Search

Most neural network-based classifiers extract features using several hidden layers and make predictions at the output layer by utilizing these extracted features. We observe that not all features are equally pronounced in all classes; we call such features class-specific features. Existing models do not fully utilize the class-specific differences in features as they feed all extracted features from the hidden layers equally to the output layers. Recent attention mechanisms allow giving different emphasis (or attention) to different features, but these attention models are themselves class-agnostic. In this paper, we propose a novel class-specific attention (CSA) module to capture significant class-specific features and improve the overall classification performance of time series. The CSA module is designed in a way such that it can be adopted in existing neural network (NN) based models to conduct time series classification. In the experiments, this module is plugged into five start-of-the-art neural network models for time series classification to test its effectiveness by using 40 different real datasets. Extensive experiments show that an NN model embedded with the CSA module can improve the base model in most cases and the accuracy improvement can be up to 42%. Our statistical analysis show that the performance of an NN model embedding the CSA module is better than the base NN model on 67% of MTS and 80% of UTS test cases and is significantly better on 11% of MTS and 13% of UTS test cases., Comment: 12 pages

Based on the joint phase interferometer and multiple signal classification (MUSIC) algorithm, two high-performance 3-dimensional (3-D) parameters estimation and localization algorithms for near field sources (NFS) are proposed in this study employing the uniform circular array (UCA). Firstly, the algebraic relation is demonstrated between the azimuth angle under the far-field assumption and the actual near-field source. Using the relation, we estimate the far-field azimuth angle using the 2-dimensional (2-D) MUSIC method and then obtain the estimation result of the azimuth angle for NFS. Secondly, we select two groups of antenna pairs whose distances are less than half of the wavelength, which are called short baselines in the interferometer method. Using a phase detector, the phase differences of the two baselines are measured, which have no phase ambiguity. Then it can estimate the elevation angle and range depending on the algebraic relations between the elevation angle, range and phase differences. The foregoing short-baseline method is qualified to localize a NF source. In addition, in order to get better performance, we propose a long-baseline method with further research. Two groups of antenna pairs whose distances are greater than half of the wavelength are selected as two long baselines. By solving the phase ambiguity, the elevation angle and range can finally be estimated with high precision. The simulation results illustrate that the proposed algorithms can achieve higher localization accuracy than the conventional 3-D MUSIC method. Moreover, the long-baseline method performs better than the short-baseline method. And it is also presented that the computational complexity of the proposed methods can be reduced owing to the fact that the 3-D spectrum peak searching process in the conventional method is not required any more.

Railway signal cables are the infrastructure of railway operations. Signal cable fault detection is one of the important guarantees for the safe operation of railways. The rapid investigation and maintenance of railway signal cable faults are of great significance to the safe operation of trains. This article uses time domain reflectometry (TDR) and signal processing technology to design a portable railway signal cable fault detector, it can quickly and conveniently detect open circuit and short-circuit faults of railway cables on site, and can locate fault points. The detector uses DSP as the main control chip, equipped with signal conditioning circuit, signal processing circuit and so on. Experiments have been carried out in the laboratory to verify that the detector can realize fault detection and location functions.

By introducing Cu and Fe ions into the Polyvinylpyrrolidone (PVP) polymer, its photonic band gap is reduced from 5.0 eV to 3.4 and 2.7 eV respectively, whereby the absorption and optical sensitivity in visible are greatly enhanced. After laser irradiation, the formed composite can be cross-linked and solidified. High quality photonic crystals are produced in Fe3+-PVP composite film. Our research will help us understanding the interaction mechanism between metallic ions and polymers under laser irradiation. The composite material system will find wide applications in polymer-based optical devices.

In China, libraries can be grouped into three major categories: (1) the public libraries, headed by the National Library of China and including the provincial, municipal, prefectural, and county libraries; (2) the academic libraries under the control of the Ministry of Education, including the university and college libraries and middle and elementary school libraries; and (3) the libraries of the Chinese Academy of Sciences (Academic Sinica), including the science and research libraries. Academic libraries play a very special role in the country among these groups of libraries and are becoming increasingly important, as they are at the centre of information and education on each campus (Ma 1989).