基于热释电红外传感器的人体动作识别方法

针对目前人体动作识别技术中存在的隐私暴露、技术复杂度高和识别精度低等相关问题,提出了一种基于热释电红外（PIR）传感器的人体动作识别方法。首先,采用一组安置在天花板上经过视场调制的PIR传感器采集人体运动时散发的红外热辐射信号,将传感器输出的电压模拟信号进行滤波放大后通过ZigBee无线模块传送到PC端打包成原始数据集;其次,将原始数据的两路传感器输出数据进行特征融合,对融合后的数据做标准化处理封装为训练集和测试集;然后,基于数据的特征提出一种两层级联的混合深度学习网络模型作为人体动作的分类算法,第一层采用一维卷积神经网络（1DCNN）对数据进行特征提取,第二层采用门控循环单元（GRU）保存历史输入信息防止丢失有效特征;最后,利用训练集来训练该网络模型得出参数最优的分类模型,通过测试集验证模型的正确性。实验结果表明,提出的该动作识别技术模型对基本动作分类的准确率高于98%,与图像动作识别或穿戴式设备动作识别相比,实现了实时、便捷、低成本和高保密性的高精度人体动作识别。     

Adaptive transfer learning framework for dense prediction of human activity recognition

Human activity recognition (HAR) for dense prediction is proven to be of good performance, but it relies on labeling every point in time series with the high cost. In addition, the performance of HAR model will show significant degradation when tested on the sensor data with different distribution from the training data, where the training data and the test data are usually collected from different sensor locations or sensor users. Therefore, the adaptive transfer learning framework for dense prediction of HAR is introduced to implement cross-domain transfer, where the proposed multi-level unsupervised domain adaptation (MLUDA) approach combines the global domain adaptation and the specific task adaptation to adapt the source and target domain in multiple levels. The multi-connected global domain adaptation architecture is proposed for the first time, which can adapt the output layer of the encoder and the decoder in dense prediction model. After this, the specific task adaptation is proposed to ensure alignment of each class centroid in source domain and target domain by introducing the cosine distance loss and the moving average method. Experiments on three public human activity recognition datasets demonstrate that the proposed MLUDA improves the prediction accuracy of target data by 20% compared to the source domain pre-trained model and it is more effective than the other three deep transfer learning methods with an improvement of 10% to 18% in accuracy.     

基于知识蒸馏与注意力图的雷达信号识别方法

针对传统雷达信号识别方法无法有效进行识别类型扩展问题,该文提出一种基于知识蒸馏与注意力图的雷达信号识别方法。首先将雷达信号的平滑伪Wigner-Ville分布(SPWVD)作为输入;然后设计了基于残差网络的增量学习网络结构,利用基于知识蒸馏与注意力图的损失函数,缓解类别增量过程中的灾难性遗忘;最后采用基于样本特征均值距离的方法对数据集进行管理,有效降低存储资源占用空间。实验表明,该方法能在存储资源有限的情况下,对扩展分类的信号快速完成训练,且对原有分类和扩展分类信号均有良好的识别准确率。     

用于SVM的RCS统计特征集约减方法

传统的支持向量机分类算法对训练样本数目敏感且不具备增量学习的能力,而空间目标识别的工程应用需要积累样本进行大样本的增量学习。根据样本在特征空间分布,提取几何意义上边缘的样本点,成功约减了应用于支持向量机学习的基于雷达散射截面统计特征的训练样本集。利用中心距离比和特征空间多维高斯分布特性,分别提取两类边界样本集和单类边缘样本集;再采用直推式实验设计方法再采样,根据样本潜在结构分布信息选择最能代表样本集结构分布的高价值样本。实验结果表明：样本初选算法能够在有效约减样本集规模的同时保持支持向量机训练分类的精度。     

体域网中一种基于压缩感知的人体动作识别方法

能量有效性是无线体域网在面向长时间健康监测应用的首要挑战。该文引入压缩感知和稀疏表示理论同时解决人体活动监测中的动作识别和数据压缩问题,探索在达到一定动作识别率的同时降低传感器节点的能耗。该文提出的压缩分类动作识别方法首先在传感器节点利用随机投影对传感数据进行压缩,传到中心节点后再利用稀疏表示对压缩采样数据进行分类与识别,可减少传感器节点处理、传输原始数据所带来的能耗。在公开的可穿戴式传感器动作识别数据库WARD(Wearable Action Recognition Database)验证算法性能,实验结果表明该动作识别方法能有效地对随机投影后的低维采样数据进行识别,具有与传统识别方法相比拟的动作识别准确率。     

Multisensor data fusion for physical activity assessment

This paper presents a sensor fusion method for assessing physical activity (PA) of human subjects, based on support vector machines (SVMs). Specifically, acceleration and ventilation measured by a wearable multisensor device on 50 test subjects performing 13 types of activities of varying intensities are analyzed, from which activity type and energy expenditure are derived. The results show that the method correctly recognized the 13 activity types 88.1% of the time, which is 12.3% higher than using a hip accelerometer alone. Also, the method predicted energy expenditure with a root mean square error of 0.42 METs, 22.2% lower than using a hip accelerometer alone. Furthermore, the fusion method was effective in reducing the subject-to-subject variability (standard deviation of recognition accuracies across subjects) in activity recognition, especially when data from the ventilation sensor were added to the fusion model. These results demonstrate that the multisensor fusion technique presented is more effective in identifying activity type and energy expenditure than the traditional accelerometer-alone-based methods.     

Human activity recognition with analysis of angles between skeletal joints using a RGB‐depth sensor

Human activity recognition (HAR) has become effective as a computer vision tool for video surveillance systems. In this paper, a novel biometric system that can detect human activities in 3D space is proposed. In order to implement HAR, joint angles obtained using an RGB‐depth sensor are used as features. Because HAR is operated in the time domain, angle information is stored using the sliding kernel method. Haar‐wavelet transform (HWT) is applied to preserve the information of the features before reducing the data dimension. Dimension reduction using an averaging algorithm is also applied to decrease the computational cost, which provides faster performance while maintaining high accuracy. Before the classification, a proposed thresholding method with inverse HWT is conducted to extract the final feature set. Finally, the K‐nearest neighbor (k‐NN) algorithm is used to recognize the activity with respect to the given data. The method compares favorably with the results using other machine learning algorithms.     

基于增量贝叶斯的双偏振气象雷达降水粒子分类方法

面对降水粒子分类过程中可能存在的样本数不足,样本质量不高的问题,提出一种基于增量贝叶斯的双偏振气象雷达降水粒子分类方法。该方法首先处理有标签的训练数据集,获取属性节点和类节点之间的条件概率表构建朴素贝叶斯分类器;接着使用朴素贝叶斯分类器分类无标签数据,判断类置信度值后将符合条件的数据追加到训练数据集中,最后修正朴素贝叶斯分类器完成增量学习,得到增量贝叶斯分类器实现降水粒子分类。增量贝叶斯分类器不仅能够增加有效的数据样本,还能够及时更新分类器从而提高其泛化性和适应性,分类结果的准确性也得到了一定的改善。     

基于自组织增量-图卷积神经网络的金相图半监督学习

采用深度学习对钢铁材料显微组织图像分类,需要大量带标注信息的训练集。针对训练集人工标注效率低下问题,该文提出一种新的融合自组织增量神经网络和图卷积神经网络的半监督学习方法。首先,采用迁移学习获取图像数据样本的特征向量集合;其次,通过引入连接权重策略的自组织增量神经网络(WSOINN)对特征数据进行学习,获得其拓扑图结构,并引入胜利次数进行少量人工节点标注;然后,搭建图卷积网络(GCN)挖掘图中节点的潜在联系,利用Dropout手段提高网络的泛化能力,对剩余节点进行自动标注进而获得所有金相图的分类结果。针对从某国家重点实验室收集到的金相图数据,比较了在不同人工标注比例下的自动分类精度,结果表明:在图片标注量仅为传统模型12%时,新模型的分类准确度可达到91%。     

Implementation of a real-time human movement classifier using a triaxial accelerometer for ambulatory monitoring.

The real-time monitoring of human movement can provide valuable information regarding an individual's degree of functional ability and general level of activity. This paper presents the implementation of a real-time classification system for the types of human movement associated with the data acquired from a single, waist-mounted triaxial accelerometer unit. The major advance proposed by the system is to perform the vast majority of signal processing onboard the wearable unit using embedded intelligence. In this way, the system distinguishes between periods of activity and rest, recognizes the postural orientation of the wearer, detects events such as walking and falls, and provides an estimation of metabolic energy expenditure. A laboratory-based trial involving six subjects was undertaken, with results indicating an overall accuracy of 90.8% across a series of 12 tasks (283 tests) involving a variety of movements related to normal daily activities. Distinction between activity and rest was performed without error; recognition of postural orientation was carried out with 94.1% accuracy, classification of walking was achieved with less certainty (83.3% accuracy), and detection of possible falls was made with 95.6% accuracy. Results demonstrate the feasibility of implementing an accelerometry-based, real-time movement classifier using embedded intelligence.     

基于受限玻尔兹曼机的专家乘积系统的一种改进算法

深度学习在高维特征向量的信息提取和分类中具有很强的能力,但深度学习训练时间也比较长,超参数搜索空间大,从而导致超参数寻优较困难。针对此问题,该文提出一种基于受限玻尔兹曼机(RBM)专家乘积系统的改进方法。先将专家乘积系统原理与RBM算法相结合,采用全是真实概率值的参数更新方式会引起模型识别效果不理想和带来密度问题,为此将其更新方式进行改进;为加快网络收敛和提高模型识别能力,采取在RBM预训练阶段和微调阶段引入不同组合方式动量项的一种改进算法。通过对MNIST数据库中的0~9的手写数字体的识别和CMU-PIE数据库的人脸识别实验,提出的算法减少了学习时间,提高了超参数寻优的效率,进而构建的深层网络能获得较好的分类效果。试验结果表明,提出的改进算法在处理高维大量的数据时,计算效率有较大提高,其算法有效。     

改进非负矩阵分解的神经网络人脸识别

为了提高人脸识别正确率,提出基于改进非负矩阵分解的神经网络人脸识别算法。首先利用改进的非负矩阵分解对人脸图像进行特征提取,提高非负矩阵分解速度。接着将提取出的特征信息作为神经网络学习入口进行特征训练,由于神经网络在学习过程中,容易出现局部最小值且收敛速度慢等问题,为此采用改进的遗传算法对神经网络进行优化处理,获得最终的人脸识别结果。实验结果表明:利用改进的非负矩阵分解方法能够降低神经网络的分类训练负荷量和运算量,提高人脸识别识别率。通过和各种方法比较可知,本方法的人脸识别率都较高。本方法人脸特征分解速度快,提高了神经网络训练前期精度和收敛速度,使得人脸识别正确率高。当特征向量个数达到40以上时,人脸识别正确率保持95%以上。     

A Supervised Learning Based Decision Support System for Multi-Sensor Healthcare Data from Wireless Body Sensor Networks

Wireless body sensor network (WBSN) is also known as wearable sensors with transmission capabilities, computation, storage and sensing. In this paper, a supervised learning based decision support system for multi sensor (MS) healthcare data from wireless body sensor networks (WBSN) is proposed. Here, data fusion ensemble scheme is developed along with medical data which is obtained from body sensor networks. Ensemble classifier is taken the fusion data as an input for heart disease prediction. Feature selection is done by the squirrel search algorithm which is used to remove the irrelevant features. From the sensor activity data, we utilized the modified deep belief network (M-DBN) for the prediction of heart diseases. This work is implemented by Python platform and the performance is carried out of both proposed and existing methods. Our proposed M-DBN technique is compared with various existing techniques such as Deep Belief Network, Artificial Neural Network and Conventional Neural Network. The performance of accuracy, recall, precision, F1 score, false positive rate, false negative and true negative are taken for both proposed and existing methods. Our proposed performance values for accuracy (95%), precision (98%), and recall (90%), F1 score (93%), false positive (72%), false negative (98%) and true negative (98%).

     

Activity classification using realistic data from wearable sensors

Automatic classification of everyday activities can be used for promotion of health-enhancing physical activities and a healthier lifestyle. In this paper, methods used for classification of everyday activities like walking, running, and cycling are described. The aim of the study was to find out how to recognize activities, which sensors are useful and what kind of signal processing and classification is required. A large and realistic data library of sensor data was collected. Sixteen test persons took part in the data collection, resulting in approximately 31 h of annotated, 35-channel data recorded in an everyday environment. The test persons carried a set of wearable sensors while performing several activities during the 2-h measurement session. Classification results of three classifiers are shown: custom decision tree, automatically generated decision tree, and artificial neural network. The classification accuracies using leave-one-subject-out cross validation range from 58 to 97% for custom decision tree classifier, from 56 to 97% for automatically generated decision tree, and from 22 to 96% for artificial neural network. Total classification accuracy is 82% for custom decision tree classifier, 86% for automatically generated decision tree, and 82% for artificial neural network.     

Enhancing Representation of Deep Features for Sensor-Based Activity Recognition

Sensor-based activity recognition (AR) depends on effective feature representation and classification. However, many recent studies focus on recognition methods, but largely ignore feature representation. Benefitting from the success of Convolutional Neural Networks (CNN) in feature extraction, we propose to improve the feature representation of activities. Specifically, we use a reversed CNN to generate the significant data based on the original features and combine the raw training data with significant data to obtain to enhanced training data. The proposed method can not only train better feature extractors but also help better understand the abstract features of sensor-based activity data. To demonstrate the effectiveness of our proposed method, we conduct comparative experiments with CNN Classifier and CNN-LSTM Classifier on five public datasets, namely the UCIHAR, UniMiB SHAR, OPPORTUNITY, WISDM, and PAMAP2. In addition, we evaluate our proposed method in comparison with traditional methods such as Decision Tree, Multi-layer Perceptron, Extremely randomized trees, Random Forest, and k-Nearest Neighbour on a specific dataset, WISDM. The results show our proposed method consistently outperforms the state-of-the-art methods.

     

一种基于超宽带雷达的日常活动分类方法

近年来,随着智能物联网应用的快速发展,雷达传感器由于具有保护隐私、全天候全天时工作、不受光线和遮挡影响等优点,在人体目标日常行为活动识别方面受到了学术界和产业界的极大重视.针对一种超低辐射的超宽带雷达(Impulse Radio Ultra-Wideband,IR-UWB),提出了一种室内人员日常活动(包含静止、坐下、...     

基于串联式一维神经网络的毫米波雷达动态手势识别方法

现有的基于雷达传感器的手势识别方法,大多先利用雷达回波对手势的距离、多普勒和角度等信息进行参数估计,得到各种数据谱图,然后再利用卷积神经网络对这些谱图进行分类,实现过程较为复杂。该文提出一种基于串联式1维神经网络(1D-ScNN)的毫米波雷达动态手势识别方法。首先基于毫米波雷达获取动态手势的原始回波,然后利用1维卷积和池化操作对手势特征进行提取,并将这些特征信息输入1维Inception v3结构。最后在网络的末端接入长短期记忆(LSTM)网络来聚合1维特征,充分利用动态手势的帧间相关性,提高识别准确率和训练收敛速度。实验结果表明,该方法实现过程简单,收敛速度快,识别准确率可以达到96.0%以上,高于现有基于数据谱图的手势分类方法。