Lower limb underwater rehabilitation training exoskeleton system

Abstract

The invention discloses a lower limb underwater rehabilitation training exoskeleton system in the technical field of rehabilitation. The lower limb underwater rehabilitation training exoskeleton system comprises exoskeleton hardware and a control system, exoskeleton hardware comprises a waterproof sealing box, a central control and energy system, hip / thigh / knee joint shank exoskeletons and the like, a waterproof design and a telescopic rod structure are adopted, different heights are adapted, and underwater operation safety is guaranteed; the control system comprises an intention analysis module, a multi-level control strategy module and a rehabilitation evaluation system, signals are collected through a multi-source sensor, motion intentions are accurately recognized after preprocessing, feature extraction and dimension reduction, bionic motion output is achieved by combining man-machine cooperative impedance control and three-ring cascade PID control, training indexes are dynamically evaluated, and parameters are adjusted. According to the lower limb rehabilitation training device, the buoyancy, resistance and hydrostatic pressure characteristics of water can be effectively utilized, a rehabilitation environment combining weight reduction training, muscle strength strengthening and circulation promotion is provided for a patient, and the application scene and intervention means of lower limb rehabilitation training are greatly expanded.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation technology, and in particular to an underwater rehabilitation training exoskeleton system for the lower limbs. Background Technology

[0002] Patients with lower limb motor dysfunction often experience gait abnormalities, decreased muscle strength, and impaired balance and coordination, severely impacting their daily living abilities and quality of life. Improving the effectiveness of rehabilitation training has become a pressing issue in clinical rehabilitation. In recent years, hydrotherapy rehabilitation techniques have emerged. Compared to traditional land-based rehabilitation techniques, these techniques utilize the buoyancy and viscous resistance of water to reduce impact and pressure on joints and soft tissues, providing variable-direction and adjustable "natural resistance" throughout the entire range of motion, which is beneficial for the gradual enhancement of lower limb muscle strength and endurance. Simultaneously, hydrostatic pressure helps promote venous return and peripheral circulation, improving respiratory and cardiopulmonary function. These advantages make underwater rehabilitation training a promising clinical application in relieving pain and improving motor function.

[0003] Existing underwater rehabilitation systems mainly include rehabilitation hydrotherapy pools, underwater treadmills, underwater exercise bikes, and wearable monitoring platforms. These systems primarily focus on providing a controlled hydrotherapy environment, basic exercise patterns, and vital sign monitoring functions. Although these underwater rehabilitation systems have made some progress in buoyancy utilization and fluid resistance training, they still have the following shortcomings when applied to practical clinical applications: (1) The current systematic research on underwater exoskeletons is insufficient, and there is a lack of mature and reliable integrated waterproof design schemes, which makes it difficult to guarantee the sealing and electrical safety of the equipment during long-term operation in the underwater environment.

[0004] (2) Existing underwater rehabilitation systems, such as underwater treadmills, are mostly fixed structures, which severely restrict patients' free movement space and multi-dimensional training possibilities, making it difficult to meet the needs of lower limb complex motor function reconstruction.

[0005] (3) At the control level, most underwater rehabilitation systems are still mainly passive rehabilitation modes, lacking accurate recognition of the user's movement intentions and collaborative assistance mechanisms. Although some devices can perform simple speed or resistance adjustments, they cannot respond in real time according to the patient's subjective exertion and gait phase, affecting the patient's training experience and neuromuscular remodeling effect.

[0006] (4) Existing underwater rehabilitation assessments rely heavily on video recordings or visual image analysis for kinematic evaluation, but overall lack systematic integration and quantitative analysis of multi-source information on mechanics, physiology and kinematics, making it difficult to form a complete objective assessment system.

[0007] Therefore, there is an urgent need to develop an underwater rehabilitation exoskeleton robot system with high waterproof reliability, large degrees of freedom of movement, accurate movement intention recognition capability, and a complete multimodal assessment mechanism in an underwater environment, so as to better meet the clinical rehabilitation needs of patients with lower limb motor dysfunction. Summary of the Invention

[0008] To address the aforementioned problems in existing technologies, this invention provides a lower limb underwater rehabilitation training exoskeleton system that combines a rigid lower limb support structure, an underwater adaptive drive mechanism, and the ability to perceive patient physiological information. This underwater rehabilitation assistive lower limb exoskeleton system can provide precise assistance, dynamically adjust training intensity, and ensure safety based on the patient's subjective participation and changes in motor ability, thereby meeting the needs of patients with lower limb movement disorders to improve the effectiveness of underwater rehabilitation training.

[0009] The technical solution of the present invention is as follows: A lower limb underwater rehabilitation training exoskeleton system includes exoskeleton hardware and a control system; the exoskeleton hardware includes a waterproof sealed box 1, a central control and energy system 2, a hip exoskeleton 3, a thigh exoskeleton 4, a knee joint lower leg exoskeleton 5, a suspension bone 6, and a bone sling 7. The central control and energy system 2 includes a main control computing module, a multi-channel signal acquisition unit, a high-energy-density battery pack, and a display screen. The high-energy-density battery pack powers the main control computing module. The multi-channel signal acquisition unit is connected to torque sensors, electromyographic electrodes, and a six-axis inertial measurement unit (IMU) mounted on the hip exoskeleton 3, thigh exoskeleton 4, and knee and lower leg exoskeleton 5, and transmits the data from the torque sensors, electromyographic electrodes, and the six-axis IMU to the main control computing module. The display screen is connected to the main control computing module and displays the data from the multi-channel signal acquisition unit in real time. The waterproof sealing box 1 has a cubic structure and is made of waterproof material. The central control and energy system 2 is located inside the waterproof sealing box 1. When the waterproof sealing box 1 is placed in water, water cannot enter the interior of the waterproof sealing box 1. The hip exoskeleton 3 includes a flexion-extension joint motor 8, an abduction-adduction joint motor 9, a hip torque sensor, a hip six-axis inertial measurement unit (IMU), and a hip joint linkage 10. The hip joint linkage 10 has an L-shaped structure; the vertical part of the L-shaped structure is called the hip joint linkage vertical beam 1001, and the horizontal part of the L-shaped structure is called the hip joint linkage horizontal beam 1002. The flexion-extension joint motor 8 is connected to the end of the hip joint linkage horizontal beam 1002, and the output shaft of the flexion-extension joint motor 8 is parallel to the horizontal plane. When the flexion-extension joint motor 8 rotates, the height of the hip joint linkage vertical beam 1001 changes. The abduction-adduction joint motor 9 is connected to the hip joint linkage horizontal beam 1002. The end of the hip joint is connected, and the output shaft of the abduction and adduction joint motor 9 is perpendicular to the horizontal plane. When the abduction and adduction joint motor 9 rotates, the hip joint link 10 rotates on the horizontal plane. There are two hip torque sensors: one is connected to the flexion and extension joint motor 8 and measures the torque of the flexion and extension joint motor 8, and the other is connected to the abduction and adduction joint motor 9 and measures the torque of the abduction and adduction joint motor 9. The hip six-axis inertial measurement unit (IMU) is set on the hip joint link vertical beam 1001 and measures the angle and angular velocity of the hip joint link vertical beam 1001. There are two hip exoskeletons 3, both connected to the bottom of the waterproof sealed box 1, and the two are mirror-symmetrical. One end of the sling 7 is connected to the hip exoskeleton 3, and the other end is connected to the suspension sling 6. The suspension sling 6 is set on the external track and moves along the external track. When the suspension sling 6 moves, the exoskeleton hardware moves with the suspension sling 7. After the exoskeleton hardware is worn on the human body, the waterproof sealing box 1 is located on the back of the human body, and the end of the hip joint connecting rod vertical beam 1001 fits against the hip joint of the human body. The thigh exoskeleton 4 includes a thigh joint motor 11, a thigh support 12, and a thigh strap. The thigh support 12 is a telescopic rod structure, and its length can be adjusted to accommodate different heights. One end of the thigh support 12 is connected to the thigh joint motor 11, and the other end is connected to the knee joint calf exoskeleton 5. The thigh joint motor 11 is connected to the end of the hip joint connecting rod vertical beam 1001. When the thigh joint motor 11 rotates, the thigh support 12 can perform circular motion around the output shaft of the thigh joint motor 11. The thigh strap is connected to the thigh support 12 and fixes the thigh exoskeleton 4 to the thigh. The part of the thigh strap that contacts the thigh is provided with electromyographic electrode pads, called thigh electromyographic electrode pads, which are used to measure the electromyographic signals of the thigh. The knee-shin exoskeleton 5 includes a calf joint motor 13, a calf torque sensor, a calf support 14, a calf strap, an ankle adjustment mechanism 15, a foot pressure sensing insole 16, and a calf six-axis inertial measurement unit (IMU). The calf support 14 is a telescopic rod structure, which can be adjusted to accommodate different heights. One end of the calf support 14 is connected to the calf joint motor 13, and the other end is connected to the ankle adjustment mechanism 15. The calf joint motor 13 is connected to the thigh support 14. When the calf joint motor 13 rotates, the calf support 14 can rotate around the output shaft of the calf joint motor 13. Circular motion; the calf torque sensor is connected to the calf joint motor 13 to measure the torque of the calf joint motor 13; the foot pressure sensing insole 16 is set on the ankle adjustment mechanism 15 to measure the pressure on the ankle adjustment mechanism 15; the calf strap is connected to the calf support 14 and fixes the knee joint calf exoskeleton 5 to the calf; the part of the calf strap that contacts the calf is provided with electromyographic electrode pads, called calf electromyographic electrode pads, to measure the electromyographic signals of the calf; the calf six-axis inertial measurement unit (IMU) is set on the calf support 14 and measures the angle and angular velocity of the calf support 14; The control system includes an intent parsing module, a multi-level control strategy module, and a rehabilitation assessment system. The intent parsing module receives data from a hip torque sensor, a hip six-axis inertial measurement unit (IMU), a thigh torque sensor, thigh electromyography (EMG) electrodes, a calf torque sensor, a calf EMG electrodes, and a calf six-axis IMU, and parses the human's movement intent. The multi-level control strategy module includes a top-level collaborative control strategy and a bottom-level motion control strategy. The top-level collaborative control strategy converts the movement intent output by the intent parsing module into action commands for the exoskeleton hardware. The bottom-level motion control strategy uses a biomechanical model-based inversion control algorithm to calculate the action commands of the top-level collaborative control strategy into the desired torque or trajectory of each joint, and achieves biomimetic motion output by reading the joint motor current and torque sensor feedback in real time. The rehabilitation assessment system dynamically adjusts the control parameters of the bottom-level motion control strategy.

[0010] Furthermore, the steps for the intent parsing module to parse motion intent are as follows: S1. Acquire joint torque signals using hip torque sensors, thigh torque sensors, and calf torque sensors. Bioelectrical signals were collected using electromyography (EMG) electrodes on the thigh and calf. Joint angles were acquired using a six-axis inertial measurement unit (IMU) at the hip and a six-axis inertial measurement unit (IMU) at the lower leg. and angular velocity The acquired signals are preprocessed as follows: (1-1) Torque signal Water resistance compensation is performed based on the underwater motion velocity v and the joint angular velocity. Calculate the water resistance torque To obtain the net human body's power ,in This is the water resistance coefficient; (1-2) Electromyographic signals A 50Hz notch filter was used to eliminate power frequency interference, and a 20~500Hz bandpass filter was used to remove motion artifacts and underwater electromagnetic noise. Full-wave rectification and a 50ms sliding window RMS calculation were then performed to obtain the electromyography amplitude sequence. ; (1-3) Perform Kalman filtering on the angle and angular velocity signals to eliminate measurement noise caused by water flow disturbance; S2. A linear interpolation method is used to resample all signals to the same frequency, and a hardware timestamp synchronization mechanism is used to ensure that the time alignment error of each mode data is less than 5ms; a sliding time window with a length of 1 second and containing m sampling points is constructed to form a multi-channel timing data matrix. , where n is the total number of sensor channels; S3. Extract the following feature vectors from the preprocessed multi-source signal: (2-1) Based on the joint torque signal Calculate the mean, variance, peak value, and zero-crossing rate of the torque at each joint to obtain the torque characteristic vector. ; (2-2) Based on bioelectrical signals Calculate the root mean square (RMS), average power frequency (MPF), median frequency (MF), and waveform length (WL) for each muscle group to obtain the electromyographic feature vector. ; (2-3) Based on joint angles and angular velocity Calculate the mean, rate of change, and acceleration of the joint angles, as well as the gait cycle duty cycle, to obtain the kinematic feature vector. ; (2-4) Based on the acceleration changes detected by the hip six-axis inertial measurement unit (IMU) and the lower leg six-axis inertial measurement unit (IMU), calculate the rate of change of water resistance coefficient and the buoyancy assist ratio to obtain the environmental feature vector. ; Concatenate the eigenvectors from (2-1) to (2-4) to obtain the original high-dimensional eigenvectors. ; S4. Principal component analysis is used to analyze the original high-dimensional eigenvectors. Dimensionality reduction is performed, retaining the top k principal components with a cumulative contribution rate of 95%, to obtain the dimensionality-reduced fusion feature vector. ; S5. Merge feature vectors Input a pre-trained Support Vector Machine (SVM) classifier, which uses a Radial Basis Function (RBF) kernel function. The mapping relationship between features and intention categories has been established through offline training. The classifier outputs the following 9 types of motion intentions: stationary standing, starting preparation, left leg swing phase, right leg swing phase, left leg support phase, right leg support phase, stop, left turn, and right turn. S6. Based on the identified movement intention and current joint angle The current gait phase is estimated using a finite state machine (FSM), and the gait cycle is divided into 8 sub-phases: initial contact phase, load response phase, mid-single support phase, late-single support phase, pre-swing phase, initial swing phase, mid-swing phase, and late-swing phase. S7. Generate control commands based on movement intention and gait phase. And send it to the multi-level control strategy module; where: M is the target assist mode, and ; The expected assist ratio for the hip joint. The expected assist ratio for the knee joint, and 0 indicates no assist and 1 indicates full assist; This is the water resistance compensation coefficient, and ; The safety limit parameter is the maximum output torque of each joint.

[0011] Furthermore, the top-level collaborative control strategy is as follows: Control commands output by the intent parsing module A human-machine collaborative impedance control algorithm is used to calculate the auxiliary torque output by the exoskeleton. in: To help the proportional coefficient, The expected joint torque is calculated based on a healthy gait database. and For position and velocity gain, and For reference trajectory, This is the feedforward compensation term for water resistance. Depending on the different target assistance mode M, the parameters are set as follows: (3-1) When M is passively driven, the exoskeleton is driven entirely along the preset trajectory. =0, =200 , =200 , =1.0; (3-2) When M is used for assistance, some active control of the human body is retained. =0.3~0.7, =50 , =5 , =0.8; (3-3) M refers to resistance training, where exoskeleton resistance is added to the water resistance. =-0.2, =10 , =2 , =0.3.

[0012] Furthermore, the underlying motion control strategy is as follows: The auxiliary torque calculated by the top-level collaborative control strategy The inverse dynamics model is used to convert the current commands to the motors at each joint: in: For gravity, For Coriolis force and centrifugal force terms, For friction, This is the motor torque constant.

[0013] The drivers for each joint motor employ a three-loop cascaded PID control system consisting of a current loop, a speed loop, and a position loop. The inner current loop has a frequency of 5kHz, the middle speed loop has a frequency of 1kHz, and the outer position loop has a frequency of 200Hz, enabling fast response and precise tracking.

[0014] Furthermore, the rehabilitation assessment system monitors the following indicators in real time: (4-1) Human-computer interaction capability That is, the difference between the sensor measurement value and the exoskeleton output value; (4-2) Electromyographic-torque synergy index , that is, the correlation coefficient between electromyographic signals and human force exertion; (4-3) Gait symmetry index That is, the difference in gait cycles between the left and right legs; among which For the left leg gait cycle, For the right leg gait cycle, This represents the average of the left leg gait cycle and the right leg gait cycle. (4-4) Training load This refers to the cumulative work done by the human body. After each training session, the rehabilitation assessment system automatically adjusts the assistance parameters for the next training session based on the above indicators: if CI < 0.6, it indicates insufficient muscle activation, and the assistance ratio coefficient is reduced. If SI > 0.15, it indicates gait asymmetry, and assistance should be increased on the weak side; if TL exceeds the preset threshold, it indicates patient fatigue, and the training intensity should be reduced.

[0015] Furthermore, in step S2, linear interpolation is used to resample all signals to 200Hz.

[0016] Furthermore, in step S2, the number of sampling points is m=200.

[0017] Furthermore, the water resistance coefficient The value range is 0.5 to 1.2. It is dynamically adjusted according to water depth and water temperature.

[0018] Furthermore, in step S5, the value range of the principal component number k is 15~20.

[0019] Furthermore, in step S6, when training the Support Vector Machine (SVM) classifier, a comprehensive dataset containing different water depths, temperatures, and movement speeds was used. The water depth ranged from 0.8 to 1.5 meters, the water temperature ranged from 28 to 35 degrees Celsius, and the movement speed ranged from 0.1 to 0.8 meters. .

[0020] The beneficial technical effects of this invention are as follows: (1) By adopting a fully waterproof design and an underwater-adaptive drive scheme, the key problem that existing rehabilitation exoskeleton systems cannot operate stably in an underwater environment has been solved; it can effectively utilize the buoyancy, resistance and hydrostatic pressure characteristics of water to provide patients with a rehabilitation environment that combines weight loss training, muscle strengthening and circulation promotion, and greatly expand the applicable scenarios and intervention methods for lower limb rehabilitation training. (2) By integrating cross-modal information detection and real-time analysis technology, a high-precision human motion intention recognition system was constructed by synchronously collecting electromyographic signals, joint torques and motion inertia data; it can accurately judge the wearer's motion needs in different gait phases, providing a reliable basis for realizing the collaborative control of "human in loop", and significantly enhancing the naturalness and responsiveness of underwater interaction of exoskeleton; (3) It realizes multi-level, customizable personalized rehabilitation training from passive driving to active resistance; it can dynamically adjust the auxiliary torque of the hip, knee, ankle and other joints according to the patient's lower limb strength level and rehabilitation stage, and support single joint directional assistance and multi-joint collaborative training, which not only ensures the safety of the training process, but also effectively improves the pertinence of rehabilitation and training efficiency. (4) A closed-loop rehabilitation management mechanism of “detection-analysis-control-evaluation-iteration” has been constructed. The system can continuously optimize rehabilitation plans based on single and long-term training data, form a patient-centered dynamic assessment and personalized adjustment capability, provide objective and visualized decision support for clinical rehabilitation, and has good clinical applicability and promotion value. Attached Figure Description

[0021] Figure 1 This is a mechanical structure diagram of an embodiment; Figure 2 This is a schematic diagram of a wearable application in an embodiment; Figure 3 This is a flowchart of the intent parsing module; Figure 4 This is a schematic diagram of the structure of a multi-level control strategy module.

[0022] In the diagram, the correspondence between component names and attached drawing numbers is as follows: 1. Waterproof sealed box; 2. Central control and energy system; 3. Hip exoskeleton; 4. Thigh exoskeleton; 5. Knee and lower leg exoskeleton; 6. Suspension frame; 7. Frame sling; 8. Flexion and extension joint motor; 9. Abduction and adduction joint motor; 10. Hip joint linkage; 11. Thigh joint motor; 12. Thigh support; 13. Lower leg joint motor; 14. Lower leg support; 15. Ankle adjustment mechanism; 16. Foot pressure sensing insole; 1001. Hip joint linkage vertical beam; 1002. Hip joint linkage horizontal beam. Detailed Implementation

[0023] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, the embodiment includes exoskeleton hardware and a control system. The exoskeleton hardware includes a waterproof sealed box 1, a central control and power system 2, a hip exoskeleton 3, a thigh exoskeleton 4, a knee and lower leg exoskeleton 5, a suspension frame 6, and a frame sling 7.

[0025] The central control and energy system 2 includes a main control computing module, a multi-channel signal acquisition unit, a high-energy-density battery pack, and a display screen. The high-energy-density battery pack powers the main control computing module; the multi-channel signal acquisition unit is connected to torque sensors, electromyographic electrodes, and a six-axis inertial measurement unit (IMU) mounted on the hip exoskeleton 3, thigh exoskeleton 4, and knee and lower leg exoskeleton 5, and transmits the data from the torque sensors, electromyographic electrodes, and the six-axis IMU to the main control computing module; the display screen is connected to the main control computing module and displays the data from the multi-channel signal acquisition unit in real time.

[0026] The waterproof sealed box 1 has a cubic structure and is made of waterproof material. The central control and energy system 2 is located inside the waterproof sealed box 1. When the waterproof sealed box 1 is placed in water, water cannot enter the interior of the waterproof sealed box 1.

[0027] The hip exoskeleton 3 includes a flexion-extension joint motor 8, an abduction-adduction joint motor 9, a hip torque sensor, a hip six-axis inertial measurement unit (IMU), and a hip joint linkage 10. The hip joint linkage 10 has an L-shaped structure; the vertical portion of the L-shaped structure is called the hip joint linkage vertical beam 1001, and the horizontal portion is called the hip joint linkage horizontal beam 1002. The flexion-extension joint motor 8 is connected to the end of the hip joint linkage horizontal beam 1002, and the output shaft of the flexion-extension joint motor 8 is parallel to the horizontal plane. When the flexion-extension joint motor 8 rotates, the height of the hip joint linkage vertical beam 1001 changes. The abduction-adduction joint motor 9 is connected to the end of the hip joint linkage horizontal beam 1002, and the output shaft of the abduction-adduction joint motor 9 is perpendicular to the horizontal plane. When the abduction-adduction joint motor 9 rotates, the hip joint linkage 10 rotates on the horizontal plane. There are two hip torque sensors: one connected to the flexion-extension joint motor 8 to measure its torque, and the other connected to the abduction-adduction joint motor 9 to measure its torque. A six-axis inertial measurement unit (IMU) is installed on the hip joint connecting rod vertical beam 1001 to measure its angle and angular velocity. There are two hip exoskeletons 3, both connected to the bottom of the waterproof sealed box 1, and they are mirror-symmetrical.

[0028] like Figure 2 As shown, one end of the sling 7 is connected to the hip exoskeleton 3, and the other end is connected to the suspension sling 6. The suspension sling 6 is mounted on an external track and moves along the track. When the suspension sling 6 moves, the exoskeleton hardware moves along with it under the action of the sling 7. After the exoskeleton hardware is worn on the body, the waterproof sealing box 1 is located on the back of the body, and the end of the hip joint connecting rod vertical beam 1001 fits against the hip joint of the body. The suspension sling 6 supports the rehabilitation patient to wear the device with low shoulder and back pressure, and also allows for low-load free movement with the wearer in the hydrotherapy pool. The suspension movement trajectory can be changed according to the patient's movement status.

[0029] The thigh exoskeleton 4 includes a thigh joint motor 11, a thigh support 12, and a thigh strap. The thigh support 12 is a telescopic rod structure, and its length can be adjusted to accommodate different heights. One end of the thigh support 12 is connected to the thigh joint motor 11, and the other end is connected to the knee joint calf exoskeleton 5. The thigh joint motor 11 is connected to the end of the hip joint connecting rod vertical beam 1001. When the thigh joint motor 11 rotates, the thigh support 12 can perform circular motion around the output shaft of the thigh joint motor 11. The thigh strap is connected to the thigh support 12 and fixes the thigh exoskeleton 4 to the thigh. The part of the thigh strap that contacts the thigh is equipped with electromyographic electrode pads, called thigh electromyographic electrode pads, which are used to measure the electromyographic signals of the thigh.

[0030] The knee-shin exoskeleton 5 includes a calf joint motor 13, a calf torque sensor, a calf support 14, a calf strap, an ankle adjustment mechanism 15, a plantar pressure sensing insole 16, and a calf six-axis inertial measurement unit (IMU). The calf support 14 is a telescopic rod structure, allowing it to be adjusted to accommodate different heights. One end of the calf support 14 is connected to the calf joint motor 13, and the other end is connected to the ankle adjustment mechanism 15. The calf joint motor 13 is connected to the thigh support 14; when the calf joint motor 13 rotates, the calf support 14 can perform circular motion around the output axis of the calf joint motor 13. The calf torque sensor is connected to the calf joint motor 13 and is used to measure the torque of the calf joint motor 13. The plantar pressure sensing insole 16 is mounted on the ankle adjustment mechanism 15 and is used to measure the pressure exerted on the ankle adjustment mechanism 15. The calf strap is connected to the calf support 14 and secures the knee-shin exoskeleton 5 to the calf. Electromyography (EMG) electrodes, called calf EMG electrodes, are installed at the point where the calf strap contacts the calf to measure EMG signals in the calf. A six-axis inertial measurement unit (IMU) is installed on the calf support 14 to measure the angle and angular velocity of the calf support 14.

[0031] The control system comprises an intent parsing module, a multi-level control strategy module, and a rehabilitation assessment system. The intent parsing module receives data from the hip torque sensor, hip six-axis inertial measurement unit (IMU), thigh torque sensor, thigh electromyography (EMG) electrodes, calf torque sensor, calf EMG electrodes, and calf six-axis IMU, and parses the human's movement intent. The multi-level control strategy module includes a top-level collaborative control strategy and a bottom-level motion control strategy. The top-level collaborative control strategy converts the movement intent output by the intent parsing module into action commands for the exoskeleton hardware. The bottom-level motion control strategy employs a biomechanical model-based inversion control algorithm to calculate the action commands from the top-level collaborative control strategy into the desired torque or trajectory of each joint, and achieves biomimetic motion output by real-time reading of joint motor current and torque sensor feedback. The rehabilitation assessment system dynamically adjusts the control parameters of the bottom-level motion control strategy.

[0032] like Figure 3 As shown, the intent parsing module achieves motion intent recognition through the following steps: Step S1: Multi-source signal acquisition and preprocessing Torque sensors placed at the hip and knee joints collect joint torque signals at a sampling frequency of 200Hz. Simultaneously, bioelectrical signals are acquired at a sampling frequency of 1000 Hz using surface electromyography electrodes attached to the surfaces of major muscle groups such as the rectus femoris, biceps femoris, and gastrocnemius. Joint angles were acquired at a sampling frequency of 100Hz using a six-axis inertial measurement unit (IMU) embedded in the hip and knee exoskeletons. and angular velocity .

[0033] In light of the characteristics of the underwater environment, the acquired signals are preprocessed as follows: (1) Torque signal Water resistance compensation is performed based on the underwater motion velocity v and the joint angular velocity. Calculate the water resistance torque (in This is the water resistance coefficient, with a value ranging from 0.5 to 1.2. (Dynamically adjusted based on water depth and temperature) to obtain net human exertion. ; (2) Electromyographic signals A 50Hz notch filter was used to eliminate power frequency interference, and a 20-500Hz bandpass filter was used to remove motion artifacts and underwater electromagnetic noise. Full-wave rectification and a 50ms sliding window RMS calculation were then performed to obtain the electromyography amplitude sequence. ; (3) Perform Kalman filtering on the angle and angular velocity signals to eliminate measurement noise caused by water flow disturbance.

[0034] Step S2: Timing Alignment and Data Synchronization Because the sampling frequencies of the various sensors differ, linear interpolation is used to resample all signals to 200Hz, and a hardware timestamp synchronization mechanism ensures that the time alignment error of each modality's data is less than 5ms. A sliding time window of 1 second (containing 200 sampling points) is constructed to form a multi-channel time-series data matrix. , where n is the total number of sensor channels.

[0035] Step S3: Multimodal Feature Extraction The following feature vectors are extracted from the preprocessed multi-source signal: (1) Torque characteristics: Calculate the mean, variance, peak value, and zero-crossing rate of the torque at each joint to form a torque characteristic vector. ; (2) Electromyographic features: The root mean square value (RMS), mean power frequency (MPF), median frequency (MF), and waveform length (WL) of each muscle group were extracted to form an electromyographic feature vector. ; (3) Kinematic features: Extract the mean, rate of change, and acceleration of joint angles, as well as the duty cycle of the gait cycle, to form a kinematic feature vector. ; (4) Underwater environment characteristics: Based on the acceleration changes detected by the IMU, the rate of change of water resistance coefficient and the buoyancy assist ratio are calculated to form an environmental characteristic vector. .

[0036] The above features are concatenated to form the original high-dimensional feature vector. .

[0037] Step S4: Feature Fusion and Dimensionality Reduction Principal component analysis (PCA) is used to reduce the dimensionality of the original high-dimensional eigenvectors, retaining the top k principal components (typically k=15-20) with a cumulative contribution rate of 95%, thus obtaining the dimensionality-reduced fused eigenvectors. This feature vector comprehensively characterizes the wearer's mechanical state, muscle activation patterns, and kinematic properties in an underwater environment.

[0038] Step S5: Motion Intent Classification and Recognition fuse feature vectors The input is a pre-trained Support Vector Machine (SVM) classifier that uses a Radial Basis Function (RBF) kernel function. The mapping between features and intention categories has been established through offline training. The classifier output includes the following motion intention categories: stationary standing, starting preparation, swing phase (left / right leg), support phase (left / right leg), stop, and turn (left / right), for a total of 9 motion intention categories.

[0039] To address the unique characteristics of the underwater environment, a comprehensive dataset containing data of varying depths (0.8-1.5m), temperatures (28-35℃), and movement speeds (0.1-0.8m / s) was used when training the classifier, making the classifier robust to changes in the underwater environment.

[0040] Step S6: Gait phase estimation and command generation Based on the identified movement intention category and current joint angle The current gait phase is estimated using a finite state machine (FSM), and the gait cycle is divided into 8 sub-phases: initial contact phase, load response phase, mid-single support phase, late-single support phase, pre-swing phase, initial swing phase, mid-swing phase (75-87%), and late-swing phase.

[0041] Based on the movement intention category and gait phase, the generated control command includes the following parameters: target assist mode. Expected assist ratio coefficient for each joint (0 indicates no assist, 1 indicates full assist); Water resistance compensation coefficient (Used to compensate for underwater motion resistance); Safety limiting parameters (Maximum output torque of each joint). This command is updated at a frequency of 50Hz and sent to the multi-level control strategy module.

[0042] Multi-level control strategy module, such as Figure 4 As shown, its strategy consists of two collaborative parts: one is the user-led "human-in-loop" high-level collaborative control of the exoskeleton, and the other is the motion control of the exoskeleton at the lower level, which is based on motion bionics.

[0043] (I) High-level collaborative control strategy Control commands output by the intent parsing module The human-machine collaborative impedance control algorithm is adopted, and its control law is as follows: in, The auxiliary torque output to the exoskeleton; To assist the proportional coefficient; The expected joint torque is calculated based on a healthy gait database; and For position and velocity gain; and For reference trajectory; This is the feedforward compensation term for water resistance.

[0044] Depending on the different assist modes, the parameter settings are as follows: Passive drive mode: =0, =200 , =200 , =1.0, the exoskeleton is driven entirely along a preset trajectory; Assist mode: =0.3-0.7 (dynamically adjusted based on muscle strength assessment), =50 , =5 , =0.8, retaining some active control of the body; resistance training mode: =-0.2 (applying reverse resistance), =10 , =2 , =0.3, which is the exoskeleton resistance added on top of the water resistance.

[0045] (II) Low-level motion control strategy The expected torque calculated at the top level The inverse dynamics model is used to convert the current commands into those of the motors at each joint. : in, This is the gravity term (buoyancy correction needs to be considered underwater). ); For Coriolis force and centrifugal force terms; Friction; This is the motor torque constant.

[0046] The motor driver adopts a three-loop cascaded PID control system consisting of a current loop, a speed loop, and a position loop. The inner current loop has a frequency of 5kHz, the middle speed loop has a frequency of 1kHz, and the outer position loop has a frequency of 200Hz, enabling fast response and precise tracking.

[0047] Furthermore, the integrated rehabilitation assessment system within the module continuously analyzes data from the entire process of "cross-modal detection and analysis - exoskeleton execution - motion assistance feedback," dynamically adjusting control parameters to ensure the safety, adaptability, and effectiveness of rehabilitation training. The system monitors the following indicators in real time: Human-computer interaction force. (Difference between sensor measurements and exoskeleton output); Electromyography-torque synergy index (Correlation coefficient between electromyographic signals and human force exertion); Gait symmetry index (Difference in gait cycle between left and right legs); Training load (The cumulative work done by the human body).

[0048] After each training session (approximately 10-30 minutes), the system automatically adjusts the assist parameters for the next training session based on the above indicators: if CI < 0.6, it indicates insufficient muscle activation, and the assist ratio should be reduced. If SI > 0.15, it indicates gait asymmetry, and assistance should be increased on the weak side; if TL exceeds the preset threshold, it indicates patient fatigue, and the training intensity should be reduced.

[0049] Through a closed-loop iterative process of "detection-analysis-control-evaluation-adjustment", personalized rehabilitation training can be dynamically optimized.

[0050] The embodiment can realize personalized, phased underwater training programs based on the wearer's lower limb strength level and specific rehabilitation needs. The system dynamically analyzes the patient's motor ability by collecting joint torque and electromyographic signals in real time, and precisely controls the output torque of motors in the hip, knee, and ankle joints to construct a multi-level rehabilitation mode from passive rehabilitation to active resistance. The specific training process is as follows: For early rehabilitation patients with weak muscles, the system adopts a passive drive mode, with the exoskeleton driving the lower limbs to complete a full range of motion according to a preset trajectory, avoiding muscle atrophy and maintaining joint flexibility; as the wearer's muscle strength improves and they enter the next stage of rehabilitation training, the system can switch to an assistive mode, providing necessary torque compensation only when the patient initiates movement, promoting neuromuscular function reconstruction; for patients in the middle and late stages of rehabilitation, whose lower limb strength has nearly recovered, the system can further provide unidirectional resistance training, such as applying controllable motor joint torque resistance during hip internal and external rotation or thigh flexion and extension movements, combined with water flow resistance to strengthen specific muscle groups. Furthermore, the system can implement targeted assistance strategies for single joints, such as independently setting hip internal and external rotation assistance, or... Figure 2 The system provides assistance only during the forward and backward leg swings, enabling highly targeted and personalized rehabilitation. Throughout the training process, the system adjusts the assistance intensity and range of motion in real time based on biomechanical and physiological signals, ensuring both the effectiveness and safety of the training while significantly improving the patient's performance and sense of participation in the rehabilitation process in the underwater environment.

[0051] The embodiment is equipped with a complete rehabilitation training effect evaluation mechanism. Through multi-channel electrode pads attached to the skin surface of the wearer's lower limbs, it collects electromyographic signals of major muscle groups in real time. Combined with data from joint torque sensors and inertial measurement units, it achieves precise monitoring and quantitative evaluation of the patient's rehabilitation progress. This system can not only analyze key indicators such as muscle activation level, joint range of motion, and motor coordination in a single training session, but also establish a patient-specific rehabilitation trend model through cross-time data accumulation. In a single training session, the system assesses the patient's muscle recruitment ability and fatigue state in real time based on the intensity and pattern characteristics of electromyographic signals, dynamically adjusting the intensity and auxiliary strategies of the current training. In long-term rehabilitation with multiple training sessions, the system identifies weaknesses and progress trends in the rehabilitation process by comparing data from multiple training sessions, automatically iteratively updating personalized training plans to achieve a closed-loop optimization of "evaluation-training-re-evaluation-optimization." This iterative evaluation mechanism based on time-series data ensures the scientific validity and adaptability of the training plan at each stage, and provides clinical rehabilitation physicians with visualized evidence for efficacy judgment, significantly improving the accuracy and long-term effectiveness of underwater rehabilitation training.

[0052] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, and for those of ordinary skill in the art, various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. Therefore, the present invention is not limited to the specific details without departing from the general concept defined by the claims and their equivalents.

Claims

1. A lower limb underwater rehabilitation training exoskeleton system, characterized in that: It includes exoskeleton hardware and control system; the exoskeleton hardware includes a waterproof sealed box (1), a central control and energy system (2), a hip exoskeleton (3), a thigh exoskeleton (4), a knee and lower leg exoskeleton (5), a suspension bone (6), and a bone sling (7); The central control and energy system (2) includes a main control computing module, a multi-channel signal acquisition unit, a high-energy-density battery pack, and a display screen; the high-energy-density battery pack powers the main control computing module; the multi-channel signal acquisition unit is connected to torque sensors, electromyographic electrodes, and a six-axis inertial measurement unit (IMU) installed on the hip exoskeleton (3), thigh exoskeleton (4), and knee joint calf exoskeleton (5), and transmits the data of the torque sensors, electromyographic electrodes, and six-axis inertial measurement unit (IMU) to the main control computing module; the display screen is connected to the main control computing module and displays the data of the multi-channel signal acquisition unit in real time; The waterproof sealing box (1) has a cubic structure and is made of waterproof material. The central control and energy system (2) is located inside the waterproof sealing box (1). When the waterproof sealing box (1) is placed in water, water cannot enter the interior of the waterproof sealing box (1). The hip exoskeleton (3) includes a flexion-extension joint motor (8), an abduction-adduction joint motor (9), a hip torque sensor, a hip six-axis inertial measurement unit (IMU), and a hip joint linkage (10). The hip joint linkage (10) has an L-shaped structure. The vertical part of the L-shaped structure is called the hip joint linkage vertical beam (1001), and the horizontal part of the L-shaped structure is called the hip joint linkage horizontal beam (1002). The flexion-extension joint motor (8) is connected to the end of the hip joint linkage horizontal beam (1002), and the output shaft of the flexion-extension joint motor (8) is parallel to the horizontal plane. When the flexion-extension joint motor (8) rotates, the height of the hip joint linkage vertical beam (1001) changes. The abduction-adduction joint motor (9) is connected to the end of the hip joint linkage horizontal beam (1002). The end of 02) is connected, and the output shaft of the abduction and adduction joint motor (9) is perpendicular to the horizontal plane. When the abduction and adduction joint motor (9) rotates, the hip joint link (10) rotates on the horizontal plane. There are two hip torque sensors. One is connected to the flexion and extension joint motor (8) and measures the torque of the flexion and extension joint motor (8). The other is connected to the abduction and adduction joint motor (9) and measures the torque of the abduction and adduction joint motor (9). The hip six-axis inertial measurement unit (IMU) is set on the hip joint link vertical beam (1001) and measures the angle and angular velocity of the hip joint link vertical beam (1001). There are two hip exoskeletons (3). Both are connected to the bottom of the waterproof sealed box (1) and the two are mirror symmetrical. One end of the sling (7) is connected to the hip exoskeleton (3), and the other end is connected to the suspension sling (6); the suspension sling (6) is set on the external track and moves along the external track; when the suspension sling (6) moves, the exoskeleton hardware moves with the suspension sling (7) under the action of the sling (7); after the exoskeleton hardware is worn on the human body, the waterproof sealing box (1) is located on the back of the human body, and the end of the hip joint connecting rod vertical beam (1001) fits against the hip joint of the human body; The thigh exoskeleton (4) includes a thigh joint motor (11), a thigh support (12), and a thigh strap. The thigh support (12) is a telescopic rod structure, and its length can be adjusted to accommodate different heights. One end of the thigh support (12) is connected to the thigh joint motor (11), and the other end is connected to the knee joint calf exoskeleton (5). The thigh joint motor (11) is connected to the end of the hip joint connecting rod vertical beam (1001). When the thigh joint motor (11) rotates, the thigh support (12) can make circular motion around the output shaft of the thigh joint motor (11). The thigh strap is connected to the thigh support (12) and fixes the thigh exoskeleton (4) to the thigh. The part of the thigh strap that contacts the thigh is provided with electromyographic electrode pads, called thigh electromyographic electrode pads, which are used to measure the electromyographic signals of the thigh. The knee joint calf exoskeleton (5) includes a calf joint motor (13), a calf torque sensor, a calf support (14), a calf strap, an ankle adjustment mechanism (15), a foot pressure sensing insole (16), and a calf six-axis inertial measurement unit (IMU). The calf support (14) is a telescopic rod structure, and its length can be adjusted to accommodate different heights. One end of the calf support (14) is connected to the calf joint motor (13), and the other end is connected to the ankle adjustment mechanism (15). The calf joint motor (13) is connected to the thigh support (14). When the calf joint motor (13) rotates, the calf support (14) can rotate around the calf joint motor (13). The output shaft of the ) performs circular motion; the calf torque sensor is connected to the calf joint motor (13) and is used to measure the torque of the calf joint motor (13); the foot pressure sensing insole (16) is set on the ankle adjustment mechanism (15) and is used to measure the pressure on the ankle adjustment mechanism (15); the calf strap is connected to the calf support (14) and fixes the knee joint calf exoskeleton (5) on the calf; the part of the calf strap that fits against the calf is provided with electromyographic electrode pads, called calf electromyographic electrode pads, which are used to measure the electromyographic signals of the calf; the calf six-axis inertial measurement unit (IMU) is set on the calf support (14) and measures the angle and angular velocity of the calf support (14); The control system includes an intent parsing module, a multi-level control strategy module, and a rehabilitation assessment system. The intent parsing module receives data from a hip torque sensor, a hip six-axis inertial measurement unit (IMU), a thigh torque sensor, thigh electromyography (EMG) electrodes, a calf torque sensor, a calf EMG electrodes, and a calf six-axis IMU, and parses the human's movement intent. The multi-level control strategy module includes a top-level collaborative control strategy and a bottom-level motion control strategy. The top-level collaborative control strategy converts the movement intent output by the intent parsing module into action commands for the exoskeleton hardware. The bottom-level motion control strategy uses a biomechanical model-based inversion control algorithm to calculate the action commands of the top-level collaborative control strategy into the desired torque or trajectory of each joint, and achieves biomimetic motion output by reading the joint motor current and torque sensor feedback in real time. The rehabilitation assessment system dynamically adjusts the control parameters of the bottom-level motion control strategy.

2. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 1, characterized in that, The steps for the intent parsing module to parse motion intent are as follows: S1. Acquire joint torque signals using hip torque sensors, thigh torque sensors, and calf torque sensors. Bioelectrical signals were collected using electromyography (EMG) electrodes on the thigh and calf. Joint angles were acquired using a six-axis inertial measurement unit (IMU) at the hip and a six-axis inertial measurement unit (IMU) at the lower leg. and angular velocity The acquired signals are preprocessed as follows: (1-1) Torque signal Water resistance compensation is performed based on the underwater motion velocity v and the joint angular velocity. Calculate the water resistance torque To obtain the net human body's power ,in This is the water resistance coefficient; (1-2) Electromyographic signals A 50Hz notch filter was used to eliminate power frequency interference, and a 20~500Hz bandpass filter was used to remove motion artifacts and underwater electromagnetic noise. Full-wave rectification and a 50ms sliding window RMS calculation were then performed to obtain the electromyography amplitude sequence. ; (1-3) Perform Kalman filtering on the angle and angular velocity signals to eliminate measurement noise caused by water flow disturbance; S2. A linear interpolation method is used to resample all signals to the same frequency, and a hardware timestamp synchronization mechanism is used to ensure that the time alignment error of each mode data is less than 5ms; a sliding time window with a length of 1 second and containing m sampling points is constructed to form a multi-channel timing data matrix. , where n is the total number of sensor channels; S3. Extract the following feature vectors from the preprocessed multi-source signal: (2-1) Based on the joint torque signal Calculate the mean, variance, peak value, and zero-crossing rate of the torque at each joint to obtain the torque characteristic vector. ; (2-2) Based on bioelectrical signals Calculate the root mean square (RMS), average power frequency (MPF), median frequency (MF), and waveform length (WL) for each muscle group to obtain the electromyographic feature vector. ; (2-3) Based on joint angles and angular velocity Calculate the mean, rate of change, and acceleration of the joint angles, as well as the gait cycle duty cycle, to obtain the kinematic feature vector. ; (2-4) Based on the acceleration changes detected by the hip six-axis inertial measurement unit (IMU) and the lower leg six-axis inertial measurement unit (IMU), calculate the rate of change of water resistance coefficient and the buoyancy assist ratio to obtain the environmental feature vector. ; Concatenate the eigenvectors from (2-1) to (2-4) to obtain the original high-dimensional eigenvectors. ; S4. Principal component analysis is used to analyze the original high-dimensional eigenvectors. Dimensionality reduction is performed, retaining the top k principal components with a cumulative contribution rate of 95%, to obtain the dimensionality-reduced fusion feature vector. ; S5. Merge feature vectors Input a pre-trained Support Vector Machine (SVM) classifier, which uses a Radial Basis Function (RBF) kernel function and has established a mapping relationship between features and intention categories through offline training; The classifier outputs the following 9 types of motion intentions: standing still, starting preparation, left leg swing phase, right leg swing phase, left leg support phase, right leg support phase, stopping, turning left, and turning right; S6. Based on the identified movement intention and current joint angle The current gait phase is estimated using a finite state machine (FSM), and the gait cycle is divided into 8 sub-phases: initial contact phase, load response phase, mid-single support phase, late-single support phase, pre-swing phase, initial swing phase, mid-swing phase, and late-swing phase. S7. Generate control commands based on movement intention and gait phase. And send it to the multi-level control strategy module; where: M is the target assist mode, and ; The expected assist ratio for the hip joint. The expected assist ratio for the knee joint, and 0 indicates no assist and 1 indicates full assist; This is the water resistance compensation coefficient, and ; The safety limit parameter is the maximum output torque of each joint.

3. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 2, characterized in that, The top-level collaborative control strategy is as follows: Control commands output by the intent parsing module A human-machine collaborative impedance control algorithm is used to calculate the auxiliary torque output by the exoskeleton. in: To help the proportional coefficient, The expected joint torque is calculated based on a healthy gait database. and For position and velocity gain, and For reference trajectory, This is the feedforward compensation term for water resistance. Depending on the different target assistance mode M, the parameters are set as follows: (3-1) When M is passively driven, the exoskeleton is driven entirely along the preset trajectory. =0, =200 , =200 , =1.0; (3-2) When M is used for assistance, some active control of the human body is retained. =0.3~0.7, =50 , =5 , =0.8; (3-3) M refers to resistance training, where exoskeleton resistance is added to the water resistance. =-0.2, =10 , =2 , =0.

3.

4. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 3, characterized in that, The underlying motion control strategy is as follows: The auxiliary torque calculated by the top-level collaborative control strategy The inverse dynamics model is used to convert the current commands to the motors at each joint: in: For gravity, For Coriolis force and centrifugal force terms, For friction, The torque constant of the motor; The drivers for each joint motor employ a three-loop cascaded PID control system consisting of a current loop, a speed loop, and a position loop. The inner current loop has a frequency of 5kHz, the middle speed loop has a frequency of 1kHz, and the outer position loop has a frequency of 200Hz, enabling fast response and precise tracking.

5. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 3, characterized in that: The rehabilitation assessment system monitors the following indicators in real time: (4-1) Human-computer interaction capability That is, the difference between the sensor measurement value and the exoskeleton output value; (4-2) Electromyographic-torque synergy index , that is, the correlation coefficient between electromyographic signals and human force exertion; (4-3) Gait symmetry index That is, the difference in gait cycles between the left and right legs; among which For the left leg gait cycle, For the right leg gait cycle, This represents the average of the left leg gait cycle and the right leg gait cycle. (4-4) Training load This refers to the cumulative work done by the human body. After each training session, the rehabilitation assessment system automatically adjusts the assistance parameters for the next training session based on the above indicators: if CI < 0.6, it indicates insufficient muscle activation, and the assistance ratio coefficient is reduced. If SI > 0.15, it indicates gait asymmetry, and the weak side should be given additional assistance. If the total training time (TL) exceeds the preset threshold, it indicates that the patient is fatigued, and the training intensity should be reduced.

6. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 1, characterized in that: In step S2, linear interpolation is used to resample all signals to 200Hz.

7. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 1, characterized in that, In step S2, the number of sampling points is m=200.

8. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 1, characterized in that: The water resistance coefficient The value range is 0.5 to 1.

2. It is dynamically adjusted according to water depth and water temperature.

9. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 2, characterized in that, In step S5, the value range of the principal component number k is 15~20.

10. The underwater rehabilitation training exoskeleton system for the lower limbs according to claim 2, characterized in that: In step S6, when training the Support Vector Machine (SVM) classifier, a comprehensive dataset containing different water depths, temperatures, and movement speeds was used. The water depth ranged from 0.8 to 1.5 meters, the water temperature ranged from 28 to 35 degrees Celsius, and the movement speed ranged from 0.1 to 0.8 meters. .

Images