基于神经网络的典型地物的偏振多光谱图像分类（特邀）

相比传统的多光谱成像探测,偏振多光谱成像探测方法可以探测目标表面的粗糙度、含水量等更多信息,给目标检测带来了很大便利,但目前主要用于目标探测,尚未广泛应用于目标分类。BP神经网络是目前常用的一种典型神经网络,可以建立从端到端的映射,在训练样本集足够大的前提下,训练完毕且效果良好的神经网络是一种高效、精确、快速的工具。首先,利用基于旋转偏振片和滤波片的偏振光谱成像探测系统获取了典型地物的偏振多光谱图像,对图像进行了预处理,建立了数据集;其次,在该数据集上进行了神经网络的训练,训练后的神经网络可以处理未知的偏振多光谱图像,并实现了对几种典型地物的分类;最后,对神经网络分类的效果进行了评价,并与其他几种典型分类方法的效果进行了对比,发现神经网络方法具有更好的分类精度和效果,相比典型的最大似然分类算法,其总体分类精度可从91.7%提升至94.2%,Kappa系数可从0.851提升至0.898。研究结果表明:基于神经网络的偏振光谱图像分类方法对于改进和优化现有的偏振多光谱图像数据处理方法具有一定的研究意义。     

Neural network approach to land cover mapping

A pattern classification method is proposed for remote sensing data using neural networks. First, the authors apply the error backpropagation (BP) algorithm to classify the remote sensing data. In this case, the classification performance depends on a training data set. In order to get stable and precise classification results, the training data set is selected based on geographical information and Kohonen's self-organizing feature map. Using the training data set and the error backpropagation algorithm, a layered neural network is trained such that the training patterns are classified with a specified accuracy. After training the neural network, some pixels are deleted from the original training data set if they are incorrectly classified and a new training data set is built up. Once training is complete, a testing data set is classified by using the trained neural network. The classification results of LANDSAT TM data show that this approach produces excellent results which are more realistic and noiseless compared with a conventional Bayesian method     

改进自训练模型在业务质差用户识别中的应用

质差用户识别是降低用户投诉率、提升用户满意度的重要环节。针对当前电信网络系统中业务感知相关的大量结构化及非结构化数据难以有效标注、质差用户标签不完备、现有监督学习模型训练样本不均衡而导致质差识别率低的问题,采用改进自训练半监督学习模型,利用少量满意度低分和投诉用户作为质差用户标签对网络数据进行标注,并通过标签迁移对大量未标注数据进行训练识别质差用户。实验表明,相比于识别准确率高但是训练成本高的全监督学习和识别准确率低的无监督学习,半监督学习可以充分利用无标签样本数据进行有效训练,保证较低训练成本的同时显著提升质差用户识别准确率。     

基于红外与雷达的夜间无人车驾驶决策方法

速度与方向的决策建议是夜间无人车驾驶研究的关键,针对夜间无人车速度与方向决策,基于红外图像与雷达信息,提出了一种包含深度信息的红外图像多任务分类网络用来给出速度与方向决策.通过红外摄像头及雷达采集的数据训练深度网络,其中雷达采集的深度图像作为训练标签,采用卷积-反卷积神经网络来进行红外图像的深度估计,进而获得深度信息.利用深度信息制作分类网络训练标签,通过AlexNet分类网络得到速度决策建议.再根据红外图像的道路信息训练方向分类网络,将无人车的驾驶决策问题转化为分类模型,并将分类模型与深度估计网络相结合.实验结果表明,网络的角度准确率及速度准确率分别为87.43%和85.89%,并且利用训练得到的模型对图像进行决策的时间为0.04 s/帧,能够达到实时性的要求。     

基于半监督联合神经网络的调制识别算法

针对基于有监督学习通信信号分类算法需要大量有标签训练样本,而在实际场合大多无法满足数量要求的问题,提出利用数据驱动模型的半监督学习方法,通过对比预测编码无监督算法预训练和有监督学习相结合,利用LSTM (long short term memory)和ResNet (residual network)联合神经网络实现小样本自动提取特征,提高小样本条件下信号识别准确率。在真实通信调制信号集上实验表明,半监督联合神经网络结构较以往方法,识别准确率提升3%-20%,小样本条件下性能提高60%,同时在低信噪比条件下识别能力突出,0dB时对11种调制信号平均识别正确率达到92%,具有明显优势。      

Distribution learning for radio network planning tool simulation

We propose a novel method that combines the simulation results of a model-based prediction tool with the knowledge contained in measurement data. This mixture of the a priori information and the posteriori knowledge aims at enhancing the prediction results by increasing their precision and quality. A multilayer perceptron (MLP) is trained to learn the mapping between the distributions of the measurement data and the simulation data. To make the complexity of the MLP tractable, we propose the utilization of independent component analysis (ICA). The ICA transformation makes the variables at the input of the MLP statistically independent so that it can perform its learning and generalization on individual one-dimensional distributions. Other contributions consist of the application of the k-means clustering algorithm on the incoming data and the use of the training data world model to enhance the generalization capability of the MLP. The world model consists of the aggregation of all the available data in the learning space. The proposed method is applied to a third generation mobile network to enhance the predictions of uplink and downlink base station loads. After a training performed on a given network configuration, mechanical antenna tilts are modified and we show that the results obtained by the supervised predictions are much closer to measurements than simulation results for cases that have not been encountered before. Copyright © 2008 John Wiley & Sons, Ltd.     

利用稀疏堆栈自编码器实现调制样式识别算法

针对传统调制样式识别方法性能受预先依靠经验设计的特征参数影响大问题,提出一种基于稀疏堆栈自编码器的数字调制样式识别算法。首先根据网络输入数据形式要求,为了利用信号幅度和相位所包含的调制样式信息,提出一种将复数信号预处理为网络可接受的实数形式的信号预处理方法。在网络训练阶段,先通过逐层训练得到每层稀疏自编码网络的初始化参数,再通过有监督算法对分类层训练,最后利用有监督算法进行整体优化。采用 作为分类层完成数字调制样式识别。7种数字调制样式识别的仿真实验表明了本文算法的有效性,相比于其他算法,本文算法在低信噪比时正确识别率较高,识别性能不受人为因素的影响。      

基于BBO-MLP和纹理特征的图像分类算法

为了提高图像分类的准确率,解决多层感知器(MLP )收敛速度缓慢等问题,提出了一种基于生物地理学优化-MLP(BBO-MLP)和纹理特征的 图像分类算法。首先,从图像库中选取 3类不同的图片,对图像分类算法运行环境进行建模;其次,选取角二阶矩(U NI)、熵(CON)、惯性矩(ENT)和 相关性(CDR)4个纹理参数构建一个四维特征矩,根据用户提供的类别号和图像 的纹理特征向量 生成训练样本文件;然后,将提取的数据作为MLP的输入数据,为MLP定义一个评估栖息地的 误差适应度函数并对适应度函数进行全局优化,利用BBO算法训练MLP,得到分类模型;最后 ,利用训练好的MLP对图像进行分类,并引入二次反馈机制进一步提高算法性能。实验结果 表明,与PSO、GA、ACO、ES和PBIL等优化算法相比,本文的BBO-MLP算法具有较高的分类正 确率。     

基于深度学习的乳液泵缺陷检测算法

为了实现对工业产品乳液泵的缺陷检测,本文采集泵顶、泵上端、泵下端、尾管4个角度的样本图像,并基于深度学习中迁移学习和卷积神经网络的原理分别构建各角度的分类模型以检测缺陷样本。首先,使用Mini-ImageNet数据集预训练网络模型。然后,调整模型结构并加载预训练网络的参数,并将乳液泵各角度的训练集和验证集经过图像预处理算法后输入至卷积神经网络中训练,根据训练过程中验证集准确率的变化调整网络超参数,得到最终网络模型。最后,将预处理后的乳液泵测试样本输入至训练好的模型中,检测最终模型的缺陷识别效果。最终4个角度检测准确率均在93%以上,单样本检测用时为2.52s,优于传统方法。本文算法可用于设计乳液泵缺陷检测系统,该系统能与工业结构相结合筛选出有缺陷的泵体,也可拓展到工业其他物件的缺陷检测。     

GSM Mobile Station Location Using Reference Stations and Artificial Neural Networks

In this paper we present a novel approach to the automatic GSM mobile station location. The approach is based on measurement of radio signal strengths from a number of the neighboring base stations (antennas) and estimation of the mobile station position using trained artificial neural network (ANN) models. First, we present an improved version of our previous positioning back propagation (BP) ANN multi-level perceptron (MLP) model that further improves positioning accuracy. Then, we extend the MLP primary ANN model by introducing correctional factors obtained from a number of reference stations with known positions. Two new models with the improved location accuracy, both aimed at real-time application, are presented. The first model is using differential range to improve the estimated location of the mobile station. The second is using small-scale secondary neural networks trained with data obtained from reference stations, in addition to the primary ANN, to correct location accuracy.     

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。     

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。     

多字体字符识别的模糊神经网络模型

提出了一种用于多字体字符识别的模糊神经网络模型．通过对一个３ 层ＭＬＰ的输入层、输出层以及学习算法的模糊化,构造出能有效处理具有模糊边界的模式分类问题的模糊神经网络．经过大量实际采样多字体字符样本的测试表明,该模型能对字体字符识别取得很高的识别率,对加噪字符的识别试验还表明该模型具有较好的鲁棒性．     

Learn++: an incremental learning algorithm for supervised neuralnetworks

We introduce Learn++, an algorithm for incremental training of neural network (NN) pattern classifiers. The proposed algorithm enables supervised NN paradigms, such as the multilayer perceptron (MLP), to accommodate new data, including examples that correspond to previously unseen classes. Furthermore, the algorithm does not require access to previously used data during subsequent incremental learning sessions, yet at the same time, it does not forget previously acquired knowledge. Learn++ utilizes ensemble of classifiers by generating multiple hypotheses using training data sampled according to carefully tailored distributions. The outputs of the resulting classifiers are combined using a weighted majority voting procedure. We present simulation results on several benchmark datasets as well as a real-world classification task. Initial results indicate that the proposed algorithm works rather well in practice. A theoretical upper bound on the error of the classifiers constructed by Learn++ is also provided     

Genetic Sequence Classification and its Application to Cross-Species Homology Detection

Although large-scale classification studies of genetic sequence data are in progress around the world, very few studies compare different classification approaches, e.g. unsupervised and supervised, in terms of objective criteria such as classification accuracy and computational complexity. In this paper, we study such criteria for both unsupervised and supervised classification of a relatively large sequence data set. The unsupervised approach involves use of different sequence alignment algorithms (e.g., Smith-Waterman, FASTA and BLAST) followed by clustering using the Maximin algorithm. The supervised approach uses a suitable numeric encoding (relative frequencies of tuples of nucleotides followed by principal component analysis) which is fed to a Multi-layer Backpropagation Neural Network. Classification experiments conducted on IBM-SP parallel computers show that FASTA with unsupervised Maximin leads to best trade-off between accuracy and speed among all methods, followed by supervised neural networks as the second best approach. Finally, the different classifiers are applied to the problem of cross-species homology detection.     

改进的U-Net算法在遥感图像典型农作物分类研究

针对传统算法提取遥感图像分类特征不全,及识别农作物分类准确率不高的问题,以无人机遥感图像为数据源,提出改进U-Net模型对研究区域薏仁米、玉米等农作物进行分类识别。实验中首先对遥感影像进行预处理,并进行数据集标注与增强;其次通过加深U-Net网络结构、引入SFAM模块和ASPP模块,多级多尺度特征聚合金字塔方法等对算进行法改进,构建改进的U-Net算法,最后进行模型训练与改进修正。实验结果表明:总体分类精度OA达到88.83%,均交并比MIoU达到0.52,较传统U-Net模型、FCN模型和SegNet模,在分类指标和精度上都有明显的提升。     

基于高光谱图像的协同分层波谱识别--以兰州、榆林地区为例

提出协同分层波谱识别法,分别从兰州、榆林市Hyperion高光谱图像上识别9种目标地类,并与SVM监督分类对比。针对Hyperion图像波谱识别的4个难点:光谱信息高保真融合、敏感谱段提取、"椒盐效应"去除、消除"同物异谱"现象导致的误判,协同应用WP-GS融合、导数变换、4尺度面向对象分割和多谱段SAM解决上述难点,并基于Hyperion导数变换图像分析波谱变化特征、提取敏感谱段、从4个尺度层依次识别9种目标地类,然后根据目视评判和定量评价,与综合使用Gram-Schmidt光谱锐化融合/Savitzky-Golay卷积滤波/PCA变换的SVM监督分类结果比较识别精度。实验结果表明WP-GS融合的光谱保真效果优于Gram-Schmidt光谱锐化;4尺度面向对象分割抑制"椒盐效应"的效果优于Savitzky-Golay卷积滤波、移动均值滤波;多谱段SAM利用导数波谱特征能够消除因照度不同对同一类别地物的误判。采用协同分层波谱识别法,兰州市Hyperion图像波谱识别的总体精度、Kappa系数分别为89.52%、0.852,较SVM分类分别提高18.68%和17.52%;榆林市Hyperion图像识别地物的总体精度、Kappa系数分别为91.12%、0.873,较SVM分类分别提高17.80%和16.89%。协同分层波谱识别法应用多种技术一体化解决Hyperion图像应用难点,有效利用导数波谱变化特征提取目标敏感谱段,在复杂环境下识别目标地类的能力优于SVM监督分类。     

卷积神经网络（CNN）训练中卷积核初始化方法研究

本文提出了一种基于样本图像局部模式聚类的卷积核初始化方法,该方法可用于卷积神经网络（Convolutional neural network, CNN）训练中卷积核的初始化。在卷积神经网络中,卷积核的主要作用可看成是利用匹配滤波提取图像中的局部模式,并将其作为后续图像目标识别的特征。为此本文在图像训练集中选取一部分典型的样本图像,在这些图像中抽取与卷积核相同大小的子图作为图像局部模式矢量集合。首先对局部模式子图集合应用拓扑特性进行粗分类,然后对粗分类后的每一子类采用势函数聚类的方法获取样本图像中的典型局部模式子图,构成候选子图模式集,用它们作为CNN的初始卷积核进行训练。实验结果表明,本文方法可以明显加速CNN网络训练初期的收敛速度,同时对最终训练后的网络识别精度也有一定程度的提高。      

Dimension reduction using global and local pattern information-based maximum margin criterion

Dimension reduction is an important research area in pattern recognition when dealing with high- dimensional data. In this paper, a novel supervised dimension reduction approach is introduced for classification. Advantages of using not only global pattern information but also local pattern information are examined in the maximum margin criterion framework. Experimental comparative results in object recognition, handwritten digit recognition, and hyperspectral image classification are presented. According to the experimental results, the proposed method can be a valuable choice for dimension reduction when considering the difficulty of obtaining training samples for some applications.     

Principal component analysis-enhanced cosine radial basis function neural network for robust epilepsy and seizure detection.

A novel principal component analysis (PCA)-enhanced cosine radial basis function neural network classifier is presented. The two-stage classifier is integrated with the mixed-band wavelet-chaos methodology, developed earlier by the authors, for accurate and robust classification of electroencephalogram (EEGs) into healthy, ictal, and interictal EEGs. A nine-parameter mixed-band feature space discovered in previous research for effective EEG representation is used as input to the two-stage classifier. In the first stage, PCA is employed for feature enhancement. The rearrangement of the input space along the principal components of the data improves the classification accuracy of the cosine radial basis function neural network (RBFNN) employed in the second stage significantly. The classification accuracy and robustness of the classifier are validated by extensive parametric and sensitivity analysis. The new wavelet-chaos-neural network methodology yields high EEG classification accuracy (96.6%) and is quite robust to changes in training data with a low standard deviation of 1.4%. For epilepsy diagnosis, when only normal and interictal EEGs are considered, the classification accuracy of the proposed model is 99.3%. This statistic is especially remarkable because even the most highly trained neurologists do not appear to be able to detect interictal EEGs more than 80% of the times.