Wrist motion recognition and sensing integrated system and method based on solar energy

Abstract

The invention provides a wrist action recognition and sensing integrated system based on solar energy and a wrist action recognition method. The system integrates at least one photovoltaic power generation device, an energy management module, a signal acquisition module and an action recognition module. The photovoltaic power generation device is used for collecting light energy in the environment, converting the light energy into electric energy and storing the electric energy in the energy management module; the energy management module is used for supplying power to other power utilization modules of the system by using electric energy from the photovoltaic power generation device; the signal acquisition module is used for acquiring a voltage signal and a current signal when the photovoltaic power generation device works, and providing the voltage signal and the current signal as original signals for wrist action recognition to the action recognition module; and the action recognition module is used for processing the original signal provided by the signal acquisition module and recognizing the wrist action. According to the invention, the use of a traditional inertial measurement sensor is cancelled, the power consumption can be effectively reduced, and the integration of energy collection and motion sensing is realized.

Description

Technical Field

[0001] This invention belongs to the field of wearable devices, specifically relating to a solar-powered wrist motion recognition and sensing system and method. Background Technology

[0002] Precise wrist movement recognition is a key technology for promoting national fitness, revolutionizing human-computer interaction, and providing personalized healthcare. In promoting national fitness, accurate monitoring of movement status through wrist movement recognition, and the provision of standardized feedback, can improve exercise effectiveness and prevent sports injuries. In revolutionizing human-computer interaction, it enables natural "gesture-based command" control, reducing screen dependence and making operation more intuitive and convenient. In providing personalized healthcare, wrist movement recognition can accurately quantify and assess hand function in conditions such as Parkinson's tremor and stroke rehabilitation, enabling remote and objective monitoring and evaluation of rehabilitation training and treatment effectiveness.

[0003] Current wrist movement recognition systems rely on externally powered inertial measurement units (IMUs) (such as accelerometers and gyroscopes) to collect raw motion data. The typical workflow involves transmitting sensor data wirelessly via Bluetooth or other protocols to an external device like a computer for offline processing and classification using machine learning models. This approach suffers from significant issues in terms of energy consumption and real-time performance, necessitating optimization and improvement. Summary of the Invention

[0004] In view of this, the present invention provides a solar-powered wrist motion recognition system and method that integrates energy harvesting and motion sensing. It utilizes the impact of wrist movements on photovoltaic power generation to capture the voltage / current ripple characteristics of the photovoltaic power generation output caused by the movement in real time, using this as the sensing signal for wrist motion recognition. This invention eliminates the use of traditional inertial measurement sensors, effectively reducing power consumption, and achieves the integration of energy harvesting and motion sensing.

[0005] To solve the above-mentioned technical problems, the present invention is implemented as follows.

[0006] A solar-powered wrist motion recognition and sensing system integrates the effects of wrist movements on photovoltaic power generation, capturing in real time the output voltage and current ripple characteristics of the photovoltaic power generation caused by the movements, and using these as sensing signals for wrist motion recognition. The system includes at least one integrated photovoltaic power generation device, an energy management module, a signal acquisition module, and a motion recognition module. The photovoltaic power generation device is used to collect light energy from the environment and convert it into electrical energy for storage in the energy management module; The energy management module is used to supply power to other power-consuming modules in the system using electrical energy from the photovoltaic power generation device; The signal acquisition module is used to acquire the voltage and current signals of the photovoltaic power generation device when it is working, and provide them to the action recognition module as the raw signals for wrist action recognition. The motion recognition module is used to process the raw signals provided by the signal acquisition module and to recognize wrist movements.

[0007] Preferably, the action recognition module includes a signal processing unit and a network model unit; The signal processing unit includes a preprocessing unit, a window segmentation unit, and a Fourier transform unit. The preprocessing unit normalizes and filters the voltage and current signals from the signal acquisition module, and then performs sliding window segmentation through the window segmentation unit to obtain segmented time-domain voltage and current signals. The time-domain voltage and current signals are converted to the frequency domain by the Fourier transform unit to obtain frequency-domain voltage and current signals. The time-domain voltage, time-domain current, frequency-domain voltage, and frequency-domain current signals are a total of four sensing signals, which serve as a multi-dimensional representation of wrist movements. The network model unit extracts features based on the four sensing signals, performs feature fusion and classification, and obtains wrist movement recognition results.

[0008] Preferably, the network model of the network model unit includes four parallel processing channels for time-domain voltage signals, time-domain current signals, frequency-domain voltage signals, and frequency-domain current signals, as well as a splicing and fusion unit and a fully connected layer; Each processing channel includes a convolutional neural network (CNN), an attention mechanism layer, and a pooling layer connected in sequence. The CNN extracts local signal features through a multi-scale convolutional structure, the attention mechanism layer is used to assign dynamic weights to key action segments to enhance the representation of weakly discriminative action signals, and finally the output is processed by the pooling layer. The features output from the four processing channels are spliced ​​and fused by the splicing and fusion unit, and then the wrist motion recognition results are output through the fully connected layer.

[0009] Preferably, the network model unit is a lightweight network model unit that uses structured channel pruning technology to remove redundant convolutional kernels, and performs feature transfer training through the knowledge distillation process of teacher network-student network; the pruning ratio and the temperature coefficient of knowledge distillation are adjusted according to the system application scenario.

[0010] Preferably, the window segmentation unit selects one or two action cycles as the sliding window length and sets a 50% overlap rate for sliding window segmentation.

[0011] Preferably, the system further includes a wireless communication module for wirelessly transmitting the wrist motion recognition results obtained through the motion recognition module.

[0012] Preferably, the energy management module, signal acquisition module, motion recognition module, and wireless communication module are integrated on a flexible circuit board; the photovoltaic power generation device and the flexible circuit board are packaged on a flexible substrate.

[0013] Preferably, the energy management module employs a maximum power point tracking algorithm.

[0014] Preferably, the motion recognition module is deployed on a microprocessor; the microprocessor is an embedded chip for wearable scenarios.

[0015] This invention also provides a solar-powered wrist movement recognition method, employing the aforementioned solar-powered wrist movement recognition sensor integration system. The method includes: Step 1: Under the set sampling frequency, different experimental objects are made to perform different actions at a stable frequency and amplitude indoors and outdoors. Only the signal acquisition module is used to collect the voltage and current signals of the solar cells in the photovoltaic power generation device and add action tags. Step 2: The signal processing unit in the motion recognition module processes the data from the voltage and current channels obtained by the signal acquisition module and obtains the frequency domain signals corresponding to the voltage and current through Fourier transform; the time domain voltage signal, time domain current signal, frequency domain voltage signal, and frequency domain current signal are respectively divided into sliding windows to obtain four sensing signals, which serve as a multi-dimensional representation of wrist movements; the multi-dimensional representation and the motion label constitute the motion sample; Step 3: Train the network model units in the action recognition module using action samples; the network model units adopt a network model with convolutional neural network and attention mechanism, and learn the mapping relationship between the multi-dimensional representation and wrist action based on the action samples; Step 4: After the network model is trained, based on the requirements of recognition accuracy and real-time performance for different application scenarios, the network model is lightweighted using pruning techniques and knowledge distillation, and then deployed to the action recognition module. Step 5: The system, trained using a network model, collects light energy from the environment and converts it into electrical energy to power other power-consuming modules in the system; at the same time, it collects voltage and current signals from the photovoltaic power generation device during operation and provides them to the motion recognition module, which then outputs the wrist motion recognition results.

[0016] Beneficial effects: (1) This invention utilizes the influence of wrist movements on photovoltaic power generation to capture the output voltage / current ripple characteristics of photovoltaic power generation caused by the movements in real time, and uses this as a sensing signal for wrist movement recognition. This invention eliminates the use of traditional inertial measurement sensors, effectively reduces power consumption, and achieves the integration of energy harvesting and movement sensing.

[0017] (2) Due to the limited range of wrist movement, voltage and current sensing signals are characterized by strong time-varying nature, unstable amplitude, and weak differences between different actions, posing significant challenges to action recognition tasks. Therefore, in a preferred embodiment, this invention extracts features such as amplitude, frequency, and energy distribution in the time and frequency domains respectively during the sensing signal processing stage, constructs a multi-dimensional highly discriminative feature space, solves the problem of insufficient single-dimensional representation ability, overcomes the challenges of unstable sensing signal amplitude caused by energy value changes and weak differences between different actions, and improves the accuracy of wrist action recognition.

[0018] (3) The wrist motion recognition and sensing system of the present invention integrates a photovoltaic power generation device, an energy management module, and a lightweight motion recognition module into one unit. The flexible photovoltaic power generation device continuously converts ambient light energy into electrical energy, replacing the traditional battery power supply mode, and designs an energy management circuit to execute the maximum power point algorithm, thereby improving the power generation efficiency of the photovoltaic power generation device and providing stable power supply for the system.

[0019] (4) In a preferred scheme, a multi-scale convolutional structure is used to extract local signal features, and a dynamic weight is assigned to key action segments through an attention mechanism to enhance the representation of weakly discriminative action signals. The model is deployed on a microprocessor through algorithm lightweighting, thereby replacing the traditional wireless transmission and offline processing methods, and has advantages such as reduced power consumption and real-time recognition. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the wrist motion recognition and transmission integration system in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the wrist motion recognition and transmission integration system in Embodiment 1 of the present invention. Figure 3 This is a schematic diagram of the signal acquisition module in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the energy management module in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the action recognition module in Embodiment 1 of the present invention; Among them, 101-flexible photovoltaic power generation device (back of wrist), 102-flexible photovoltaic power generation device (side of wrist), 103-flexible photovoltaic power generation device (side of wrist), 2-flexible circuit board, 3-flexible substrate. Detailed Implementation

[0021] This invention provides a solar-powered wrist motion recognition system that integrates energy harvesting and motion sensing. The core idea is that the system utilizes the impact of wrist movements on photovoltaic power generation. While the photovoltaic system powers the system, it also captures the output voltage and current ripple characteristics of the photovoltaic power generation triggered by the movement in real time, using these as sensing signals for motion recognition. This invention eliminates the use of traditional inertial measurement sensors, effectively reducing power consumption and achieving a unified system of energy harvesting and motion sensing.

[0022] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] Example 1 This embodiment provides a solar-powered wrist motion recognition and sensing system, such as... Figure 1 As shown, this system is illustrated using a wristband device as an example. Combined with... Figure 2 The system integrates photovoltaic power generation devices 101-103, an energy management module, a signal acquisition module, and a motion recognition module. In this embodiment, the system also includes a wireless communication module. The energy management module, signal acquisition module, motion recognition module, and wireless communication module are all integrated on the flexible circuit board 2.

[0024] Photovoltaic power generation devices 101-103 collect light energy from the environment based on the photoelectric effect of flexible solar cells, convert it into electrical energy, and store it in an energy management module.

[0025] In a preferred embodiment, the flexible solar cells of the photovoltaic power generation devices 101-103 are evenly arranged on the back and sides of the wristband corresponding to the wrist, and are connected in parallel to ensure power generation conversion efficiency while reducing the hot spot effect caused by series connection.

[0026] Alternatively, the photovoltaic power generation device 101-103 may also be equipped with three or more flexible solar cells.

[0027] Preferably, the flexible solar cells of the photovoltaic power generation devices 101-103 should be selected with a thickness of less than 2mm and a bending radius of less than 50mm to ensure the wearability of the device and avoid breakage or performance degradation.

[0028] Furthermore, the photovoltaic power generation device 101-103 and the flexible circuit board 2 are encapsulated using a flexible substrate 3, which ensures good conductivity while improving wearability.

[0029] The energy management module is used to power other electrical modules in the system using electricity from the photovoltaic power generation device. For example... Figure 4As shown, the energy management module includes an MPPT execution unit, a buck-boost unit, energy storage elements, and an energy storage management unit. The energy management module uses the Maximum Power Point Tracking (MPPT) algorithm to maintain the solar cells of the photovoltaic power generation device at their maximum power point voltage and stores electrical energy in the energy storage elements, thereby powering the signal acquisition module and the motion recognition module. The MPPT algorithm can employ the open-circuit voltage method to reduce energy consumption. By sampling the open-circuit voltage at fixed periods, the maximum power point operating voltage is calculated, and the operating voltage of the solar cells follows this maximum power point operating voltage.

[0030] The charging and discharging thresholds of the energy storage element are set by the energy storage management unit. When the charging voltage of the energy storage element reaches the threshold, it discharges; when the discharging voltage reaches the threshold, it charges. Optionally, the energy storage element can be a supercapacitor.

[0031] The signal acquisition module is used to acquire the voltage and current signals of the photovoltaic power generation device during operation, and provide them as the raw signals for wrist motion recognition to the motion recognition module. For example... Figure 3 As shown, the signal acquisition module includes a voltage acquisition unit and a current acquisition unit. The voltage and current acquisition units acquire the real-time operating voltage and current of the photovoltaic power generation device 101-103. Different wrist movements have varying impacts on the power generation performance of the photovoltaic power generation device 101-103; therefore, the voltage and current signals themselves can serve as sensing signals reflecting wrist movements.

[0032] Optionally, to fully acquire signals from various wrist movements, the sampling frequency of the signal acquisition module is set to 50-200Hz during operation, preferably 100Hz.

[0033] The motion recognition module is used to process the raw signals provided by the signal acquisition module and to recognize wrist movements. See also... Figure 2 The motion recognition module comprises a signal processing unit and a network model unit. The signal processing unit processes data from two channels—voltage and current—to obtain four sensing signals in the time and frequency domains, which are then transmitted to the network model unit. The network model unit extracts features from each of the four sensing signals, performs feature fusion and concatenation, and then classifies the signals to obtain the wrist motion recognition result.

[0034] Due to the limited range of wrist movement, voltage and current sensing signals are characterized by strong time-varying properties, unstable amplitude, and subtle differences between different movements, posing significant challenges to action recognition tasks. Traditional signal processing methods often rely solely on time-domain features (such as statistical quantities like mean, variance, and maximum values) for modeling. These methods are easily affected by noise when faced with non-stationary signals and struggle to extract essential features from subtle differences between movements, resulting in limited classification performance. To address this issue, the signal processing unit of this invention normalizes and filters the acquired raw voltage and current signals, then performs a Fourier transform to obtain the frequency domain signal. This yields sensing signals in both the time and frequency domains, constructing a multi-dimensional, highly discriminative feature space. This enhances the model's ability to recognize subtle differences in movements and improves the system's robustness in unstable signal environments.

[0035] Therefore, as Figure 5 As shown, the signal processing unit includes a preprocessing unit, a window segmentation unit, and a Fourier transform unit. The preprocessing unit normalizes and filters the voltage and current signals from the signal acquisition module to obtain processed time-domain voltage and current signals. The window segmentation unit segments the time-domain data into windows, with each window ideally encompassing the complete wrist movement cycle. The segmented time-domain voltage and current signals are then converted to the frequency domain by the Fourier transform unit to obtain frequency-domain voltage and current signals. These four sensing signals—time-domain voltage, time-domain current, frequency-domain voltage, and frequency-domain current—serve as a multi-dimensional representation of wrist movement and are input to the network model unit.

[0036] The network model unit is used for efficient action recognition based on the multi-dimensional representation of wrist movements from the signal processing unit. (Still...) Figure 5 As shown, the network model unit structure includes four parallel processing channels, which process time-domain voltage signals, time-domain current signals, frequency-domain voltage signals, and frequency-domain current signals, respectively. The network model unit also includes a splicing and fusion unit and a fully connected layer.

[0037] Each processing channel consists of a convolutional neural network (CNN), an attention layer, and a pooling layer connected in sequence. The CNN extracts local signal features of the corresponding modality through multi-scale convolutional structures, with the time-domain processing channel focusing on identifying instantaneous changes and the frequency-domain processing channel used to capture the spectral features of periodic actions.

[0038] To enhance the model's ability to focus on key feature segments, an attention mechanism is introduced into the output features of each processing channel before entering the fusion stage, which strengthens the signal segments with the ability to distinguish and discriminate through weighting.

[0039] The features output from the four processing channels are spliced ​​and fused by the splicing and fusion unit, and then the final wrist motion recognition result is output through the fully connected layer.

[0040] Furthermore, the network model units can be lightweighted, including by using structured channel pruning to remove redundant convolutional kernels, resulting in lightweight network model units. These lightweight network model units are then trained through feature transfer using a knowledge distillation process between the teacher and student networks. Depending on the system application scenario, the pruning ratio and the temperature coefficient of knowledge distillation are adjusted to ensure both high accuracy and real-time performance.

[0041] Furthermore, the signal processing unit and lightweight network model unit are deployed on a low-power microprocessor with limited computing power, eliminating the need to rely on cloud computing resources and enabling real-time response. The microprocessor can be an embedded chip used in wearable applications.

[0042] The wireless communication module transmits the motion recognition results from the motion recognition module, and the wrist motion recognition results are displayed on the receiving end. Optionally, the wireless communication module can be a low-power Bluetooth module or a WIFI module.

[0043] In summary, the embodiments of the present invention capture solar energy and convert it into electrical energy to power the entire system. At the same time, by utilizing the influence of wrist movements on power generation performance, the operating voltage and current of the solar cells are used as signals to identify wrist movements, thus achieving the integration of energy sensing and control.

[0044] Example 2 This embodiment provides a wrist movement recognition method based on the above-described sensor-sensor integration system. The method uses the system of Embodiment 1 and specifically includes the following steps: Step 1: Acquisition and calibration of electrical signals: With a set sampling frequency, different experimental subjects were instructed to perform different actions at stable frequency and amplitude indoors and outdoors. The signal acquisition module was used to collect the working voltage and current signals of the solar cells in the photovoltaic power generation device under each action as the raw signals, without the need to install a traditional inertial measurement unit. Then, action labels were manually added to the raw signals according to time periods.

[0045] Step Two: Electrical Signal Processing and Dataset Establishment The signal processing unit in the motion recognition module processes the data from the two channels of voltage and current signals obtained by the signal acquisition module and obtains the frequency domain signals corresponding to the voltage and current through Fourier transform; For each action sample, both its temporal and frequency domain features are preserved. Since wrist movements are limited in amplitude, signal changes are weak, and they are susceptible to external interference, relying solely on the time domain may result in insufficient action discriminative power. Therefore, a time-frequency fusion approach enhances feature representation. Thus, a suitable window length (preferably containing 1-2 action cycles) is selected for the time-domain voltage and current signals, and a 50% overlap rate is set for sliding window segmentation. Each window segment is then subjected to Fourier transform to obtain the frequency-domain voltage and current signals. These four sensing signals—time-domain voltage, time-domain current, frequency-domain voltage, and frequency-domain current—serve as a multi-dimensional representation of wrist movements. This multi-dimensional representation, along with action labels, constitutes the action sample. This approach improves data utilization while maintaining the integrity of sample information. By using simple cross-validation to divide the action samples into training and validation sets, a wrist action dataset for building machine learning models can be established.

[0046] Step 3: Machine Learning Model Training The network model units in the action recognition module are trained using action samples; the network model units adopt a network model with convolutional neural network and attention mechanism, and learn the mapping relationship between the multi-dimensional representation and wrist action based on the action samples.

[0047] Step 4: Lightweighting the machine learning model: After the model training is completed, pruning techniques and knowledge distillation are flexibly used to reduce the model and network model according to the requirements of recognition accuracy and real-time performance in different application scenarios. The model is then converted into a microprocessor-executable format and capacity before being deployed on the microprocessor.

[0048] In scenarios such as health monitoring and posture reminders, the accuracy requirements for action recognition are relatively relaxed, and the system can adopt a larger proportion of pruning to significantly improve real-time performance. However, in scenarios such as sports training and rehabilitation assessment, which require higher accuracy in action classification, a gentler pruning ratio is adopted. At the same time, by adjusting the temperature coefficient of knowledge distillation, etc., high accuracy is ensured while also taking into account real-time recognition.

[0049] Step 5: Motion Perception The system, trained using a network model, collects ambient light energy and converts it into electrical energy to power other electrical modules within the system. Simultaneously, it collects voltage and current signals from the photovoltaic power generation device and provides them to the motion recognition module, which outputs wrist motion recognition results. A wireless communication module transmits the real-time wrist motion recognition results, which are then displayed at the receiving end.

[0050] The specific embodiments described above only illustrate the design principles of the present invention. The shapes and names of the components in this description may differ and are not limited. Therefore, those skilled in the art can modify or make equivalent substitutions to the technical solutions described in the foregoing embodiments; and these modifications and substitutions do not depart from the inventive spirit and technical solutions of the present invention, and should all fall within the protection scope of the present invention.

Claims

1. A solar-powered wrist motion recognition and sensing system, characterized in that, By utilizing the impact of wrist movements on photovoltaic power generation, the system captures the output voltage and current ripple characteristics of the photovoltaic power generation caused by the movements in real time, and uses these as sensing signals for wrist movement recognition. The system includes at least one integrated photovoltaic power generation device, an energy management module, a signal acquisition module, and a movement recognition module. The photovoltaic power generation device is used to collect light energy from the environment and convert it into electrical energy for storage in the energy management module; The energy management module is used to supply power to other power-consuming modules in the system using electrical energy from the photovoltaic power generation device; The signal acquisition module is used to acquire the voltage and current signals of the photovoltaic power generation device when it is working, and provide them to the action recognition module as the raw signals for wrist action recognition. The motion recognition module is used to process the raw signals provided by the signal acquisition module and to recognize wrist movements.

2. The solar-powered wrist motion recognition and sensing system as described in claim 1, characterized in that, The action recognition module includes a signal processing unit and a network model unit; The signal processing unit includes a preprocessing unit, a window segmentation unit, and a Fourier transform unit. The preprocessing unit normalizes and filters the voltage and current signals from the signal acquisition module, and then performs sliding window segmentation through the window segmentation unit to obtain segmented time-domain voltage and current signals. The time-domain voltage and current signals are converted to the frequency domain by the Fourier transform unit to obtain frequency-domain voltage and current signals. Four sensing signals—time-domain voltage signal, time-domain current signal, frequency-domain voltage signal, and frequency-domain current signal—serve as a multi-dimensional representation of wrist movements. The network model unit extracts features based on the four sensing signals, performs feature fusion and classification, and obtains wrist movement recognition results.

3. The solar-powered wrist motion recognition and sensing system as described in claim 2, characterized in that, The network model unit includes four parallel processing channels for time-domain voltage signals, time-domain current signals, frequency-domain voltage signals, and frequency-domain current signals, as well as a splicing and fusion unit and a fully connected layer. Each processing channel includes a convolutional neural network (CNN), an attention mechanism layer, and a pooling layer connected in sequence. The CNN extracts local signal features through a multi-scale convolutional structure, the attention mechanism layer is used to assign dynamic weights to key action segments to enhance the representation of weakly discriminative action signals, and finally the output is processed by the pooling layer. The features output from the four processing channels are spliced ​​and fused by the splicing and fusion unit, and then the wrist motion recognition results are output through the fully connected layer.

4. The solar-powered wrist motion recognition and sensing system as described in claim 2 or 3, characterized in that, The network model unit is a lightweight network model unit that uses structured channel pruning technology to remove redundant convolutional kernels, and performs feature transfer training through the knowledge distillation process of teacher network-student network. Adjust the pruning ratio and the temperature coefficient of knowledge distillation according to the system application scenario.

5. The solar-powered wrist motion recognition and sensing system as described in claim 2, characterized in that, The window segmentation unit selects one or two action cycles as the sliding window length and sets a 50% overlap rate for sliding window segmentation.

6. The solar-powered wrist motion recognition and sensing system as described in claim 1, characterized in that, The system further includes a wireless communication module for wirelessly transmitting the wrist movement recognition results obtained through the motion recognition module.

7. The solar-powered wrist motion recognition and sensing system as described in claim 6, characterized in that, The energy management module, signal acquisition module, motion recognition module, and wireless communication module are integrated on a flexible circuit board; the photovoltaic power generation device and the flexible circuit board are packaged on a flexible substrate.

8. The solar-powered wrist motion recognition and sensing system as described in claim 1, characterized in that, The energy management module uses the maximum power point tracking algorithm.

9. The solar-powered wrist motion recognition and sensing system as described in claim 1, characterized in that, The motion recognition module is deployed on a microprocessor; the microprocessor is an embedded chip for wearable scenarios.

10. A solar-powered wrist motion recognition method, employing the solar-powered wrist motion recognition sensor integration system as described in any one of claims 1-9, characterized in that, The method includes: Step 1: Under the set sampling frequency, different experimental objects are made to perform different actions at a stable frequency and amplitude indoors and outdoors. Only the signal acquisition module is used to collect the voltage and current signals of the solar cells in the photovoltaic power generation device and add action tags. Step 2: The signal processing unit in the motion recognition module processes the data from the voltage and current channels obtained by the signal acquisition module and obtains the frequency domain signals corresponding to the voltage and current through Fourier transform; the time domain voltage signal, time domain current signal, frequency domain voltage signal, and frequency domain current signal are respectively divided into sliding windows to obtain four sensing signals, which serve as a multi-dimensional representation of wrist movements; the multi-dimensional representation and the motion label constitute the motion sample; Step 3: Train the network model units in the action recognition module using action samples; the network model units adopt a network model with convolutional neural network and attention mechanism, and learn the mapping relationship between the multi-dimensional representation and wrist action based on the action samples; Step 4: After the network model is trained, based on the requirements of recognition accuracy and real-time performance for different application scenarios, the network model is lightweighted using pruning techniques and knowledge distillation, and then deployed to the action recognition module. Step 5: The system, trained using a network model, collects light energy from the environment and converts it into electrical energy to power other power-consuming modules in the system; at the same time, it collects voltage and current signals from the photovoltaic power generation device during operation and provides them to the motion recognition module, which then outputs the wrist motion recognition results.

Images