Approximating support vector machine with artificial neural network for fast prediction

Support vector machine (SVM) is a powerful algorithm for classification and regression problems and is widely applied to real-world applications. However, its high computational load in the test phase makes it difficult to use in practice. In this paper, we propose hybrid neural network (HNN), a method to accelerate an SVM in the test phase by approximating the SVM. The proposed method approximates the SVM using an artificial neural network (ANN). The resulting regression function of the ANN replaces the decision function or the regression function of the SVM. Since the prediction of the ANN requires significantly less computation than that of the SVM, the proposed method yields faster test speed. The proposed method is evaluated by experiments on real-world benchmark datasets. Experimental results show that the proposed method successfully accelerates SVM in the test phase with little or no prediction loss.     

支持向量机与神经网络相结合的板带凸度预测

为提高热轧生产过程中板带凸度的预测精度,提出了一种将粒子群优化算法(particle swarm optimization, PSO)、支持向量回归(support vector regression, SVR)和BP神经网络(back propagation neural network, BPNN)相结合的板带凸度预测模型。采用PSO算法优化SVR模型的参数,建立了PSO-SVR板带凸度预测模型,提出采用BPNN建立板带凸度偏差模型与PSO-SVR板带凸度模型相结合的方法对板带凸度进行预测。采用现场数据对模型的预测精度进行验证,并采用统计指标评价模型的综合性能。仿真结果表明,与PSO-SVR、SVR、BPNN和GA-SVR模型进行比较,PSO-SVR+BPNN模型具有较高的学习能力和泛化能力,并且比GA-SVR模型运算时间短。     

基于偏好型支持向量机的遥感图像云检测技术

针对以往遥感图像云检测方法对雪地等特殊地貌识别效果不理想的问题,基于偏好型支持向量机(preference support vector machine,PSVM)提出一种云检测技术。利用图像的灰度特征和纹理特征,提取图像的能量、对比度、逆差矩、熵、自相关性以及平均灰度值6个分类指标,使用偏好训练的方式,提高对云和普通地貌的识别准确率,准确区分云和雪地等特殊地物。实验结果表明,PSVM方法综合识别准确率达到了97.66%,特殊地貌识别准确率达到了99.31%,相比于传统的云检测算法性能提升显著。     

基于支持向量机的遥感影像薄云去除方法

提出了一种基于支持向量机的遥感影像薄云去除方法。利用支持向量机对边缘点的极强捕捉能力,将含云遥感影像进行多尺度分解,获得不同尺度上的变换系数,再结合方向滤波器组得到丰富的高频信息。采用自适应阈值的图像增强方法处理这片区域,重构时通过对高频区域的增强和低频区域的抑制,得到去云图像。实验结果表明,采用该方法能有效地去除遥感影像中的薄云。     

基于K型支持向量机的遥感图像分类新算法

为了提高遥感图像的分类精度和识别速度,提出了一种基于K型支持向量机（SVM）的遥感图像分类新算法,该算法将灰度共生矩阵提取的纹理特征与光谱特征相结合进行分类。对两组Landsat ETM+数据进行分类仿真实验,结果表明,在多光谱遥感图像的分类中,新算法提高了分类效率、分类精度和泛化能力,K型SVM是一种优于径向基函数SVM的分类器。     

基于支持向量机的遥感影像湿地信息提取研究*

以ETM 影像数据为例,采用基于支持向量机的方法对黄河中上游区域湿地信息进行提取,并将该方法与传统的最大似然分类提取方法以及面向对象的提取方法进行对比分析。结果表明:基于支持向量机方法的提取精度高达93.57%,Kappa系数也超过了0.9,比单纯的最大似然分类方法或者面向对象的方法提取精度高得多,而且该方法操作性和实用性也很强。     

基于全卷积神经网络的多源高分辨率遥感道路提取

针对半自动道路提取方法人工参与较多、提取精度不高且较为耗时的问题提出一种基于全卷积神经网络（FCN）的多源高分辨率遥感道路提取方法。首先,对高分二号和World View图像进行分割,用卷积神经网络（CNN）分类出包含道路的图像;然后,用Canny算子提取道路的边缘特征信息;最后,结合RGB、Gray和标签图放入FCN中训练,将现有的FCN模型拓展为多卫星源输入及多特征源输入的FCN模型。选取西藏日喀则地区作为研究区域,实验结果显示,所提方法在对高分辨率遥感影像进行道路提取时能够达到99.2%的提取精度,并且有效地减少了提取所需的时间。     

基于多源遥感数据融合的土壤水分反演研究

土壤含水量作为地表的重要参量之一,对地球能量循环、水循环、碳循环及生态环境都有十分重要的意义。以南京市金川河流域为研究区,融合哨兵 2 号 L2A 数据和 Landsat 8 遥感数据 2 种数据源,分别采用偏最小二乘法(PLSR)、最小二乘-支持向量机(LS-SVM)、反向传播神经网络(BPNN)和随机森林(RF)4 种建模方法,建立遥感数据与土壤含水量之间的关系,并进行模型的验证与评价。结果表明：1)土壤含水量与哨兵 2 号和 Landsat 8 各波段反射率均呈负相关关系,和海岸带监测波段(波长为 430~450 nm)和近红外波段(波长为 2 100~2 300 nm)相关性最佳;2)融合后的遥感数据相较于单一遥感数据源,预测土壤含水量的能力更佳, 最优模型 R2 达 0.996,均方根误差仅为 0.003 g/g;3)4 种建模方法中,建模效果从好到差依次为 PLSR,RF, LS-SVM,BPNN。融合哨兵 2 号 L2A 和 Landsat 8 数据,结合 PLSR 建模方法可进行土壤含水量的精准反演, 相较于现有研究反演精度大大提升,对研究该地区地表与地下水循环和生态环境治理有一定参考价值。     

Fast-forward solver for inhomogeneous media using machine learning methods: artificial neural network,support vector machine and fuzzy logic

Encountering with a nonlinear second-order differential equation including ϵ _r and μ _r spatial distributions, while computing the fields inside inhomogeneous media, persuaded us to find their known distributions that give exact solutions. Similarities between random distributions of electric properties and known functions lead us to estimate them using three mathematical tools of artificial neural networks (ANNs), support vector machines (SVMs) and Fuzzy Logic (FL). Assigning known functions after fitting with minimum error to arbitrary inputs using results of machine learning networks leads to achieve an approximate solution for the field inside materials considering boundary conditions. A comparative study between the methods according to the complexity of the structures as well as the accuracy and the calculation time for testing of unforeseen inputs, including classification, prediction and regression is presented. We examined the extracted pairs of ϵ _r and μ _r with ANN, SVM networks and FL and got satisfactory outputs with detailed results. The application of the presented method in zero reflection subjects is exemplified.

     

Comparison of support vector machine,artificial neural network,and spectral angle mapper algorithms for crop classification using LISS IV data

The Resourcesat-2 is a highly suitable satellite for crop classification studies with its improved features and capabilities. Data from one of its sensors, the linear imaging and self-scanning (LISS IV), which has a spatial resolution of 5.8 m, was used to compare the relative accuracies achieved by support vector machine (SVM), artificial neural network (ANN), and spectral angle mapper (SAM) algorithms for the classification of various crops and non-crop covering a part of Varanasi district, Uttar Pradesh, India. The separability analysis was performed using a transformed divergence (TD) method between categories to assess the quality of training samples. The outcome of the present study indicates better performance of SVM and ANN algorithms in comparison to SAM for the classification using LISS IV sensor data. The overall accuracies obtained by SVM and ANN were 93.45% and 92.32%, respectively, whereas the lower accuracy of 74.99% was achieved using the SAM algorithm through error matrix analysis. Results derived from SVM, ANN, and SAM classification algorithms were validated with the ground truth information acquired by the field visit on the same day of satellite data acquisition.     

Soft multimedia anomaly detection based on neural network and optimization driven support vector machine

Software multimedia anomaly detection model based on neural network and optimization driven support vector machine is discussed in this paper. For multimedia information, most traditional information security technology has its limitations. For example, the limitation of the encryption technology is that on the one hand, the encrypted files resulting from the incomprehension of attributes interfere with the transfer of multimedia information. On the other hand, the encrypted multimedia information is likely to attract the attacker’s curiosity and attention, and is likely to be cracked, and once it is cracked, the system loses control of the information. To deal with these challenges, this study integrates soft computing techniques to finalize the enhanced multimedia anomaly detection model. With respect to the neural network, a random system with random factors is referred to as a random system. These practical systems are generally described and modeled by stochastic differential equations. In this study, we combined the double support vector machine and decision tree support vector machine to construct a new double support vector machine decision tree classifier. Kernel function and convex optimization were integrated to guarantee an optimal solution. Experimental results demonstrated the robustness of the model compared with other recent techniques.

     

Retina disease prediction using modified convolutional neural network based on Inception-ResNet model with support vector machine classifier

Artificial intelligence and deep learning have aided ocular disease through experiments including automatic illness recognition from images of the iris, fundus, or retina. Automated diagnosis systems (ADSs) provide services for the benefit of humanity and are essential in the early detection of harmful diseases. In fact, early detection is essential to avoid total blindness. In real life, several diagnostic tests such as visual ocular tonometry, retinal exam, and acuity test are performed, but they are conclusively time demanding and stressful for the patient. To consume time and detect the retinal disease earlier, an efficient prediction method is designed. In this proposed model, the first process is data collection that consists of a retinal disease dataset for testing and training. The second process is pre-processing, which executes image resizing and noise filter for feature extraction. The third step is feature extraction, which extracts the image's form, size, color, and texture for classification with CNN based on Inception-ResNet V2. The classification process is done by using the SVM with the extracted features. The prediction of diseases is classified such as normal, cataract, glaucoma, and retinal disease. The suggested model's performance is assessed using performance indicators such as accuracy, error, sensitivity, precision, and so forth. The suggested model's accuracy, error, sensitivity, and precision are 0.96, 0.962, 0.964, and 0.04, respectively, higher than existing techniques such as VGG16, Mobilenet V1, ResNet, and AlexNet. Thus, the proposed model instantly predicts retinal disease.     

基于支持向量数据描述算法的SVM多分类新方法*

提出一种基于支持向量数据描述算法(SVDD)的多分类方法(S-MSVM).受SVDD的启发,该方法对每类样本建立一个超球来界定,但训练好的超球在所有情况下都是相交的.选择相交区域的样本单独建立超球,重复该步骤,直到相交区域消失或相交区域内没有样本点.给出了该方法的时间复杂度分析,并通过实验验证了该方法具有相对较好的训练精度.     

Mapping urban land surface temperature using remote sensing techniques and artificial neural network modelling

The objective of the present study is to monitor and predict the changes in land surface temperature (LST) in the North of Jordan during the Period 2000 to 2016. Due to political instability in the nearby countries Syria and Iraq, Jordan has witnessed increased influxes of refugees, starting from the year 2003, which has been led to the urban expansion in the area that reflected on the climatic conditions and affected the LST values. Satellite images were used for providing LST, the acquired images represented two seasons of each year, namely summer and winter. Simulation and prediction of LST values for the next 10 years were carried out using nonlinear autoregressive exogenous (NARX) artificial neural network (ANN) model. The inputs to the model consist of meteorological data collected from eight stations in the study area, population, and land use and land cover (LULC). In fact, LULC was expressed in terms of normalized difference building index (NDBI) and normalized difference vegetation index (NDVI) that were obtained from satellite images. The model showed a high correlation between these parameters and the values of simulated LST, where the correlation coefficient for the training set, validation set, testing set and for the entire data ranged from 0.91 to 0.92. Based on the predicted LST values, LST maps for the next 10 years were developed and compared with the present actual LST maps for the year 2016. The comparison has shown an average increase of 1.1 ^°C in the average LST values, which is considered a significant increase compared with previous studies.     

基于人工鱼群算法的孪生支持向量机

孪生支持向量机（twin support vector machine, TWSVM）是在支持向量机的基础上产生的机器学习算法,具有训练速度快、分类性能优越等优点。但是孪生支持向量机无法很好地处理参数选择问题,不合适的参数会降低分类能力。人工鱼群算法（artificial fish swarm algorithm, AFSA）是一种群智能优化算法,具有较强的全局寻优能力和并行处理能力。本文将孪生支持向量机与人工鱼群算法结合,来解决孪生支持向量机的参数选择问题。首先将孪生支持向量机的参数作为人工鱼的位置信息,同时将分类准确率作为目标函数,然后通过人工鱼的觅食、聚群、追尾和随机行为来更新位置和最优解,最后迭代结束时得到最优参数和最优分类准确率。该算法在训练过程中自动确定孪生支持向量机的参数,避免了参数选择的盲目性,提高了孪生支持向量机的分类性能。     

基于支持向量机的恶意软件行为评估系统

为解决恶意软件行为分析系统中分类准确率较低的问题,提出了一种基于支持向量机(SVM)的恶意软件分类方法。首先人工建立了一个以软件行为结果作为特征的危险行为库;然后捕获软件所有行为,并与危险行为库进行匹配,通过样本转换算法将匹配结果变成适合SVM处理的数据,再利用SVM进行分类。在SVM模型、核函数以及参数对(C,g)的选择方面先进行理论分析确定大致范围,再使用网格搜索和遗传算法(GA)相结合的方式进行寻优。为验证所提恶意软件分类方法的有效性,设计了一个基于SVM模型的恶意软件行为评估系统。实验结果表明,该系统的误报率和漏报率分别为5.52%和3.04%,比K近邻(KNN)、朴素贝叶斯(NB)算法更好,与反向传播(BP)神经网络相当,但比BP神经网络的训练和分类效率更高。     

基于SVM动态集成的高光谱遥感图像分类

在Bagging支持向量机(SVM)的基础上,将动态分类器集选择技术用于SVM的集成学习,研究了SVM动态集成在高光谱遥感图像分类中的应用。结合高光谱数据特性,通过随机选取特征子空间和反馈学习改进了Bagging SVM方法;通过引进加性复合距离改善了K近邻局部空间的计算方法;通过将错分的训练样本添加到验证集增强了验证集样本的代表性。实验结果表明,与单个优化的SVM和其他常见的SVM集成方法相比,改进后的SVM动态集成分类精度最高,能有效地提高高光谱遥感图像的分类精度。     

基于自编码神经网络的航空物探遥感数据分类方法研究

航空物探遥感数据的采集过程中受到电磁波辐射等外界因素的影响,导致航空物探遥感数据分类准确率较低,为此提出基于自编码神经网络的航空物探遥感数据分类方法;根据航空物探对象的基本特征,设置遥感数据的分类标准;通过辐射校正、几何纠正、噪声消除等步骤,完成航空物探遥感数据的预处理;构建自编码神经网络,利用自编码神经网络算法,从光谱、形状、纹理等方面提取遥感数据特征,通过特征匹配确定航空物探遥感数据的所属类型;通过分类性能测试实验得出结论：所提方法的全局遥感数据分类成功率和错误率的平均值分别为99.8%和0.6%,局部遥感数据分类的成功率和错误率的平均值分别为99.8%和0.3%,即所提方法在分类性能方面具有明显优势。     

基于改进卷积神经网络与支持向量机结合的面部表情识别算法

针对当前卷积神经网络（CNN）利用端层特征进行面部表情识别存在模型结构繁琐、训练参数过多、识别不够理想的问题,提出一种基于改进CNN与支持向量机（SVM）相结合的优化算法。首先,利用连续卷积的思想设计网络模型,以获取更多非线性激活;然后,采用自适应全局平均池化（GAP）层取代传统CNN中的全连接层,以减少网络参数量;最后,用SVM分类器代替传统Softmax函数实现表情识别,以提高模型泛化能力。实验结果表明,所提算法在Fer2013和CK+数据集上分别取得了73.4%和98.06%的识别准确率,与传统LeNet-5算法相比,在Fer2013数据集上提升了2.2个百分点,且该网络模型结构简单、参数量较少,具有良好的鲁棒性。