In order to more effectively assess the effectiveness of ankle functional rehabilitation training for patients, a model combining a convolutional neural network module with a multi-head graph attention residual module based on dynamic gating was proposed by the research team, a specialized database comprising 20 categories of ankle rehabilitation (covering typical features such as dorsiflexion and adduction) was established to address the scarcity of existing clinical image datasets, and a multimodal network architecture was adopted in which local image features extracted by a convolutional neural network were converted into graph relationship data via a dynamic graph constructor, followed by fine-grained information fusion using a multi-layer gating mechanism and a multi-head graph attention mechanism to capture different types of neighborhood relationships from multiple perspectives, while residual connections were incorporated to mitigate the vanishing gradient problem; Subsequently, experimental validation was conducted using the established dataset in a Python environment, and this experimental setup was used to investigate the model’s performance. The results show that, compared to a typical CNN-ResNet model, the proposed model achieves a 3.77% increase in accuracy, a 3.13% increase in macro precision, a 3.35% increase in macro recall, and a 3.23% increase in macro F1 score. It is concluded that the model demonstrates validity and effectiveness, making it suitable for practical engineering applications.
LAO Xiang-yi, MA Jia-qing, ZHAN Yu, HE Zhi-qin, WU Qin-mu, LIU Hong-ju, LI Yong-jie.
Recognition of Ankle Rehabilitation Training Based on Convolutional Neural Networks Combined with Multi-head Graph Attention Residual Networks[J].
Science Technology and Engineering, 2026, 26(27): 11830-11838 DOI:10.12404/j.issn.1671-1815.2507656
从图7中能够清晰直观地展示优化模型训练后20种脚踝状态的精确率、召回率和F1分数。该模型对不同状态下脚踝康复动作的识别评估指标数值均较高,其中跖屈状态下的表现尤为突出,各项指标均接近100%。从图8的ROC曲线(receiver operating characteristic curve)可以看出,各类状态识别的曲线均远离对角线,且整体靠近左上角,说明该预测模型的性能良好;同时各状态的AUC值(area under the curve)均接近1,证明该预测模型能够有效识别出所有脚踝康复状态。尽管对部分动作的识别仍存在一定偏差,但整体识别准确度始终保持在94%以上,满足临床康复训练的精度要求。
ShengY X, WangJ X, TanG G, et al. Muscle synergy plasticity in motor function recovery after stroke[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2024, 32: 1657-1667.
[2]
RahmanS, SarkerS, HaqueA K, et al. AI-driven stroke rehabilitation systems and assessment: a systematic review[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 192-207.
HuangLinqing, HeChen, HuangRenkai. An ankle rehabilitation robot based on a 2-SPS-PU parallel mechanism[J]. Automation Applications, 2025, 66(10): 12-16.
[5]
LiuY, LuW J, DengY J, et al. Configuration design of ankle rehabilitation robots: a systematic review and future prospects[J]. Journal of NeuroEngineering and Rehabilitation, 2025, 22(1): 244.
ZhangTeng, DongChao, JiangShuangpeng, et al. Measurement of gait parameters and analysis of related factors by wearable inertial sensors in patients with chronic external ankle instability[J]. Journal of Practical Orthopaedics, 2025, 31(2): 132-137.
[8]
LeH V. Selected hand gesture recognition model based on cross-evaluation of deep learning from large RGB image datasets[J]. Multimedia Tools and Applications, 2025, 84(32): 1-50.
LiuJinzhi, DongHao, FangWenxuan, et al. Research on an intelligent classification model for lymph node metastasis in rectal cancer based on convolutional neural networks[J]. Science Technology and Engineering, 2022, 22(32): 14328-14338.
Cui ZhanheElisha, MaXinyu, et al. Research on lower limb exoskeleton gait information recognition based on 1D-CNN-SVM[J]. Electronic Measurement Technology, 2025, 48(12): 71-78.
LiJincheng, DaiXuejing, YanRui’ao. A gait recognition algorithm based on 3D-CNN and a fused Transformer[J]. Science Technology and Engineering, 2025, 25(17): 7276-7284.
[17]
张焱. 基于图神经网络的连续手语识别方法研究[D]. 天津: 天津理工大学, 2025.
[18]
ZhangYan. A study on continuous sign language recognition methods based on graph neural networks[D]. Tianjin: Tianjin University of Technology, 2025.
JiaWenwen, WuDong, FanZhanyong. A probabilistic traffic flow forecasting model coupling graph multi-head attention with t-distributed GARCH[J]. Journal of Transportation Science and Engineering, 2026, 42(2): 79-87.
ZhuPing, ZhengWeiguo, QianSongrong. Hierarchical multi-head attention-driven graph-to-text generation[J]. Small and Micro Computer Systems, 2023, 44(10): 2323-2330.
[23]
HeJ, ZhaoL, YangH, et al. HSI-BERT: Hyperspectral image classification using the bidirectional encoder representation from transformers[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58: 165-178.
HuYuchen, LiQiusheng. Research on face expression recognition based on structured deep clustering network[J]. Journal of Gannan Normal University, 2023, 44(6): 56-63.
YanZhilin, NieRencan. Cross-modal contrast representation network for infrared and visible image fusion based on rolling MLP feature extraction[J]. Computer Science, 2025, 52(11): 123-130.
ZhangHuiyi, ZhangJin, HuangJun. A multi-label image classification model based on graph attention network[J]. Journal of Chongqing Technology and Business University(Natural Science Edition), 2022, 39(1): 34-41.
ChenTing, WangTong, ZhangJiwu, et al. Face image restoration based on edge-apriori fusion and dynamic gated features[J]. Research on Computer Applications, 2023, 40(11): 3478-3484.
[34]
许宝卉. 显示器色彩特性分析及色彩空间转换技术研究[D]. 西安: 西安理工大学, 2010.
[35]
XuBohui. Color characterization and color space conversion technology for displays[D]. Xi’an: Xi’an University of Technology, 2010.
LiTao, ZhuWenzhong, CheXuan. Image recognition of apple leaf diseases based on an improved ResNet-50 and transfer learning[J]. Science Technology and Engineering, 2024, 24(24): 10370-10381.
ChenNingning. Research on high-similarityimage recognition technology supported byconvolutional neural networks[J]. Innovation and Application in Science and Technology, 2025, 15(27): 50-53, 59.
[40]
方燕燕. 基于深度学习的多标签胸部X光图像分类方法研究[D]. 淮南: 安徽理工大学, 2025.
[41]
FangYanyan. Research on multi-label chest X-ray image classification methods based on deep learning[D]. Huainan: Anhui University of Science and Technology, 2025.
WangDi. Research on dermoscopy image classification based on an improved ResNet variant network[D]. Baotou: Inner Mongolia University of Science and Technology, 2025.
LiFengshuo. Design and implementation of a deep learning-based image classification system for diabetic retinopathy[D]. Guiyang: Guizhou University, 2025.