摘要
随着人口老龄化加剧,肢体功能障碍患者增加,特别是因慢性病导致行走困难的患者,需要一种能准确识别人体运动意图并辅助康复训练的设备。提出一种融合数据增强与注意力机制的卷积神经网络-双向长短期记忆(Convolutional Neural Network-Bidirectional Long Short Term Memory,CNN-BiLSTM)模型,用于下肢髋膝关节角度预测。通过惯性测量单元(Inertial Measurement Unit,IMU)采集步态信号,并采用多种数据增强技术如高斯噪声、随机遮挡、缩放及时域扭曲,模拟真实干扰情况以提升模型性能。实验结果表明,新模型在处理原始及增强数据时均优于传统网络模型,特别是在膝关节预测上,误差显著降低,其误差指标均方根误差(Root Mean Square Error,RMSE)、平均绝对误差(Mean Absolute Error,MAE)、决定系数(R 2)最优状态下分别为1.89、1.23、0.9878,平均误差较传统方法减少超过10%。不同增强策略的应用进一步增强了模型对个体差异和步态变化的适应能力,提高了预测精度和稳定性,为智能康复外骨骼系统的发展提供了新的方法和技术支持,有助于提升康复训练效果和患者生活质量。
With the intensification of population aging and the increasing number of patients with physical disabilities,particularly those with mobility difficulties due to chronic diseases,there is a need for a device that can accurately recognize human movement intentions and assist in rehabilitation training.A Convolutional Neural Network-Bidirectional Long Short Term Memory(CNN-BiLSTM)model integrated with data augmentation and attention mechanisms is proposed for predicting hip and knee joint angles in lower limbs.Gait signals are collected using Inertial Measurement Unit(IMU),and various data augmentation techniques,including Gaussian noise,random occlusion,scaling,and time-domain warping,are applied to simulate real-world interference and enhance model performance.Experimental results show that the new model outperforms traditional network models in handling both original and augmented data,particularly showing significant error reduction in knee joint predictions.The optimal error metrics,namely Root Mean Square Error(RMSE),Mean Absolute Error(MAE),and coefficient of determination(R 2),are 1.89,1.23,and 0.9878,respectively,with an average error reduction exceeding 10%compared to conventional methods.The application of different augmentation strategies further enhances the model’s adaptability to individual differences and gait variations,improving prediction accuracy and stability.This provides new methodologies and technical support for the development of intelligent rehabilitation exoskeleton systems,contributing to enhanced rehabilitation outcomes and patients’quality of life.
作者
洪涛
吴钦木
HONG Tao;WU Qinmu(College of Electrical Engineering,Guizhou University,Guiyang 550025,China)
出处
《无线电工程》
2025年第12期2479-2487,共9页
Radio Engineering
基金
贵州省科技支撑计划项目(黔科合支撑[2023]一般179)。
