecutive signal segments to be discarded as the amplitude of the retained signal segment to be discarded with the smallest amplitude, until the counted number of reserved signal segments to be processed retained by the channel reaches the preset number.

[0029] In this scheme, the occurrence of signal segments to be discarded is divided into two situations: one is that there are signal segments to be processed that are reserved before and after, with a signal segment to be discarded in the middle; the other is that there are signal segments to be processed that are reserved before and after, with several consecutive signal segments to be discarded in the middle. When the number of reserved segments does not reach the preset number, the signal segments to be discarded in the situation where there are signal segments to be processed that are reserved before and after, with a signal segment to be discarded in the middle are processed first. After processing, if the number of reserved segments reaches the preset number, the other situation is no longer processed. When the number of reserved segments after processing does not reach the preset number, the signal segments to be discarded in the situation where there are signal segments to be processed that are reserved before and after, with several consecutive signal segments to be discarded in the middle are processed until the number of reserved segments after processing reaches the preset number.

[0030] The target deep vein thrombosis risk identification model is constructed as follows: the controller is also used to obtain multi-channel electromyographic signals of the calf detected by the multi-channel electromyographic sensors of multiple historical ankle pump passive trainers during ankle pump passive training, as well as the risk of lower limb deep vein thrombosis for each historical ankle pump passive trainer, and some historical ankle pump passive trainers have both risks of lower limb deep vein thrombosis and those who do not have risks of lower limb deep vein thrombosis; the electromyographic signals of each historical ankle pump passive trainer are preprocessed with noise reduction filtering and artifact removal, and the features of each channel electromyographic signal after preprocessing are extracted to obtain the electromyographic time of each channel. The myoelectric feature set is constructed by using the multidimensional features of each historical ankle pump passive trainer and the myoelectric feature set is divided into an myoelectric feature training set and an myoelectric feature verification set. A deep vein thrombosis risk identification model is constructed based on the deep learning model. The myoelectric feature training set is input into the deep vein thrombosis risk identification model for model training, and the myoelectric feature verification set is input into the trained deep vein thrombosis risk identification model for model verification. The corresponding deep vein thrombosis risk identification model is used as the target deep vein thrombosis risk identification model for model verification.

[0031] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that these are only examples, and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. An ankle pump training monitoring and recognition system based on a deep learning model of electromyographic signals, characterized in that: The invention relates to a fixing bracket, wherein a clamping bracket for fixing on a footboard of a bed is installed on a vertical part of the fixing bracket, a horizontal part of the fixing bracket is fixed with an arc-shaped mounting cavity with a mounting protrusion fixedly extending from the front end, a first driving component is fixed to the bottom of the inner cavity of the arc-shaped mounting cavity, a calf supporting arc base matching the shape of a calf of a human body is fixed to the rear end of the arc-shaped mounting cavity, an arc-shaped rotating seat with a rotating protrusion fixedly extending from the front end and a rotating shell fixed to the top outer side of the arc-shaped rotating seat are arranged in the arc-shaped mounting cavity, the first driving component is connected to the bottom of the arc-shaped rotating seat, a first rotating shaft is fixed to the rear outer side of the rotating protrusion, the first rotating shaft is penetrated by the mounting protrusion, an ankle and foot sole bearing seat is arranged in the arc-shaped rotating seat, and the ankle and foot sole bearing seat is rotationally connected with the arc-shaped rotating seat, a second driving component is installed in the rotating shell, the second driving component is connected to the ankle and foot sole bearing seat, and an angle sensor is installed on the ankle and foot sole bearing seat; The system also includes a controller and an electromyographic sensor for attaching to the calf of a bedridden patient; The controller is used to pre-store the identity information and ankle pump passive training program of each bedridden patient, match the input information with the pre-stored identity information to match the target bedridden patient, and obtain the ankle pump passive training program of the target bedridden patient as the target ankle pump passive training program, the target ankle pump passive training program includes ankle flexion and extension training and its training time and ankle left-right swing training and its training time, the ankle flexion and extension training includes the ankle flexion and extension speed and angle, and the ankle left-right swing training includes the ankle swing speed and angle; The controller is also used to control the first drive component to start based on the target ankle pump passive training program, drive the arc-shaped rotating seat to rotate, drive the ankle and foot support seat to rotate accordingly, and the ankle of the target bedridden patient on the ankle and foot support seat is swung left and right accordingly, and the training speed and training angle fed back by the angle sensor are received; control the second drive component to start, drive the ankle and foot support seat to move forward and backward, and the ankle of the target bedridden patient on the ankle and foot support seat is flexed and extended accordingly, and the training speed and training angle fed back by the angle sensor are received; The controller is also used to receive the electromyographic signal of the calf of the target bedridden patient detected by the electromyographic sensor within a set time at the beginning of training, preprocess the electromyographic signal, perform feature extraction on the preprocessed electromyographic signal, obtain electromyographic time domain features and electromyographic frequency domain features, form one-dimensional features, and input the one-dimensional features into a target deep vein thrombosis risk identification model constructed based on the deep learning model for training to identify whether the target bedridden patient has a risk of deep vein thrombosis in the lower limbs, increase the intensity of ankle flexion and extension training and the intensity of ankle left and right swing training sometimes, and maintain the target ankle pump passive training program for training when not in use.

2. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals according to claim 1, characterized in that: The controller is also used to increase the ankle flexion and extension training intensity and training time and the ankle left-right swing training intensity and training time when it is identified that the target bedridden patient has a risk of lower limb deep vein thrombosis, and the increased ankle flexion and extension training time is greater than the increased ankle left-right swing training time.

3. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals according to claim 1, characterized in that: There are multiple electromyographic sensors, which are attached to different positions of the calf to form a multi-channel electromyographic sensor. The controller is also used to receive multi-channel electromyographic signals of the calf of a target bedridden patient detected by a multi-channel electromyographic sensor within a set time at the beginning of training, perform noise reduction filtering and artifact removal preprocessing on each channel electromyographic signal, perform feature extraction on each channel electromyographic signal after preprocessing, obtain electromyographic time domain features and electromyographic frequency domain features of each channel, form multidimensional features corresponding to the number of channels, and input the multidimensional features into a target deep vein thrombosis risk identification model for training to identify whether the target bedridden patient is at risk of lower limb deep vein thrombosis.

4. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals as claimed in claim 3, characterized in that: The controller is also used to divide the preprocessed electromyographic signals of each channel according to the angle fed back by the angle sensor, and use the preprocessed electromyographic signals corresponding to an ankle flexion and extension movement or an ankle left-right swinging movement as a signal segment to be processed, and judge whether the amplitude of any signal segment to be processed is within a preset amplitude range, retain the signal segment to be processed when the answer is yes, and discard the signal segment to be processed when the answer is no, and splice the signal segments to be processed retained by the channel in time sequence to obtain the spliced ​​electromyographic signals of the channel, thereby obtaining the spliced ​​electromyographic signals of each channel, and perform feature extraction on the spliced ​​electromyographic signals of each channel to obtain the electromyographic time domain features and electromyographic frequency domain features of each channel, so as to form multidimensional features corresponding to the number of channels.

5. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals as claimed in claim 4, characterized in that: Determine whether the amplitude of any signal segment to be processed is within the preset amplitude range. If yes, retain the signal segment to be processed; if no, treat it as a signal segment to be discarded. Count the number of reserved segments of the signal segments to be processed reserved by the channel. When the number of reserved segments reaches the preset number, discard the signal segment to be discarded corresponding to the channel. When the number of reserved segments does not reach the preset number, for the signal segment to be discarded that is both the reserved signal segment to be processed and the discarded signal segment, use the average amplitude of the reserved signal segments to be processed and the discarded signal segment is changed to the reserved signal segment. The number of signal segments to be processed retained by the channel is continued to be counted; when the number of reserved segments reaches the preset number, the remaining signal segments to be discarded corresponding to the channel are discarded; when the number of reserved segments does not reach the preset number, for the consecutive signal segments to be discarded, the signal segment to be discarded with the smallest amplitude is retained and the remaining signal segments to be discarded are discarded, and the average amplitude of the signal segments to be processed retained before and after is used as the amplitude of the signal segment to be discarded with the smallest amplitude, until the counted number of signal segments to be processed retained by the channel reaches the preset number.

6. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals according to claim 1, characterized in that: The controller is also used to obtain the electromyographic signals of the calves detected by the electromyographic sensors of multiple historical ankle pump passive trainers during ankle pump passive training, as well as the risk of deep vein thrombosis of the lower limbs for each historical ankle pump passive trainer, including those who have both the risk of deep vein thrombosis of the lower limbs and those who do not have the risk of deep vein thrombosis of the lower limbs among the historical ankle pump passive trainers, preprocess the electromyographic signals of each historical ankle pump passive trainer, perform feature extraction on the preprocessed electromyographic signals, obtain electromyographic time domain features and electromyographic frequency domain features, form one-dimensional features, construct an electromyographic feature set using the one-dimensional features corresponding to each historical ankle pump passive trainer, divide the electromyographic feature set into an electromyographic feature training set and an electromyographic feature verification set, construct a deep vein thrombosis risk identification model based on a deep learning model, input the electromyographic feature training set into the deep vein thrombosis risk identification model for model training, input the electromyographic feature verification set into the trained deep vein thrombosis risk identification model for model verification, and use the corresponding deep vein thrombosis risk identification model as the target deep vein thrombosis risk identification model for model verification.

7. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals according to claim 1, characterized in that: The first driving assembly includes a first driving motor, a first gear and an arc-shaped rack. The first driving motor is fixed to the bottom of the inner cavity of the arc-shaped installation cavity. The output end of the first driving motor is fixed with a first gear. The bottom of the arc-shaped rotating seat is fixed with an arc-shaped rotating block. The bottom of the arc-shaped rotating block is provided with an arc-shaped rack matching the first gear. The controller is used to control the start-up of the first drive motor to drive the first gear to rotate, so that the arc-shaped rack meshing with the first gear drives the arc-shaped rotating seat to rotate, and then drives the ankle and foot support seat to rotate accordingly, so that the ankle of the target bedridden patient on the ankle and foot support seat is swung left and right for training.

8. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals according to claim 7, characterized in that: A mounting seat is fixed at the bottom of the inner cavity of the arc-shaped mounting cavity, the first drive motor is fixed in the mounting seat, and a first gear is fixed to the output end of the first drive motor after passing through the mounting seat.

9. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals according to claim 1, characterized in that: The second driving assembly includes a second driving motor and a small box body fixed in a rotating shell, a second gear is fixed to the end of the output shaft of the second driving motor, a second rotating shaft is axially connected to the small box body, a third gear is fixed to the second rotating shaft, the second gear is meshed with the third gear, one end of the second rotating shaft extends out of the small box body, a fourth gear is fixed to the extended end of the second rotating shaft, the fourth gear is meshed with a fifth gear, the fifth gear is fixed to the outside of one side of the ankle and foot support seat through the third rotating shaft passing through one side of the arc-shaped rotating seat, a fourth rotating shaft is fixed to the outside of the other side of the ankle and foot support seat, and the fourth rotating shaft passes through the other side of the arc-shaped rotating seat; The controller is used to control the start of the second drive motor to drive the second gear to rotate, and the third gear meshing with the second gear and the second rotating shaft rotate accordingly. The fourth gear on the second rotating shaft is linked to rotate the fifth gear meshing with the fourth gear, driving the ankle and foot support seat to move forward and backward, so that the ankle of the target bedridden patient on the ankle and foot support seat can be flexed and extended.

10. The ankle pump training monitoring and identification system based on the deep learning model of electromyographic signals according to claim 1, characterized in that: The fixed bracket includes support columns located at four corner positions, a first connecting rod is fixed between every two adjacent support columns among the four support columns, and two support columns located at the front end are respectively extended upward and fixed with vertical pipes; The clamping frame includes a first clamping frame body and a second clamping frame body, and the first clamping frame body and the second clamping frame body are both composed of a long vertical rod extending upward, a transverse rod extending forward and a short vertical rod extending downward, which are sequentially spliced ​​together. The long vertical rods of the first clamping frame body and the second clamping frame body are both inserted into corresponding vertical tubes and the bottom ends of the long vertical rods are clamped in the corresponding vertical tubes to form a stretchable structure, and a second connecting rod is fixed between the first clamping frame body and the second clamping frame body.

Images