摘要:
在诸多物联网实际应用中，原始采集信号数据多含有大量噪声，特别是在运动相关场景里.需从含大量噪声的一维时序信号中对有效信号活动区域起止点进行准确识别，以支持相关分析.已有的基于双阈值规则的识别方法对噪声十分敏感，噪声的存在会导致计算出的识别阈值无法匹配非噪声段的原始数据，从而导致将随机噪声数据识别为信号活动区间或者漏检信号活动区间.基于机器学习和深度学习的识别方法需要大量的样本数据，在样本量较小的物联网场景中模型会产生欠拟合问题，从而降低识别精度.为了对含有大量噪声且数据量少的一维时序信号中的信号活动区间进行准确识别，提出了一种基于局部动态阈值的信号活动区间识别方法EasiLTOM(signal activity interval recognition based on local dynamic threshold).该方法基于局域信号计算识别阈值，并使用最短信号长度对噪声尖峰进行过滤，可避免随机噪声对信号活动区间识别的影响，解决漏检和误检问题，从而提高识别精度.此外，EasiLTOM方法所需数据量小，适用于数据稀少的物联网场景.为验证EasiLTOM方法的有效性，该研究于3个月间采集了14人次的表面肌电数据，并使用2个公开数据集进行了对比实验.结果表明：EasiLTOM方法对信号活动区间可达到平均93.17%的识别精度，相对于现有的双阈值和机器学习方法，分别提升了15.03%和4.70%，在运动分析相关场景中具有实用价值.
Abstract:
In many practical applications of the Internet of things, the original collected signal data contains a lot of noise, especially in motion-related scenes. It is necessary to accurately identify the start and end points of the effective signal activity area from the one-dimensional time series signal with a lot of noise to support the relevant analysis. Existing recognition methods based on dual threshold rules are very sensitive to noise. The presence of noise will cause the calculated recognition threshold to fail to match the original data of the non-noise segment, which leads to the recognition of random noise data as signal activity intervals or missed signal activity interval. Recognition methods based on machine learning and deep learning require a large amount of sample data. In IoT scenarios with a small sample size, the model will have underfitting problems, thereby reducing recognition accuracy. In order to accurately identify the signal activity interval in a one-dimensional time series signal with a lot of noise and a small amount of data, a signal activity interval recognition based on local dynamic threshold EasiLTOM is proposed. This method calculates the recognition threshold based on the local signal, and it uses the shortest signal length to filter noise spikes, which can avoid the influence of random noise on the recognition of signal activity intervals, solve the problems of missed detection and false detection, and improve the recognition accuracy. In addition, EasiLTOM requires a small amount of data, which is suitable for IoT scenarios with scarce data. In order to verify the effectiveness of EasiLTOM, this study collects surface EMG data of 14 people in 3 months, and conducts comparative experiments using two public data sets. The results show that EasiLTOM method can achieve an average recognition accuracy of 93.17% for the signal activity range, which is 15.03% and 4.70% higher than the existing dual threshold and machine learning methods, and has practical value in motion analysis related scenes.
百度学术
下载:
