中文摘要：

步态冻结（FOG）是晚期帕金森病患者最常见的症状,FOG的突然发作会造成患者的行走障碍。为患者佩戴FOG检测可穿戴设备助手是一种有效可行的治疗途径,当检测到FOG发作时,可穿戴设备助手为患者提供一段有节奏的听觉信号刺激患者恢复行走。针对FOG检测,提出一种系统式的特征学习方法。该方法采用一个基于深度学习的卷积神经网络,对原始输入信号自动地进行特征学习。采用监督式学习方法利用标签信息,使学习到的特征更具识别能力。在整个网络模型中,特征学习和分类互相加强使整个网络更加稳定,更具智能化。通过DAPHNet数据集进行验证,结果表明,该方法可以自动地进行特征学习并识别出步态冻结。与以往的阈值法实验结果相比,平均正确率提高到91.43%,灵敏性提高到85.58%,特异性提高到了93.63%。该方法可以在一定程度上代替人工干预,在处理频繁出现FOG症状的帕金森患者的治疗中具有重要意义。

英文摘要：

The freezing of gait（FOG） is the most common symptoms of late-stage Parkinson＇s disease. The sudden attack of FOG can cause patients walking trouble. It is an effective and feasible treatment method to equip patients with wearable device assistant which can detect FOG. When the FOG attack is detected, the wearable device provides patients with the rhythm of the auditory signal to HELP the recovery of walking. In this article, in view of FOG detection, we proposed a systemic feature learning method. This method used a convolutional neural network based on deep learning to automatically conduct feature learning from the original input signals. And a supervised learning method was adopted to improve learned features＇ recognition capability using tag information. In the entire network model, feature learning and classification reinforced each other to make the whole network more stable and more intelligent, which was verified by the DAPHNet datasets. Compared with the previous threshold method, the average correct rate was increased to 91.43% , the sensitivity was increased to 85.58% and the specificity was increased to 93.63%. To some extent, the proposed method could alleviate the FOG of patients with Parkinson＇s disease, and reduce the number of falls, which is of great significance to improve the ability of daily life of the patients and the quality of life.