The effect of training data set size and the complexity of the separation function on neural network classification capability: The two-group case

Classification among groups is a crucial problem in managerial decision making. Classification techniques are used in: identifying stressed firms, classifying among consumer types, and rating of firms' bonds, etc. Neural networks are recognized as important and emerging methodologies in the area of classification. In this paper, we study the effect of training sample size and the neural network topology on the classification capability of neural networks. We also compare neural network capabilities with those of commonly used statistical methodologies. Experiments were designed and carried out on two-group classification problems to find answers to these questions. The prediction capability of the neural network models are better than traditional statistical models. The learning capability of the neural networks is improving compared to traditional models because the discriminate function is more complex. For real world classification problems, the usage of neural networks is highly recommended, for two reasons: learning capability and flexibility. Learning capability: Neural network classifies better in sterile experiments as performed in this research. Flexibility: Real life data are rarely not contaminated with noise, such as unknown distributions, and missing variables, etc. Neural networks differ from a statistical model that it is not dependent on any assumption concerning the data set distribution. © 1997 John Wiley & Sons, Inc. Naval Research Logistics 44: 699–717, 1997     

基于RBF网络的导弹攻击区解算及对比分析

研究了神经网络在空空导弹攻击区处理中的运用,在提出多层前馈(MF)网络和径向基函数(RBF)网络设计方案的基础上,着重对MF网络的BP算法、改进BP算法以及RBF网络的应用进行了仿真,并从收敛速度、逼近精度和实时性三个方面,对计算结果进行了对比分析,结论表明RBF网络在导弹攻击区处理中具有更高的可行性,更适用于快速高精度的火控解算。     

BP人工神经网络的应用及其实现技术

简述了BP人工神经网络的算法原理 ,利用Batchnet程序对一个三层BP人工神经网络进行了设计 ,应用Matlab中人工神经网络工具箱设计了一个用于函数逼近的BP人工神经网络 ,最后 ,运用Matlab中可视化工具Simulink对一个BP人工神经网络例子进行了仿真 ,通过例子 ,探讨了BP人工神经网络的应用并介绍目前几种实用的BP人工神经网络实现技术 .     

一种基于神经网络的磁性目标定位方法

磁性目标定位问题可归结为一类非线性规划问题的求解 .该问题的最大特点就是其目标函数的计算过程极为繁琐 .而目标函数计算的快慢对磁定位的实时性有很大影响 .在详细研究了目标函数之后 ,给出了目标函数的一种神经网络结构实现 .由于该网络结构易于并行计算和VLSI实现 ,从而可使磁定位的实时性得到改善 .     

导弹故障智能诊断的一种实现方法

研究了基于神经网络的故障诊断专家系统的理论和方法及具体实现过程。将此方法应用于导弹武器系统故障智能诊断之中 ,结果表明该方法是一种行之有效的故障智能诊断方法。     

跨模态检索中的相似性漂移问题

为了降低“相似性漂移”问题的影响,提出一种基于“邻域传播”的匹配策略,将待查询项的模态内近邻映射到目标空间中,并将它们在目标空间中的最近邻作为查询项的跨模态近邻。基于邻域传播的匹配策略在不改变跨模态映射函数的条件下,可以有效地降低“相似性漂移”带来的误匹配现象。理论和实验分析证明,跨模态映射函数的“相似性漂移”问题广泛存在,而基于“邻域传播”的匹配策略可以有效降低其影响,提高匹配的准确率。     

图像超分辨率卷积神经网络加速算法

为了实现模型的实时和嵌入式运行,提出了一种轻量级的卷积神经网络结构。通过采用较小的滤波器尺寸和引入深度可分离卷积,可大量减少模型参数,提高模型非线性表达能力;在网络末端引入子像素卷积层,直接从原始低分辨率图像学习到高分辨率图像的映射,计算成本为原来的1/k²(k为放大因子)。在Set5数据集上的实验表明,所提模型的速度较经典的图像超分辨率重建算法速度提高了25.8倍,能够在通用GPU上实时运行,峰值信噪比平均提高了0.17 dB,并且参数只有它的35%。     

基于多通道卷积神经网络的磁性舰船目标运动参数估计

传统磁性目标运动估计效果依赖于目标的初始状态信息,为克服这一缺陷,建立磁性运动目标三分量投影模型,并据此生成磁性舰船运动目标在运动速度、航向、信噪比等参数变化情况下的10类目标的训练数据集、验证数据集以及测试数据集。设计多通道卷积神经网络,对目标的正横距离和运动速度进行估计,并比较和分析了不同的学习方式和激活函数对网络性能的影响。结果表明:Adam+tanh组合方式的估计性能要优于其他组合方式,而且对磁性目标运动参数的估计效果比较精确,此方法相较于卡尔曼滤波、粒子滤波等估计算法的优越性在于运算复杂度低以及参数估计不需要目标初始状态信息。     

Arbitrary-oriented target detection in large scene sar images

Target detection in the field of synthetic aperture radar (SAR) has attracted considerable attention of researchers in national defense technology worldwide, owing to its unique advantages like high resolution and large scene image acquisition capabilities of SAR. However, due to strong speckle noise and low signal-to-noise ratio, it is difficult to extract representative features of target from SAR images, which greatly inhibits the effectiveness of traditional methods. In order to address the above problems, a framework called contextual rotation region-based convolutional neural network (RCNN) with multilayer fusion is proposed in this paper. Specifically, aimed to enable RCNN to perform target detection in large scene SAR images efficiently, maximum sliding strategy is applied to crop the large scene image into a series of sub-images before RCNN. Instead of using the highest-layer output for proposal generation and target detection, fusion feature maps with high resolution and rich semantic information are constructed by multilayer fusion strategy. Then, we put forwards rotation anchors to predict the minimum circumscribed rectangle of targets to reduce redundant detection region. Furthermore, shadow areas serve as contextual features to provide extraneous information for the detector identify and locate targets accurately. Experimental results on the simulated large scene SAR image dataset show that the proposed method achieves a satisfactory performance in large scene SAR target detection.     

二维矩阵卷积在向量处理器中的设计与实现

为了加快卷积神经网络模型的计算速度,便于大规模神经网络模型在嵌入式微处理器中的实现,以FT-matrix2000向量处理器体系结构为研究背景,通过对多核向量处理器体系结构的分析和对卷积神经网络算法的深入研究,提出将规模较小的卷积核数据置于标量存储体,尺寸较大的卷积矩阵置于向量存储体的数据布局方案。针对矩阵卷积中数据难以复用的问题,提出根据卷积核移动步长的不同动态可配置的混洗模式,通过对所取卷积矩阵元素进行不同的移位操作,进而大幅提高卷积矩阵数据的复用率。针对二维矩阵卷积由于存在数据相关性进而难以多核并行的问题,提出将卷积矩阵多核共享,卷积核矩阵多核独享的多核并行方案。设计了卷积核尺寸不变、卷积矩阵规模变化和卷积矩阵尺寸不变、卷积核规模变化的两种计算方式,并在主流CPU、GPU、TI6678、FT-matrix2000平台进行了性能对比与分析。实验结果表明:FT-matrix2000相比CPU最高可加速238倍,相比TI6678可加速21倍,相比GPU可加速663 805倍。