First and second order dynamics in a hierarchical SOM system for action recognition

Human recognition of the actions of other humans is very efficient and is based on patterns of movements. Our theoretical starting point is that the dynamics of the joint movements is important to action categorization. On the basis of this theory, we present a novel action recognition system that employs a hierarchy of Self-Organizing Maps together with a custom supervised neural network that learns to categorize actions. The system preprocesses the input from a Kinect like 3D camera to exploit the information not only about joint positions, but also their first and second order dynamics. We evaluate our system in two experiments with publicly available datasets, and compare its performance to the performance with less sophisticated preprocessing of the input. The results show that including the dynamics of the actions improves the performance. We also apply an attention mechanism that focuses on the parts of the body that are the most involved in performing the actions.     

Output feedback controller for polytopic systems exploiting the direct searching of the design space

This paper deals with the problem of designing a robust static output feedback controller for polytopic systems. The current research that tackled this problem is mainly based on LMI method, which is conservative by nature. In this paper, a novel approach is proposed, which considers the design space of the controller parameters and iteratively partitions the space to small simplexes. Then, by assessing the stability in each simplex, the solution space for design parameters is directly determined. It has been theoretically proved that, if there exists a feasible solution in the design space, the algorithm can find it. To validate the result of the proposed approach, comparative simulation examples are given to illustrate the performance of the design methodology as compared to those of previous approaches.     

Speech / music classification using speech-specific features

This paper proposes the use of speech-specific features for speech / music classification. Features representing the excitation source, vocal tract system and syllabic rate of speech are explored. The normalized autocorrelation peak strength of zero frequency filtered signal, and peak-to-sidelobe ratio of the Hilbert envelope of linear prediction residual are the two source features. The log mel energy feature represents the vocal tract information. The modulation spectrum represents the slowly-varying temporal envelope corresponding to the speech syllabic rate. The novelty of the present work is in analyzing the behavior of these features for the discrimination of speech and music regions. These features are non-linearly mapped and combined to perform the classification task using a threshold based approach. Further, the performance of speech-specific features is evaluated using classifiers such as Gaussian mixture models, and support vector machines. It is observed that the performance of the speech-specific features is better compared to existing features. Additional improvement for speech / music classification is achieved when speech-specific features are combined with the existing ones, indicating different aspects of information exploited by the former.     

基于MED和边际谱的自动倾斜器轴承故障诊断

针对直升机自动倾斜器轴承早期微弱故障特征易被强烈背景噪声淹没的问题，提出了一种基于最小熵反褶积（Minimum Entropy Deconvolution，MED）和边际谱的自动倾斜器轴承故障诊断方法。采用MED对采集的振动信号进行滤波降噪，提高了信号的信噪比，突出了轴承早期微弱故障特征；通过Hilbert变换和经验模态分解（Empirical Mode Decomposition，EMD）获取去噪包络信号的本征模态函数（Intrinsic Mode Functions，IMF）集，并引入峭度筛选准则选取合理IMF集计算局部Hilbert边际谱，有效地提取了故障特征频率，能够通过故障特征频率进行故障类型判别。通过某型直升机自动倾斜器故障诊断试验系统验证了该诊断方法的合理性和可行性。     

Explicit bounds for separation between Oseledets subspaces

We consider a two-sided sequence of bounded operators in a Banach space which are not necessarily injective and satisfy two properties (SVG) and (FI). The singular value gap (SVG) property says that two successive singular values of the cocycle at some index d admit a uniform exponential gap; the fast invertibility (FI) property says that the cocycle is uniformly invertible on the fastest d-dimensional direction. We prove the existence of a uniform equivariant splitting of the Banach space into a fast space of dimension d and a slow space of codimension d. We compute an explicit constant lower bound on the angle between these two spaces using solely the constants defining the properties (SVG) and (FI). We extend the results obtained by Bochi and Gourmelon in the finite-dimensional case for bijective operators and the results obtained by Blumenthal and Morris in the infinite dimensional case for injective norm-continuous cocycles, in the direction that the operators are not required to be globally injective, that no dynamical system is involved and no compactness of the underlying system or smoothness of the cocycle is required. Moreover we give quantitative estimates of the angle between the fast and slow spaces that are new even in the case of finite-dimensional bijective operators in Hilbert spaces.     

基于SWT和Hilbert的密封磁阻马达式CRDM振动信号去噪

针对控制棒驱动机构滚轮振动信号采集过程受到噪声干扰的问题，提出了小波半软阈值（SWT）和Hilbert变换法相结合的去噪方法。该方法首先利用小波半软阈值算法在时频域对滚轮振动信号进行降噪处理，然后进行Hilbert变换求出其包络谱，分析寿命试验与缺陷验证试验中不同的滚轮振动信号。试验结果表明，该方法可有效消除噪声对振动信号的干扰，证明了小波半软阈值与Hilbert变换相结合的方法在驱动机构滚轮状态识别和故障诊断中的有效性，为驱动机构的状态判别提供了理论支持。     

Time–frequency interpretation of multi-frequency signal from rotating machinery using an improved Hilbert–Huang transform

The Hilbert–Huang transform (HHT) has proven to be a promising tool for the analysis of non-stationary signals commonly occurred in industrial machines. However, in practice, multi-frequency intrinsic mode functions (IMFs) and pseudo IMFs are likely generated and lead to grossly erroneous or even completely meaningless instantaneous frequencies, which raise difficulties in interpreting signal features by the HHT spectrum. To enhance the time–frequency resolution of the traditional HHT, an improved HHT is proposed in this study. By constructing a bank of partially overlapping bandpass filters, a series of filtered signals are obtained at first. Then a subset of filtered signals, each associated with certain energy-dominated components, are selected based on the maximal-spectral kurtosis–minimal-redundancy criterion and the information-related coefficient, and further decomposed by empirical mode decomposition to extract sets of IMFs. Furthermore, IMF selection scheme is applied to select the relevant IMFs on which the HHT spectrum is constructed. The novelty of this method is that the HHT spectrum is just constructed with the relevant, almost monochromatic IMFs rather than with the IMFs possibly with multiple frequency components or with pseudo components. The results on the simulated data, test rig data, and industrial gearbox data show that the proposed method is superior to the traditional HHT in feature extraction and can produce a more accurate time–frequency distribution for the inspected signal.     

Boolean operations on rasterized shapes represented by chain codes using space filling curves

This paper introduces a new algorithm for Boolean operations on rasterized geometric shapes that are represented with chain codes. The algorithm works in three steps. Firstly, the chain code symbols are transformed in the Hilbert space, where the overlaid chain code symbols are recognised. After that, a suitable starting cell is determined. Finally, the walk-about through the sequence of the initial chain code symbols is performed to obtain the sequence of chain code symbols representing the shape of the required Boolean operation. The algorithm is demonstrated on Freeman chain code in four directions. The time and space complexity of the proposed algorithm is linear, which was proven theoretically and confirmed by experiments.     

Model development in scientific discovery learning with a computer-based physics task

Based on theories of scientific discovery learning (SDL) and conceptual change, this study explores students' preconceptions in the domain of torques in physics and the development of these conceptions while learning with a computer-based SDL task. As a framework we used a three-space theory of SDL and focused on model space, which is supposed to contain the current conceptualization/model of the learning domain, and on its change through hypothesis testing and experimenting. Three questions were addressed: (1) What are students' preconceptions of torques before learning about this domain? To do this a multiple-choice test for assessing students' models of torques was developed and given to secondary school students (N = 47) who learned about torques using computer simulations. (2) How do students' models of torques develop during SDL? Working with simulations led to replacement of some misconceptions with physically correct conceptions. (3) Are there differential patterns of model development and if so, how do they relate to students’ use of the simulations? By analyzing individual differences in model development, we found that an intensive use of the simulations was associated with the acquisition of correct conceptions. Thus, the three-space theory provided a useful framework for understanding conceptual change in SDL.