采用8031作主控器的时域FIR数字滤波器

长期以来，ＦＩＲ数字滤波器大多是在频域上实现。因为，在时域上实现ＦＩＲ数字滤波所遇到的首要问题，是输入信号序列与冲激响应序列的卷积运算速度难以提高。然而，随着超大规模集成电路的飞速发展，硬件集成度与运算速度获得极大的改观，在时域上实现ＦＩＲ数字滤波已成为可能。ＩＭＳＡ１００是高速、高精度３２级数字信号处理器，是完成卷积运算的理想器件。本文论证了用ＩＭＳＡ１００实现时域ＦＩＲ数字滤波器的可行性和硬件设计中的一些问题，并给出了应用举例。在设计中选用８０３１单片机做主控器，大大提高了性能价格比，使这一设计具有很高的实用价值。     

Discrete singular convolution and its application to the analysis of plates with internal supports. Part 2: Applications

Part 2 of this series of two papers presents the applications of the discrete singular convolution (DSC) algorithm. The main purpose of this paper is to explore the utility, test the accuracy and examine the convergence of the proposed approach for the vibration analysis of rectangular plates with internal supports. Both partial internal line supports and complex internal supports are considered for 21 square plates of various combinations of edge support conditions. The effects of different size, shape and topology of the internal supports and different boundary conditions on the vibration response of plates are investigated. The partial internal line supports may vary from a central point support to a full range of cross or diagonal line supports. Several closed‐loop supports, such as ring, square and rhombus, and their combinations are studied for complex internal supports. Convergence and comparison studies are carried out to establish the correctness and accuracy of the DSC algorithm. The DSC results are compared with those in the available literature obtained by using other methods. Numerical results indicate that the DSC algorithm exhibits controllable accuracy for plate analysis and shows excellent flexibility in handling complex geometries, boundary conditions and support conditions. Copyright © 2002 John Wiley & Sons, Ltd.     

基于滑窗结构的软输出维特比算法的实现

文章在对软输出维特比算法(SOVA)进行推导的基础上，分析了软信息的提取过程。同时从硬件实现的角度考虑，提出了一种基于滑动窗1：3结构的SOVA算法实现方案，该算法大大降低了算法实现复杂度和译码延迟，同时通过调整滑窗参数，可以取得与非滑窗SOVA算法几乎相同的性能。     

信号时域积分性质的一个推广应用

对信号进行运算（卷积运算，傅氏变换，拉纸变换）时，应用时域积分性质，得出由函数导数的运算式求取原函数的相应运算结果的一般性计算公式，使时域积分性质的应用得到推广。     

Convolutional solutions of partial difference equations

A significant part of the theory of one-dimensional linear shift-invariant systems is based on the concept of weighting function (or impulse response): the output is the convolution of the weighting function with the input. This paper introduces the concept of linear translation-invariant systems and uses this notion in studying impulse response, z-transforms, and transfer functions for multidimensional systems.     

A combination of differential equations and convolution in understanding the spread of an epidemic

Nonlinear dynamical method of projecting the transmission of an epidemic is accurate if the input parameters and initial value variables are reliable. Here, such a model is proposed for predicting an epidemic. A method to supplement two variables and two parameters for this proposed model is demonstrated through a robust statistical approach. The method described here worked well in case of three continuous distributions. Model predictions could be lower estimates due to under-reporting of disease cases. Anad hoc procedure with a technical note is provided in the appendix An earlier version of this paper was presented at the annual conference of the Indian Society for Mathematical Modelling & Computer Simulation, Bangalore, November 14–15, 2002     

稀疏差分网络和多监督哈希用于高效图像检索

针对基于深度哈希的图像检索中卷积神经网络（CNN）特征提取效率较低和特征相关性利用不充分的问题,提出一种融合稀疏差分网络和多监督哈希的新方法SDNMSH（sparse difference networks and multi-supervised hashing）,并将其用于高效图像检索。SDNMSH以成对的图像作为训练输入,通过精心设计的稀疏差分卷积神经网络和一个监督哈希函数来指导哈希码学习。稀疏差分卷积神经网络由稀疏差分卷积层和普通卷积层组成。稀疏差分卷积层能够快速提取丰富的特征信息,从而实现整个网络的高效特征提取。同时,为了更加充分地利用语义信息和特征的成对相关性,以促进网络提取的特征信息能够更加有效地转换为具有区分性的哈希码、进而实现SDNMSH的高效图像检索,采用一种多监督哈希（MSH）函数,并为此设计了一个目标函数。在MNIST、CIFAR-10和NUS-WIDE三个广泛使用的数据集上进行了大量的对比实验,实验结果表明,与其他先进的深度哈希方法相比,SDNMSH取得了较好的检索性能。     

Deconvolution with partially known kernel of nonnegative signals

This paper extends the Bayesian direct deconvolution method to the case in which the convolution kernel depends on few unknown nuisance parameters. Moreover, an acceleration procedure is proposed that drastically reduces the computational burden. Finally, the implementation of the method by means of the fast Fourier transform is fully discussed in the multidimensional case.     

Multi-Scale Dilated Convolutional Neural Network for Hyperspectral Image Classification

基于多尺度空洞卷积神经网络的高光谱图像分类

郑姗姗¹,刘文¹,单锐¹,赵静一²,江国乾³,张智⁴

（1. 燕山大学理学院,河北 秦皇岛 066004;2. 燕山大学机械工程学院,河北 秦皇岛 066004;3. 燕山大学电气工程学院,河北 秦皇岛 066004;4. 北京航天研究所,北京 100094）

创新点说明：

1）将图像分割方法——空洞卷积用于卷积神经网络进行高光谱图像分类,提取更加广泛、抽象的图像特征。

2）构建基于多尺度空洞卷积神经网络的高光谱图像分类方法。搭建多尺度聚合结构,在每个通道中使用快捷连接和空洞卷积结构,有效提取图像特征,避免信息丢失。

研究目的：

针对图像信息丢失问题,得到高精度的高光谱图像分类方法。

研究方法：

在Indian Pines和Pavia University数据集上,与4个已有的高光谱图像分类方法进行对比实验,比较OA, AA和Kappa值。

研究结果：

多尺度空洞卷积神经网络在Indian Pines和Pavia University数据集上OA值分别达到了99.58%,99.92%。AA值分别达到了99.57%,99.90%。Kappa分别达到了99.52%,99.89%。

结论：

1）在卷积神经网络中引入空洞卷积,可以有效避免图像信息丢失。

2）多尺度空洞卷积神经网络能提取更佳的判别性特征,实现高分类性能。

关键词：多尺度聚合;空洞卷积;高光谱图像分类;快捷连接