面向连续意图的智能下肢假肢运动过渡感知软标签与动态加权预测方法

doi:10.3969/j.issn.1006-9771.2026.09.012

《中国康复理论与实践》 ›› 2026, Vol. 32 ›› Issue (9): 1100-1107.doi: 10.3969/j.issn.1006-9771.2026.09.012

面向连续意图的智能下肢假肢运动过渡感知软标签与动态加权预测方法

苏本跃¹^,²^,³(), 王保乾¹^,²^,³^,⁴, 盛敏¹^,⁵

¹ 安徽省智能感知与计算重点实验室，安徽安庆市 246133
² 铜陵学院人工智能学院，安徽铜陵市 244061
³ 安徽省铜基材料数字化智能制造工程研究中心，安徽铜陵市 244061
⁴ 安庆师范大学人工智能与计算机学院，安徽安庆市 246133
⁵ 安庆师范大学数学与统计学院，安徽安庆市 246133

收稿日期:2026-06-06 修回日期:2026-08-05 接受日期:2026-08-07 出版日期:2026-09-25 发布日期:2026-09-20
通讯作者: 苏本跃, E-mail： subenyue@sohu.com
作者简介:苏本跃(1971-)，男，汉族，安徽芜湖市人，博士，教授，硕士研究生导师，主要研究方向：康复医学、模式识别、人工智能。
基金资助:
安徽省高校优秀科研创新团队项目(2023AH010056);铜陵学院联合培养硕士研究生创新基金项目(25tlcb04)

Soft labeling and dynamic weighting prediction for locomotion transition perception in intelligent lower-limb prostheses facing continuous intent

Su Benyue¹^,²^,³(), Wang Baoqian¹^,²^,³^,⁴, Sheng Min¹^,⁵

¹ University Key Laboratory of Intelligent Perception and Computing of Anhui Province, Anqing, Anhui 246133, China
² School of Artificial Intelligence, Tongling University, Tongling, Anhui 244061, China
³ Anhui Engineering Research Center of Intelligent Manufacturing of Copper-based Materials, Tongling, Anhui 244061, China
⁴ School of Artificial Intelligence and Computer Science, Anqing Normal University, Anqing, Anhui 246133, China
⁵ School of Mathematics and Statistics, Anqing Normal University, Anqing, Anhui 246133, China

Received:2026-06-06 Revised:2026-08-05 Accepted:2026-08-07 Published:2026-09-25 Online:2026-09-20
Contact: Su Benyue, E-mail: subenyue@sohu.com
Supported by:
Excellent Innovative Research Team of Universities in Anhui Province(2023AH010056);Tongling University Innovation Fund for Joint Postgraduate Training(25tlcb04)

摘要/Abstract

摘要：

目的提出一种智能下肢假肢运动意图预测过渡感知软标签结合动态加权的运动意图预测算法，解决离散硬标签所致的过渡区预测震荡。

方法 2025年1月至7月，招募铜陵学院7例健康受试者佩戴下肢假肢模拟支具参与试验，采用6个惯性测量单元采集21种下肢运动模式下的加速度和角速度。采用余弦函数构造软标签，还原步态模式转换的平滑演化梯度；提取软标签概率分布特征构建过渡感知加权机制，引导模型训练阶段重点关注模式转换边界样本；以时间卷积网络构建预测模型。

结果全部运动模式总体预测准确率94.03%。该方法可提升模式转换阶段预测性能，有效抑制判决边界附近预测概率抖动，输出平滑连续的意图转换概率轨迹。

结论过渡感知软标签动态加权方法能够克服传统离散标签带来的过渡区预测震荡缺陷，可为智能下肢假肢输出平稳、鲁棒且具备超前量的运动意图决策信号。

关键词: 智能下肢假肢, 连续运动意图, 惯性测量单元, 动态加权, 模式转换

Abstract:

Objective To propose a prediction method combining transition-aware soft labels and dynamic weighting to mitigate transition-zone prediction jitter of hard labels in locomotion intent prediction for intelligent lower-limb prostheses.

Methods Seven healthy subjects were recruited from Tongling University from January to July, 2025, wearing simulated lower-limb-prosthesis braces. Six inertial measurement units were used to acquire acceleration and angular-velocity under 21 locomotion modes. Cosine-based soft labels were introduced to reconstruct the smooth evolutionary gradient of locomotion-mode transitions. Probability-distribution features derived from soft labels were used to build a transition-aware weighting mechanism, which drived the model to focus dynamically on transition boundaries during training. A temporal convolutional network served as the prediction backbone.

Results The proposed method achieved an overall prediction accuracy of 94.03% across all locomotion modes, which substantially improved prediction performance during mode transitions, and suppressed probability jitter near decision boundaries and yielded smooth continuous probability trajectories of intent evolution.

Conclusion The proposed transition-aware soft-labeling and dynamic-weighting approach alleviates transition-region prediction oscillation caused by traditional discrete labels. It can provide smooth, robust and anticipatory intent-decision signals for intelligent lower-limb prostheses.

Key words: intelligent lower-limb prosthesis, continuous locomotion intent, inertial measurement unit, dynamic weighting, mode transition

中图分类号:

R496

苏本跃, 王保乾, 盛敏. 面向连续意图的智能下肢假肢运动过渡感知软标签与动态加权预测方法[J]. 《中国康复理论与实践》, 2026, 32(9): 1100-1107.

Su Benyue, Wang Baoqian, Sheng Min. Soft labeling and dynamic weighting prediction for locomotion transition perception in intelligent lower-limb prostheses facing continuous intent[J]. Chinese Journal of Rehabilitation Theory and Practice, 2026, 32(9): 1100-1107.

图/表 6

表1

表2

表3

表4

图1

图2

参考文献 33

[1]	Sullivan L M, Cowan M, Gabert L, et al. Unified control of a powered knee-ankle prosthesis enables walking, stairs, transitions, and other daily ambulation activities[J]. IEEE Trans Neural Syst Rehabil Eng, 2025, 33: 3024-3039. doi: 10.1109/TNSRE.2025.3595496
[2]	黄品高. 智能下肢假肢运动意图的感测与识别关键技术研究[D]. 深圳: 中国科学院大学(中国科学院深圳先进技术研究院), 2020.
	Huang P G. Study on the key technologies of motion intention sensing and recognition of intelligent lower limb prostheses[D]. Shenzhen: The University of Chinese Academy of Sciences (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences), 2020.
[3]	唐易, 陈奕希, 喻洪流, 等. 一种面向下肢假肢的运动意图识别方法及验证[J]. 信息与控制, 2023, 52(5): 598-606.
	Tang Y, Chen Y X, Yu H L, et al. Research and verification of a motion intention recognition method for lower limb prosthesis[J]. Inform Control, 2023, 52(5): 598-606.
[4]	顾洪, 李伟达, 李娟. 智能膝关节假肢研究现状及发展趋势[J]. 中国康复理论与实践, 2016, 22(9): 1080-1085.
	Gu H, Li W D, Li J. State-of-the-art and development of intelligent knee prosthesis (review)[J]. Chin J Rehabil Theory Pract, 2016, 22(9): 1080-1085.
[5]	王启宁, 郑恩昊, 陈保君, 等. 面向人机融合的智能动力下肢假肢研究现状与挑战[J]. 自动化学报, 2016, 42(12): 1780-1793.
	Wang Q N, Zheng E H, Chen B J, et al. Recent progress and challenges of robotic lower-limb prostheses for human-robot integration[J]. Acta Automat Sinic, 2016, 42(12): 1780-1793.
[6]	Tang C Y, Xu Z Y, Occhipinti E, et al. From brain to movement: wearables-based motion intention prediction across the human nervous system[J]. Nano Energy, 2023, 115: 108712. doi: 10.1016/j.nanoen.2023.108712
[7]	张意彬, 吕杰, 喻洪流. 基于模糊逻辑算法的智能膝关节假肢步态相位识别[J]. 中国康复理论与实践, 2023, 29(8): 896-902. doi: 10.3969/j.issn.1006-9771.2023.08.005
	Zhang Y B, Lü J, Yu H L. Gait phase recognition in intelligent above-knee prosthesis based on fuzzy logic algorithm[J]. Chin J Rehabil Theory Pract, 2023, 29(8): 896-902.
[8]	孙为双, 路知远, 公维军, 等. 表面肌电信号在脑卒中手功能康复机器人中的研究进展[J]. 中国康复医学杂志, 2025, 40(5): 790-794.
[9]	张尧. 智能动力膝关节假肢的人体运动意图识别技术研究[D]. 长春: 吉林大学, 2024.
	Zhang Y. Research on human locomotion intent recognition technology for intelligent powered knee prosthesis[D]. Changchun: Jilin University, 2024.
[10]	Li J, Gu X, Qiu S, et al. A survey of wearable lower extremity neurorehabilitation exoskeleton: sensing, gait dynamics, and human-robot collaboration[J]. IEEE Trans Syst Man Cybernetics Syst, 2024, 54(6): 3675-3693.
[11]	杨平, 陈浩源, 高睿馨, 等. 基于低成本惯性测量单元和力敏电阻的足部三维运动数据建模[J]. 中国康复理论与实践, 2024, 30(10): 1224-1231. doi: 10.3969/j.issn.1006-9771.2024.10.013
	Yang P, Chen H Y, Gao R X, et al. Three-dimensional modeling of foot motion based on low-cost inertial measurement unit and force sensing resistor[J]. Chin J Rehabil Theory Pract, 2024, 30(10): 1224-1231.
[12]	Wang E, Chen X, Li Y, et al. Lower limb motion intent recognition based on sensor fusion and fuzzy multitask learning[J]. IEEE Trans Fuzzy Syst, 2024, 32(5): 2903-2914. doi: 10.1109/TFUZZ.2024.3364382
[13]	Su B Y, Wang J, Liu S Q, et al. A CNN-based method for intent recognition using inertial measurement units and intelligent lower limb prosthesis[J]. IEEE Trans Neural Syst Rehabil Eng, 2019, 27(5): 1032-1042. doi: 10.1109/TNSRE.7333
[14]	盛敏, 夏安琦, 王可林, 等. 基于几何与物理特征融合的智能下肢假肢运动意图识别[J]. 控制与决策, 2022, 37(4): 953-961.
	Sheng M, Xia A Q, Wang K L, et al. Movement intention recognition of intelligent lower limb prosthesis based on the fusion of geometric and physical features[J]. Control Decision, 2022, 37(4): 953-961.
[15]	盛敏, 刘双庆, 王婕, 等. 基于GMM-HMM模型的智能下肢假肢运动意图识别[J]. 仪器仪表学报, 2019, 40(5): 169-178.
	Sheng M, Liu S Q, Wang J, et al. Motion intent recognition of intelligent lower limb prosthesis based on GMM-HMM[J]. Chin J Sci Instrument, 2019, 40(5): 169-178.
[16]	苏本跃, 张利, 何清旋, 等. 基于小波特征匹配的短时人体行为识别[J]. 系统仿真学报, 2023, 35(1): 158-168. doi: 10.16182/j.issn1004731x.joss.22-0176
	Su B Y, Zhang L, He Q X, et al. Short-time human activity recognition based on wavelet features matching[J]. J Syst Simulat, 2023, 35(1): 158-168.
[17]	苏本跃, 王婕, 刘双庆, 等. 惯性动捕数据驱动下的智能下肢假肢运动意图识别方法[J]. 自动化学报, 2020, 46(7): 1517-1530.
	Su B Y, Wang J, Liu S Q, et al. An improved motion intent recognition method for intelligent lower limb prosthesis driven by inertial motion capture data[J]. Acta Automat Sinic, 2020, 46(7): 1517-1530.
[18]	苏本跃, 宗文杰, 刘文瑶, 等. 基于足底动力相数据和小波变换自适应分解的下肢意图识别方法[J]. 控制与决策, 2025, 40(10): 3005-3018.
	Su B Y, Zong W J, Liu W Y, et al. Lower limb prosthesis intention recognition method based on powered plantarflexion phase data and wavelet transform adaptive decomposition[J]. Control Decision, 2025, 40(10): 3005-3018.
[19]	苏本跃, 刘文瑶, 宗文杰, 等. 基于自适应时序对齐的智能下肢假肢运动意图识别[J]. 中国康复理论与实践, 2025, 31(9): 1101-1115. doi: 10.3969/j.issn.1006-9771.2025.09.014
	Su B Y, Liu W Y, Zong W J, et al. Adaptive temporal alignment-based motion intention recognition for intelligent lower-limb prostheses[J]. Chin J Rehabil Theory Pract, 2025, 31(9): 1101-1115.
[20]	Cheng S H, Laubscher C A, Gregg R D, et al. Controlling powered prosthesis kinematics over continuous inter-leg transitions between walking and stair ascent/descent[J]. IEEE Trans Neural Syst Rehabil Eng, 2024, 32: 3891-3901. doi: 10.1109/TNSRE.2024.3485643
[21]	Ma X, Liu Y, Zhang X, et al. Real-time continuous locomotion mode recognition and transition prediction for human with lower limb exoskeleton[J]. IEEE J Biomed Health Inform, 2025, 29(2): 1074-1086. doi: 10.1109/JBHI.2024.3462826
[22]	Xue D F, Wang Q N. Noninvasive human-prosthesis interfaces for locomotion intent recognition: a review[J]. Cyborg Bionic Syst, 2021, 2021: 9863761.
[23]	Batista P G, Vieira A C, Gaspar P D. Robotic prostheses and neuromuscular interfaces: a review of design and technological trends[J]. Machines, 2025, 13(9): 804. doi: 10.3390/machines13090804
[24]	Wu J, Maurenbrecher H, Schaer A, et al. Human gait-labeling uncertainty and a hybrid model for gait segmentation[J]. Front Neurosci, 2022, 16: 976594. doi: 10.3389/fnins.2022.976594
[25]	Haque M R, Islam M R, Sazonov E, et al. Swing-phase detection of locomotive mode transitions for smooth multi-functional robotic lower-limb prosthesis control[J]. Front Robot AI, 2024, 11: 1267072. doi: 10.3389/frobt.2024.1267072
[26]	Zheng E, Wan J, Gao S, et al. Adaptive locomotion transition recognition with wearable sensors for lower limb robotic prosthesis[J]. IEEE/ASME Trans Mechatron, 2024, 29(5): 3481-3492.
[27]	Chen Z H, Zhang T Y. Evaluation of basic sports actions for students based on DTW posture matching algorithm[J]. Syst Soft Comput, 2025, 7: 200196.
[28]	Wang X M, Bai S P, Li L R, et al. Enhancing locomotion-mode recognition and transition prediction with (bio)mechanical sensor fusion for intelligent prosthetic knees[J]. IEEE J Biomed Health Inform, 2025, 29(11): 7955-7968. doi: 10.1109/JBHI.2025.3583319
[29]	Huang H, Zhang F, Hargrove L J, et al. Continuous locomotion-mode identification for prosthetic legs based on neuromuscular-mechanical fusion[J]. IEEE Trans Biomed Eng, 2011, 58(10): 2867-2875. doi: 10.1109/TBME.2011.2161671 pmid: 21768042
[30]	Young A J, Simon A M, Fey N P, et al. Intent recognition in a powered lower limb prosthesis using time history information[J]. Ann Biomed Eng, 2014, 42(3): 631-641. doi: 10.1007/s10439-013-0909-0 pmid: 24052324
[31]	杜妍辰, 杨良栋, 汪晓铭, 等. 基于肌压信号与惯性测量单元的下肢运动意图识别[J]. 信息与控制, 2024, 53(6): 761-773.
	Du Y C, Yang L D, Wang X M, et al. Recognition of lower limb movement intention based on force myography signal and inertial measurement unit[J]. Inform Control, 2024, 53(6): 761-773.
[32]	Liu Y X, Gutierrez-Farewik E M. Joint kinematics, kinetics and muscle synergy patterns during transitions between locomotion modes[J]. IEEE Trans Biomed Eng, 2023, 70(3): 1062-1071. doi: 10.1109/TBME.2022.3208381
[33]	Hu N H, Englebienne G, Lou Z Y, et al. Learning to recognize human activities using soft labels[J]. IEEE Trans Pattern Anal Mach Intell, 2017, 39(10): 1973-1984. doi: 10.1109/TPAMI.2016.2621761

编号	性别	年龄/岁	身高/cm	体质量/kg
Sub01	男	24	183	83
Sub02	男	27	180	70
Sub03	男	26	180	85
Sub04	男	30	172	65
Sub05	女	22	172	59
Sub06	女	26	161	60
Sub07	女	25	165	48

模式类别	运动模式	运动模式描述
转换模式	行走—上楼	平地行走到上楼转换
	行走—上坡	平地行走到上坡转换
	行走—下楼	平地行走到下楼转换
	行走—下坡	平地行走到下坡转换
	行走—站立	平地行走到站立转换
	上楼—行走	上楼到平地行走转换
	上坡—行走	上坡到平地行走转换
	下楼—行走	下楼到平地行走转换
	下坡—行走	下坡到平地行走转换
	站立—行走	站立到平地行走转换
	坐—站立	坐到站立转换
	站立—坐	站立到坐转换
稳态模式	行走	平地行走
	上楼	稳步上台阶
	上坡	稳步上坡
	下楼	稳步下台阶
	下坡	稳步下坡
	右转弯	平地右转弯
	左转弯	平地左转弯
	坐	保持坐姿
	站立	保持站立

方法	准确率/%	加权F1/%	过渡帧F1/%	过渡帧MSE
硬标签CE	93.66±0.73	93.47±0.76	—	—
硬标签CE+TC	93.94±0.65	93.72±0.71	—	—
软标签CE (γ = 0)	93.89±0.64	93.65±0.67	78.17±2.10	0.00790
软标签CE+TAW (γ = 2)	94.03±0.66	93.81±0.69	78.71±1.82	0.00755
MSE (γ = 0)	93.99±0.60	93.77±0.63	77.95±1.38	0.00789
MSE+TAW (γ = 2)	94.03±0.58	93.80±0.60	78.40±1.64	0.00742

方法	稳态帧准确率	过渡帧准确率	稳态帧F1	过渡帧F1
软标签CE (γ = 0)	96.70	78.64	96.53	78.17
软标签CE+TAW (γ = 2)	96.76	79.13	96.61	78.71
MSE (γ = 0)	96.88	78.40	96.72	77.95
MSE+TAW (γ = 2)	96.81	78.89	96.66	78.40

面向连续意图的智能下肢假肢运动过渡感知软标签与动态加权预测方法

Soft labeling and dynamic weighting prediction for locomotion transition perception in intelligent lower-limb prostheses facing continuous intent

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 6

参考文献 33

相关文章 3

Metrics

本文评价

推荐阅读 0

[1]	苏本跃, 刘文瑶, 宗文杰, 王保乾, 盛敏. 基于自适应时序对齐的智能下肢假肢运动意图识别[J]. 《中国康复理论与实践》, 2025, 31(9): 1101-1115.
[2]	杨平, 陈浩源, 高睿馨, 王欣平. 基于低成本惯性测量单元和力敏电阻的足部三维运动数据建模[J]. 《中国康复理论与实践》, 2024, 30(10): 1224-1231.
[3]	杨波;付成龙 . 惯性测量单元在老年人跌倒风险评估中的应用进展[J]. 《中国康复理论与实践》, 2015, 21(07): 780-784.