Intelligent speed control system and method of underwater treadmill based on deep learning

Abstract

The present invention discloses an underwater treadmill intelligent speed regulation system and method based on deep learning, the method comprising: obtaining the heart rate and image of the user; using an image processing algorithm to perform motion posture analysis on the image to obtain the motion posture of the user; using the heart rate and the motion posture as input data, inputting them into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the motion state of the user on the underwater treadmill; inputting the motion state into a trained underwater treadmill speed regulation model to obtain the optimal training water flow speed of the user on the underwater treadmill; and controlling the current water flow speed of the underwater treadmill to adjust to the optimal training water flow speed. The present invention can dynamically adjust the water flow speed according to the real-time physiological state and sports performance of the athlete, effectively improving the training effect of underwater sports.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an underwater treadmill intelligent speed regulation system and method based on deep learning. Background Art

[0002] In the field of modern fitness and rehabilitation, underwater treadmills have become the preferred tool for athlete training and rehabilitation treatment due to their low impact and high efficiency. However, existing technologies have obvious limitations in achieving intelligent speed regulation of underwater treadmills. Specifically, most underwater treadmills use preset training modes and manually adjust the water flow speed. They cannot respond to the athlete's physiological state changes and sports performance requirements in real time, and lack personalization and intelligence.

[0003] Therefore, the existing intelligent speed control systems of underwater treadmills generally lack real-time monitoring, in-depth analysis and adaptive speed control capabilities, which limits training efficiency and safety, and cannot meet the athletes' needs for personalized and intelligent training, so they cannot effectively improve the training effect of underwater sports. Summary of the invention

[0004] The embodiment of the present invention provides an underwater treadmill intelligent speed control system and method based on deep learning, which can dynamically adjust the water flow speed according to the real-time physiological state and athletic performance of the athlete, thereby effectively improving the training effect of underwater sports.

[0005] An embodiment of the present invention provides an underwater treadmill intelligent speed control system based on deep learning, comprising:

[0006] A heart rate detection device, used to detect the heart rate of a user exercising on an underwater treadmill;

[0007] A camera, used to capture an image of the user;

[0008] A control device is connected to the heart rate detection device and the camera for:

[0009] acquiring the heart rate and the image;

[0010] Performing motion posture analysis on the image using an image processing algorithm to obtain the motion posture of the user;

[0011] The heart rate and the movement posture are used as input data and input into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the movement state of the user on the underwater treadmill;

[0012] Inputting the motion state into a trained underwater treadmill speed regulation model to obtain the optimal training water flow speed of the user on the underwater treadmill;

[0013] The current water flow speed of the underwater treadmill is controlled to be adjusted to the optimal training water flow speed.

[0014] As an improvement of the above solution, the heart rate and the motion posture are used as input data and input into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the motion state of the user on the underwater treadmill, including:

[0015] Extracting features of the heart rate and the movement posture to obtain heart rate features and movement posture features;

[0016] Calculate the attention weights of the heart rate feature and the motion posture feature and perform feature splicing to obtain an attention-weighted fusion feature;

[0017] The fusion features are input into a constructed multi-layer perceptron neural network model with an attention mechanism, and the motion state of the user on the underwater treadmill is output through multi-layer nonlinear transformation.

[0018] As an improvement of the above solution, the attention weight calculation and feature splicing of the heart rate feature and the motion posture feature to obtain the attention-weighted fusion feature includes:

[0019] Use linear transformation to transform the heart rate feature vector O HR and motion posture feature vector O STA Convert to key vector and value vector;

[0020] The heart rate feature vector O is calculated by the following formula HR and motion posture feature vector O STA The attention weight w i :

[0021] Among them, Q k , K i , V i Represent the query vector, key vector and value vector respectively, which come from the heart rate feature vector O HR and motion posture feature vector O STA Linear transformation of k is the dimension of the key vector, which is used to scale the dot product result to avoid gradient disappearance;

[0022] The feature vector v after attention weighting is obtained by feature concatenation using the following formula:

[0023] N is the heart rate feature vector O HR and motion posture feature vector O STA The linearly transformed value vector V i The number of .

[0024] As an improvement of the above scheme, the multi-layer perceptron neural network model with attention mechanism consists of multiple hidden layers, and the calculation formula of each hidden layer l is:

[0025] h l =f(W l ·v+b l ), l=1,...,L

[0026] Among them, h l-1 is the output of the previous layer, W l and b l are the weight matrix and bias vector of the lth layer, respectively, f is the nonlinear activation function, and the output of the last layer L is used as the final layer h L The final layer h L The calculation formula is:

[0027] S = g(W S ·h L +b S )W S and b S is the final layer h L The weight matrix and bias vector, g

[0028] is the output activation function, and S is the output motion state of the user on the underwater treadmill.

[0029] As an improvement of the above scheme, the model formula of the underwater treadmill speed regulation model is:

[0030]

[0031] Among them, V opt is the output of the optimal training water flow speed of the user currently on the underwater treadmill; S is the motion state; is a deep learning model with parameters Θ, including the weight matrix W h , W f and the bias vector b h 、b f ; σ(·) is the activation function of the output layer, and tanh(·) is the activation function of the hidden layer, which is used to introduce nonlinear transformation and enhance the expressive power of the model.

[0032] Another embodiment of the present invention provides an underwater treadmill intelligent speed regulation method based on deep learning, which is applied to the underwater treadmill intelligent speed regulation system based on deep learning as described above, including:

[0033] acquiring the heart rate and the image;

[0034] Performing motion posture analysis on the image using an image processing algorithm to obtain the motion posture of the user;

[0035] The heart rate and the movement posture are used as input data and input into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the movement state of the user on the underwater treadmill;

[0036] Inputting the motion state into a trained underwater treadmill speed regulation model to obtain the optimal training water flow speed of the user on the underwater treadmill;

[0037] The current water flow speed of the underwater treadmill is controlled to be adjusted to the optimal training water flow speed.

[0038] As an improvement of the above solution, the heart rate and the motion posture are used as input data and input into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the motion state of the user on the underwater treadmill, including:

[0039] Extracting features of the heart rate and the movement posture to obtain heart rate features and movement posture features;

[0040] Calculate the attention weights of the heart rate feature and the motion posture feature and perform feature splicing to obtain an attention-weighted fusion feature;

[0041] The fusion features are input into a constructed multi-layer perceptron neural network model with an attention mechanism, and the motion state of the user on the underwater treadmill is output through multi-layer nonlinear transformation.

[0042] As an improvement of the above solution, the attention weight calculation and feature splicing of the heart rate feature and the motion posture feature to obtain the attention-weighted fusion feature includes:

[0043] Use linear transformation to transform the heart rate feature vector O HR and motion posture feature vector O STA Convert to key vector and value vector;

[0044] The heart rate feature vector O is calculated by the following formula HR and motion posture feature vector O STA The attention weight w i :

[0045] Among them, Q k , K i , V i Represent the query vector, key vector and value vector respectively, which come from the heart rate feature vector O HR and motion posture feature vector O STALinear transformation of k is the dimension of the key vector, which is used to scale the dot product result to avoid gradient disappearance;

[0046] The feature vector ν is obtained by concatenating the features using the following formula:

[0047] N is the heart rate feature vector O HR and motion posture feature vector O STA The linearly transformed value vector V i The number of .

[0048] As an improvement of the above scheme, the multi-layer perceptron neural network model with attention mechanism consists of multiple hidden layers, and the calculation formula of each hidden layer l is:

[0049] h l =f(W l ·v+b l ),l=1,...,L

[0050] Among them, h l-1 is the output of the previous layer, W l and b l are the weight matrix and bias vector of the lth layer, respectively, f is the nonlinear activation function, and the output of the last layer L is used as the final layer h L The final layer h L The calculation formula is:

[0051] S = g(W S ·h L +b S )W S and b S is the final layer h L The weight matrix and bias vector, g

[0052] is the output activation function, and S is the output motion state of the user on the underwater treadmill.

[0053] As an improvement of the above scheme, the model formula of the underwater treadmill speed regulation model is:

[0054]

[0055] Among them, V opt is the output of the optimal training water flow speed of the user currently on the underwater treadmill; S is the motion state; is a deep learning model with parameters Θ, including the weight matrix W h , W f and the bias vector b h 、b f; σ(·) is the activation function of the output layer, and tanh(·) is the activation function of the hidden layer, which is used to introduce nonlinear transformation and enhance the expressive power of the model.

[0056] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0057] The embodiment of the present invention captures the athlete's heart rate changes and movement posture in real time, fuses the heart rate data with the movement posture features, and sends them as input to the trained multi-layer perceptron neural network model with attention mechanism, which can deeply analyze the athlete's movement state; the obtained movement state information is then input into the specially trained underwater treadmill speed regulation model, which uses deep learning technology to learn and predict the current optimal training water flow speed to achieve personalized speed regulation. Finally, according to the current optimal training water flow speed, the water flow speed of the underwater treadmill is adjusted in real time to ensure that the athlete trains in the best state, which not only improves the training efficiency but also ensures the safety of the training. From the above analysis, it can be seen that the embodiment of the present invention solves the problem that the speed regulation strategy in the prior art is too static and lacks personalization, and can dynamically adjust the water flow speed according to the athlete's real-time physiological state and sports performance, effectively improving the training effect of underwater sports. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the architecture of an underwater treadmill intelligent speed regulation method system based on deep learning provided by an embodiment of the present invention;

[0059] Figure 2 It is a flow chart of an underwater treadmill intelligent speed regulation method based on deep learning provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 creative work are within the scope of protection of the present invention.

[0061] See also Figure 1, is a schematic diagram of the architecture of an underwater treadmill intelligent speed regulation method system based on deep learning provided by an embodiment of the present invention. The underwater treadmill intelligent speed regulation system based on deep learning includes: a heart rate detection device 10, a camera 11 and a control device 12; the heart rate detection device 10 is used to detect the heart rate of a user who is exercising on the underwater treadmill. The heart rate detection device 10 can be an intelligent wearable device with a heart rate detection function. When the user exercises on the underwater turret, the device can be worn on the body, for example, the user can wear a smart watch with a heart rate detection function on the hand; the camera 11 is used to capture the image of the user who is exercising on the underwater treadmill. The camera 11 can be waterproof and set in the water, and aimed at the underwater treadmill at the same time; the control device 12 (such as a desktop computer or a control motherboard for controlling the underwater treadmill, etc.) is connected to the heart rate detection device and the camera in communication (such as a wired connection or a wireless communication connection), and is used to:

[0062] acquiring the heart rate and the image;

[0063] Performing motion posture analysis on the image using an image processing algorithm to obtain the motion posture of the user;

[0064] The heart rate and the movement posture are used as input data and input into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the movement state of the user on the underwater treadmill;

[0065] Inputting the motion state into a trained underwater treadmill speed regulation model to obtain the optimal training water flow speed of the user on the underwater treadmill;

[0066] The current water flow speed of the underwater treadmill is controlled to be adjusted to the optimal training water flow speed.

[0067] The embodiment of the present invention captures the athlete's heart rate changes and movement posture in real time, fuses the heart rate data with the movement posture features, and sends them as input to the trained multi-layer perceptron neural network model with attention mechanism, which can deeply analyze the athlete's movement state; the obtained movement state information is then input into the specially trained underwater treadmill speed regulation model, which uses deep learning technology to learn and predict the current optimal training water flow speed to achieve personalized speed regulation. Finally, according to the current optimal training water flow speed, the water flow speed of the underwater treadmill is adjusted in real time to ensure that the athlete trains in the best state, which not only improves the training efficiency but also ensures the safety of the training. From the above analysis, it can be seen that the embodiment of the present invention solves the problem that the speed regulation strategy in the prior art is too static and lacks personalization, and can dynamically adjust the water flow speed according to the athlete's real-time physiological state and sports performance, effectively improving the training effect of underwater sports.

[0068] It should be noted that the image processing algorithm used to perform motion posture analysis on the image may be an existing technology, such as OpenPose, DeepLab, Hourglass Network, etc.

[0069] Understandably, underwater treadmills (or underwater treadmills) are usually equipped with a water return system that can adjust the speed and direction of the water flow, thereby affecting the runner's resistance and training intensity. Here is the basic principle of adjusting the water flow rate: The water return system of an underwater treadmill is designed to circulate water and generate water flow. The system may include a water pump, pipes, nozzles, valves and filters. The water is pumped from the bottom of the pool, filtered to remove impurities, and then re-injected into the pool through pipes and nozzles to form a water flow. The key to controlling the water flow rate is to adjust the power of the water pump and / or change the path of the water flow. This can be achieved in the following ways:

[0070] Adjust the speed of the water pump motor to change the water flow and pressure.

[0071] Use variable area nozzles or valves to change the cross-sectional area of ​​the water flow, thereby affecting the flow rate.

[0072] Use guide plates or baffles to change the direction of water flow, affecting the density and speed of the water flow in front of the runner.

[0073] In one embodiment, the heart rate and the motion posture are input as input data into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the motion state of the user on the underwater treadmill, including:

[0074] Extracting features of the heart rate and the movement posture to obtain heart rate features and movement posture features;

[0075] Calculate the attention weights of the heart rate feature and the motion posture feature and perform feature splicing to obtain an attention-weighted fusion feature;

[0076] The fusion features are input into a constructed multi-layer perceptron neural network model with an attention mechanism, and the motion state of the user on the underwater treadmill is output through multi-layer nonlinear transformation.

[0077] In this embodiment, through deep learning technology, especially the multi-layer perceptron neural network model with attention mechanism, the heart rate data and motion posture data obtained in real time are deeply analyzed to obtain the precise motion state of the athlete on the underwater treadmill. Specifically, first, the original heart rate data and motion posture data are pre-processed to extract the key features that can reflect the physiological state and motion performance of the athlete. This stage aims to convert complex data into a form that is easy to handle by the machine learning model to ensure the accuracy of subsequent analysis. Next, the heart rate features and motion posture features are weighted using the attention mechanism to highlight the feature information that is most critical to the current motion state. Through the calculation of the attention weight, the model can independently determine which features are more important in the current situation, thereby performing intelligent screening and fusion of features. This weighted feature splicing not only enhances the model's ability to understand complex motion scenes, but also improves the distinguishability of features, laying a solid foundation for subsequent motion state prediction. Finally, the fusion features after attention weighting are input into the pre-trained multi-layer perceptron neural network model with attention mechanism. Through multiple layers of nonlinear transformations, the model can capture the complex correlations between features, thereby outputting the athlete's movement status on the underwater treadmill, including but not limited to status indicators in multiple dimensions such as fatigue level and movement efficiency.

[0078] As an improvement of the above solution, the attention weight calculation and feature splicing of the heart rate feature and the motion posture feature to obtain the attention-weighted fusion feature includes:

[0079] Use linear transformation to transform the heart rate feature vector O HR and motion posture feature vector O STA Convert to key vector and value vector;

[0080] The heart rate feature vector O is calculated by the following formula HR and motion posture feature vector O STA The attention weight w i :

[0081] Among them, Q k , K i , V i Represent the query vector, key vector and value vector respectively, which come from the heart rate feature vector O HR and motion posture feature vector O STA Linear transformation of k is the dimension of the key vector, which is used to scale the dot product result to avoid gradient disappearance;

[0082] The feature vector v after attention weighting is obtained by feature concatenation using the following formula:

[0083] N is the heart rate feature vector O HR and motion posture feature vector O STA The linearly transformed value vector V i The number of .

[0084] In this embodiment, the deep fusion of heart rate features and motion posture features is realized by utilizing the attention mechanism to enhance the accuracy and personalization of the model's understanding of the athlete's motion state. Specifically: First, the extracted heart rate feature vector and motion posture feature vector are converted into key vectors (Keys) and value vectors (Values) through linear transformation. This process is intended to lay the foundation for the subsequent attention mechanism calculation, so that the model can process feature information from different sources in a unified manner. Then, by calculating the dot product between the query vector (Query) and the key vector (Keys), and after scaling and Softmax function processing, the attention weights of the heart rate feature vector and the motion posture feature vector are obtained. Among them, the query vector can be regarded as the focus of the model on the current task, and the key vector represents the potential meaning of the feature vector. This calculation process enables the model to adaptively adjust the degree of attention to different features, thereby giving more important features higher weights when the features are fused. Finally, according to the calculated attention weight, the value vector is weighted and summed to obtain the feature vector after attention weighting. This weighted fusion process not only enhances the interaction between features, but also ensures that the model can make more comprehensive and accurate decisions from a global perspective. In summary, this embodiment, by introducing the attention mechanism, not only realizes the intelligent fusion of heart rate characteristics and motion posture characteristics, but also significantly improves the performance of the model in the field of athlete motion status analysis.

[0085] As an improvement of the above scheme, the multi-layer perceptron neural network model with attention mechanism consists of multiple hidden layers, and the calculation formula of each hidden layer l is:

[0086] h l =f(W l ·v+b l ), l=1,...,L

[0087] Among them, h l-1 is the output of the previous layer, W l and b l are the weight matrix and bias vector of the lth layer, respectively, f is a nonlinear activation function (such as ReLU or tanh function), and the output of the last layer L is used as the final layer h L The final layer h L The calculation formula is:

[0088] S = g(W S ·h L +bS ), where W S and b S is the final layer h L The weight matrix and bias vector, g is the output activation function (softmax or linear function can be used), and S is the output motion state of the user on the underwater treadmill.

[0089] In this embodiment, a multi-layer perceptron neural network model with an attention mechanism is constructed to deeply process and fuse heart rate features and motion posture features to accurately predict the user's motion state on the underwater treadmill. Among them, the multi-layer perceptron neural network (MLP) model structure consists of multiple hidden layers, each hidden layer is linearly transformed by a weight matrix and a bias vector, and then the expression ability of the model is increased by a nonlinear activation function to learn complex feature mapping relationships. The attention mechanism enables the model to focus on the key parts of the input features by calculating the attention weights, thereby improving the model's sensitivity to important features and learning efficiency. Through attention weighting, the model can adaptively adjust the importance of features, so as to better capture patterns closely related to motion state prediction in a complex feature space. Therefore, by introducing the attention mechanism, this embodiment can automatically screen and learn the most valuable information for motion state prediction from complex heart rate and motion posture features, significantly improving the accuracy of the prediction. At the same time, the deep structure and nonlinear transformation ability of the multi-layer perceptron neural network model enable the model to learn the deep abstract features of the data, so as to maintain good prediction performance under different individuals and environmental conditions. In addition, the attention mechanism enables the model to adaptively adjust the prediction strategy according to different user characteristics and motion scenarios, thereby achieving personalized analysis of each user's motion status.

[0090] As an improvement of the above scheme, the model formula of the underwater treadmill speed regulation model is:

[0091]

[0092] Among them, V opt is the output of the optimal training water flow speed of the user currently on the underwater treadmill; S is the motion state; is a deep learning model with parameters Θ, including the weight matrix W h , W f and the bias vector b h 、b f ; σ(·) is the activation function of the output layer (Sigmoid, ReLU or linear function can be selected), tanh(·) is the activation function of the hidden layer, which is used to introduce nonlinear transformation and enhance the expressive power of the model.

[0093] In this embodiment, a deep learning model is used to construct a special underwater treadmill speed control model to predict the user's current optimal training water flow speed. Among them, the deep learning model is designed to handle complex nonlinear relationships, and can automatically learn and extract key features from the input motion state to predict the optimal training water flow speed. The structure of the model includes a hidden layer and an output layer, wherein the hidden layer introduces the complexity of the model through a nonlinear activation function, enhancing its modeling ability for nonlinear relationships. By predicting the water flow speed that best matches the user's current motion state, the system can provide personalized training conditions for each user, maximizing the training effect while reducing unnecessary physical exertion and injury risks. The real-time adjustment of the optimal water flow speed ensures that the water flow speed is always in the most suitable state during training, which helps to improve the athlete's endurance, speed and skills, thereby improving the overall training efficiency.

[0094] See also Figure 2 , is a flow chart of an underwater treadmill intelligent speed regulation method based on deep learning provided by an embodiment of the present invention. The underwater treadmill intelligent speed regulation method based on deep learning is applied to the underwater treadmill intelligent speed regulation system based on deep learning as described above, including steps S10 to S14:

[0095] S10, acquiring the heart rate and the image;

[0096] S11, using an image processing algorithm to perform motion posture analysis on the image to obtain the motion posture of the user;

[0097] S12, inputting the heart rate and the motion posture as input data into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the motion state of the user on the underwater treadmill;

[0098] S13, inputting the motion state into a trained underwater treadmill speed regulation model to obtain the optimal training water flow speed of the user on the underwater treadmill;

[0099] S14, controlling the current water flow speed of the underwater treadmill to adjust to the optimal training water flow speed.

[0100] The embodiment of the present invention captures the athlete's heart rate changes and movement posture in real time, fuses the heart rate data with the movement posture features, and sends them as input to the trained multi-layer perceptron neural network model with attention mechanism, which can deeply analyze the athlete's movement state; the obtained movement state information is then input into the specially trained underwater treadmill speed regulation model, which uses deep learning technology to learn and predict the current optimal training water flow speed to achieve personalized speed regulation. Finally, according to the current optimal training water flow speed, the water flow speed of the underwater treadmill is adjusted in real time to ensure that the athlete trains in the best state, which not only improves the training efficiency but also ensures the safety of the training. From the above analysis, it can be seen that the embodiment of the present invention solves the problem that the speed regulation strategy in the prior art is too static and lacks personalization, and can dynamically adjust the water flow speed according to the athlete's real-time physiological state and sports performance, effectively improving the training effect of underwater sports.

[0101] As an improvement of the above solution, the heart rate and the motion posture are used as input data and input into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the motion state of the user on the underwater treadmill, including:

[0102] Extracting features of the heart rate and the movement posture to obtain heart rate features and movement posture features;

[0103] Calculate the attention weights of the heart rate feature and the motion posture feature and perform feature splicing to obtain an attention-weighted fusion feature;

[0104] The fusion features are input into a constructed multi-layer perceptron neural network model with an attention mechanism, and the motion state of the user on the underwater treadmill is output through multi-layer nonlinear transformation.

[0105] As an improvement of the above solution, the attention weight calculation and feature splicing of the heart rate feature and the motion posture feature to obtain the attention-weighted fusion feature includes:

[0106] Use linear transformation to transform the heart rate feature vector O HR and motion posture feature vector O STA Convert to key vector and value vector;

[0107] The heart rate feature vector O is calculated by the following formula HR and motion posture feature vector O STA The attention weight w i :

[0108] Among them, Q k , K i , V iRepresent the query vector, key vector and value vector respectively, which come from the heart rate feature vector O HR and motion posture feature vector O STA Linear transformation of k is the dimension of the key vector, which is used to scale the dot product result to avoid gradient disappearance;

[0109] The following formula is used to concatenate features and obtain the attention-weighted feature vector v:

[0110] N is the heart rate feature vector O HR and motion posture feature vector O STA The linearly transformed value vector V i The number of .

[0111] As an improvement of the above scheme, the multi-layer perceptron neural network model with attention mechanism consists of multiple hidden layers, and the calculation formula of each hidden layer l is:

[0112] h l =f(W l ·v+b l ), l=1,...,L

[0113] Among them, h l-1 is the output of the previous layer, W l and b l are the weight matrix and bias vector of the lth layer, respectively, f is the nonlinear activation function, and the output of the last layer L is used as the final layer h L The final layer h L The calculation formula is:

[0114] S = g(W S ·h L +b S )W S and b S is the final layer h L , g is the output activation function, and S is the output motion state of the user on the underwater treadmill.

[0115] As an improvement of the above scheme, the model formula of the underwater treadmill speed regulation model is:

[0116]

[0117] Among them, V opt is the output of the optimal training water flow speed of the user currently on the underwater treadmill; S is the motion state; is a deep learning model with parameters Θ, including the weight matrix W h , W fand the bias vector b h 、b f ; σ(·) is the activation function of the output layer, and tanh(·) is the activation function of the hidden layer, which is used to introduce nonlinear transformation and enhance the expressive power of the model.

[0118] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An underwater treadmill intelligent speed control system based on deep learning, characterized in that: include: A heart rate detection device, used to detect the heart rate of a user exercising on an underwater treadmill; A camera, used to capture an image of the user; A control device is connected to the heart rate detection device and the camera for: acquiring the heart rate and the image; Performing motion posture analysis on the image using an image processing algorithm to obtain the motion posture of the user; The heart rate and the movement posture are used as input data and input into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the movement state of the user on the underwater treadmill; The motion state is input into the trained underwater treadmill speed regulation model to obtain the optimal training water flow speed of the user on the underwater treadmill. The model formula of the underwater treadmill speed regulation model is: Among them, V opt is the output of the optimal training water flow speed of the user currently on the underwater treadmill; S is the motion state; is a deep learning model with parameters Θ, including the weight matrix W h , W f and the bias vector b h 、b f ; σ(·) is the activation function of the output layer, and tanh(·) is the activation function of the hidden layer, which is used to introduce nonlinear transformation and enhance the expressiveness of the model; The current water flow speed of the underwater treadmill is controlled to be adjusted to the optimal training water flow speed.

2. The underwater treadmill intelligent speed control system based on deep learning as claimed in claim 1, characterized in that: The heart rate and the motion posture are input as input data into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the motion state of the user on the underwater treadmill, including: Extracting features of the heart rate and the movement posture to obtain heart rate features and movement posture features; Calculate the attention weights of the heart rate feature and the motion posture feature and perform feature splicing to obtain an attention-weighted fusion feature; The fusion features are input into a constructed multi-layer perceptron neural network model with an attention mechanism, and the motion state of the user on the underwater treadmill is output through multi-layer nonlinear transformation.

3. The underwater treadmill intelligent speed control system based on deep learning as claimed in claim 2, characterized in that: The step of calculating the attention weights of the heart rate feature and the motion posture feature and performing feature splicing to obtain the attention-weighted fusion feature includes: Use linear transformation to transform the heart rate feature vector O HR and motion posture feature vector O STA Convert to key vector and value vector; The heart rate feature vector O is calculated by the following formula HR and motion posture feature vector O STA The attention weight w i : Among them, Q k , K i , V i Represent the query vector, key vector and value vector respectively, which come from the heart rate feature vector O HR and motion posture feature vector O STA Linear transformation of k is the dimension of the key vector, which is used to scale the dot product result to avoid gradient disappearance; The feature vector v after attention weighting is obtained by feature concatenation using the following formula: N is the heart rate feature vector O HR and motion posture feature vector O STA The linearly transformed value vector V i The number of .

4. The underwater treadmill intelligent speed control system based on deep learning as claimed in claim 3, characterized in that: The multi-layer perceptron neural network model with attention mechanism consists of multiple hidden layers, and the calculation formula of each hidden layer l is: h l =f(W l ·v+b l ),l=1,...,L Among them, h l-1 is the output of the previous layer, W l and b l are the weight matrix and bias vector of the lth layer, respectively, f is the nonlinear activation function, and the output of the last layer L is used as the final layer h L The final layer h L The calculation formula is: S = g(W S ·h L +b S ), where W S and b S is the final layer h L , g is the output activation function, and S is the output motion state of the user on the underwater treadmill.

5. A method for intelligent speed regulation of an underwater treadmill based on deep learning, applied to the intelligent speed regulation system of an underwater treadmill based on deep learning as claimed in any one of claims 1 to 4, characterized in that: include: acquiring the heart rate and the image; Performing motion posture analysis on the image using an image processing algorithm to obtain the motion posture of the user; The heart rate and the movement posture are used as input data and input into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the movement state of the user on the underwater treadmill; Inputting the motion state into a trained underwater treadmill speed regulation model to obtain the optimal training water flow speed of the user on the underwater treadmill; The current water flow speed of the underwater treadmill is controlled to be adjusted to the optimal training water flow speed.

6. The underwater treadmill intelligent speed regulation method based on deep learning as claimed in claim 5, characterized in that: The heart rate and the motion posture are input as input data into a trained multi-layer perceptron neural network model with an attention mechanism to obtain the motion state of the user on the underwater treadmill, including: Extracting features of the heart rate and the movement posture to obtain heart rate features and movement posture features; Calculate the attention weights of the heart rate feature and the motion posture feature and perform feature splicing to obtain an attention-weighted fusion feature; The fusion features are input into a constructed multi-layer perceptron neural network model with an attention mechanism, and the motion state of the user on the underwater treadmill is output through multi-layer nonlinear transformation.

7. The underwater treadmill intelligent speed regulation method based on deep learning as claimed in claim 6, characterized in that: The step of calculating the attention weights of the heart rate feature and the motion posture feature and performing feature splicing to obtain the attention-weighted fusion feature includes: Use linear transformation to transform the heart rate feature vector O HR and motion posture feature vector O STA Convert to key vector and value vector; The heart rate feature vector O is calculated by the following formula HR and motion posture feature vector O STA The attention weight w i : Among them, Q k , K i , V i Represent the query vector, key vector and value vector respectively, which come from the heart rate feature vector O HR and motion posture feature vector O STA Linear transformation of k is the dimension of the key vector, which is used to scale the dot product result to avoid gradient disappearance; The feature vector υ is obtained by concatenating the features using the following formula: N is the heart rate feature vector O HR and motion posture feature vector O STA The linearly transformed value vector V i The number of .

8. The underwater treadmill intelligent speed regulation method based on deep learning as claimed in claim 7, characterized in that: The multi-layer perceptron neural network model with attention mechanism consists of multiple hidden layers, and the calculation formula of each hidden layer l is: h l =f(W l ·v+b l ),l=1,...,L Among them, h l-1 is the output of the previous layer, W l and b l are the weight matrix and bias vector of the lth layer, respectively, f is the nonlinear activation function, and the output of the last layer L is used as the final layer h L The final layer h L The calculation formula is: S = g(W S ·h L +b S ), where W S and b S is the final layer h L , g is the output activation function, and S is the output motion state of the user on the underwater treadmill.

Images