Hepatic CT image query based on threshold-based classification scheme with gabor features

Hepatic computed tomography(CT) images with Gabor function were analyzed.Then a threshold-based classification scheme was proposed using Gabor features and proceeded with the retrieval of the hepatic CT images.In our experiments, a batch of hepatic CT images containing several types of CT findings was used and compared with the Zhao's image classification scheme, support vector machines(SVM) scheme and threshold-based scheme.     

基于形貌特征的CO2腐蚀速率预测方法

针对CO2腐蚀过程复杂，难以利用实测数据有效预测腐蚀速率问题，文中以腐蚀形貌图像为对象，利用支持向量机（SVM）构建预测模型，实现对CO2腐蚀速率的预测。对N80钢的CO2腐蚀图像进行灰度处理、灰度增强及二值化处理，提取蚀孔数目和孔蚀面积。经计算获得孔蚀密度及孔蚀率，结合工作温度及CO2分压作为腐蚀速率预测的四维特征向量。以SVM构建预测模型，经测试，可准确预测CO2腐蚀速率，并与神经网络预测结果进行对比，验证了该方法的优越性。     

Identifying multiclass vehicles using global positioning system data

It has been previously evidenced that global positioning system (GPS) data can be used to distinguish passenger cars from delivery trucks. In this paper, a machine learning approach is proposed to use GPS data to identify multiclass vehicles, including passenger cars, single unit trucks, and multi-trailer trucks. The method is acceleration and deceleration-based since it considers the variations of acceleration and deceleration as the most effective features to classify vehicles. The overall classification result for the three vehicle classes is about 75%. The major challenge is to distinguish single unit trucks from multitrailer trucks due to their somewhat similar mobility patterns. The paper also explores the impacts of GPS sampling frequency on vehicle classification. It is found that the proposed multiclass vehicle classification can be reasonably conducted if the data are collected frequently enough (i.e., every five seconds or more frequently) to capture the major acceleration and deceleration processes. The proposed method can be considered as a low-cost and non-intrusive approach to collect vehicle class information and to potentially supplement the existing classification schemes in urban areas.     

Rollover risk prediction of heavy vehicles by reliability index and empirical modelling

This paper focuses on a combination of a reliability-based approach and an empirical modelling approach for rollover risk assessment of heavy vehicles. A reliability-based warning system is developed to alert the driver to a potential rollover before entering into a bend. The idea behind the proposed methodology is to estimate the rollover risk by the probability that the vehicle load transfer ratio (LTR) exceeds a critical threshold. Accordingly, a so-called reliability index may be used as a measure to assess the vehicle safe functioning. In the reliability method, computing the maximum of LTR requires to predict the vehicle dynamics over the bend which can be in some cases an intractable problem or time-consuming. With the aim of improving the reliability computation time, an empirical model is developed to substitute the vehicle dynamics and rollover models. This is done by using the SVM (Support Vector Machines) algorithm. The preliminary obtained results demonstrate the effectiveness of the proposed approach.     

基于混沌时序最小二乘支持向量机的汽油机瞬态空燃比预测模型研究

针对由氧传感器构成的瞬态空燃比反馈控制系统无法满足实时性要求的问题,提出了基于混沌时序最小二乘支持向量机(LS-SVM)的瞬态空燃比预测模型。对试验采集到的一维空燃比数据利用相空间重构技术构造多维空间数据,恢复空燃比时间序列的多维非线性特性,然后采用LS-SVM对重构后的数据进行训练及预测,得出预测结果。仿真结果表明:与Elman神经网络预测模型及前馈BP神经网络预测模型相比较,混沌时序LS-SVM预测模型具有更强的非线性预测能力,能够有效地提高瞬态空燃比的预测精度。     

车用汽油机瞬态空燃比的混沌时序非线性组合辨识模型

为提高汽油机瞬态空燃比的辨识精度,提出了混沌时序非线性组合辨识模型.采用2种单项辨识方法,包括最小二乘支持向量机(LS-SVM)及径向基函数(RBF)前向型神经网络,分别对瞬态空燃比时间序列进行建模与辨识.采用非线性组合方法利用BP神经网络对2种单项辨识方法的结果进行组合辨识,并与Elman神经网络模型及最小二乘辨识模型进行比较.结果表明:混沌时序非线性组合辨识模型的辨识精度优于Elman神经网络模型及最小二乘辨识模型,具有更强的非线性辨识能力,能提高瞬态空燃比的辨识精度,为空燃比反馈控制的成功实行提供了有力依据.     

基于SVM分类器的道路湿滑图像分类方法研究

在恶劣天气下路面湿滑状况将发生显著变化,导致驾驶条件恶化,极易引发恶劣的交通事故.为了帮助交通管理部门实时获取路面湿滑状况信息,了解路面附着性能,并以此制定合理的车速控制和交通诱导策略,通过采集道路湿滑图像,利用设计的具体SVM分类器结合3种训练算法对道路湿滑图像进行分类学习训练,并通过训练后的SVM分类器对大量道路湿滑图像进行分类实验,对道路湿滑状态进行分类研究.实验结果表明:(1)寻找合适的核函数,选择合适的参数是使用SVM进行高效分类的一个重要因素;(2)在训练过程中,均方误差(MSE)能反映出分类器实现的正确率,SVM的训练本身的误差决定了分类的正确率,而且训练个数的增多带来了特征空间维数的增加,从而导致计算量的增大.     

一种基于贝叶斯特征的视频图像人脸检测

提出了一种基于仿射运动模型和贝叶斯理论的视频图像人脸检测方法.建立仿射运动模型进行运动估计,提取运动对象区域;对训练图像提取人脸与非人脸的统计特征,利用贝叶斯准则建立概率模型;根据贝叶斯分类器和支持向量机分类器将图像特征分为人脸类与非人脸类,从而检测出视频运动图像中的人脸区域.     

����PSO-SVM�ľ�����з�ʽԤ��ģ�ͪ�

居民出行方式选择是一个较为复杂的非线性问题,受到的影响因素众多。提出采用支持向量机方法构建了居民出行方式选择模型,并以交叉验证意义下的分类准确率作为适应度函数,利用粒子群算法对支持向量机参数优化选择,避免参数设定的随机性,减少参数选择的工作量.通过实证研究表明,利用粒子群算法优化支持向量机的参数是可行的,支持向量机方法相对于BP神经网络,对居民出行方式预测有更高的精度.预测精度比BP神经网络提高了将近5个百分点,建模样本和测试样本的分类精度分别达到86.20%和82.31%.所构建的模型可用于居民出行方式预测,这对城市交通规划,出行需求预测具有现实指导意义.     

基于DAGSVM的装备不常用备件分类方法

针对装备不常用备件需求样本数据有限、影响不常用备件需求的因素众多,且各因素间的关系多为非线性等特点,分析了现有备件分类方法中的不足,并引入在小样本学习方面具有优势的支持向量机(Support Vector Machines,SVM)方法,利用有向无环图支持向量机(Directed Acyclic Graph SVM,DAGSVM)方法对装备的不常用备件进行分类,并进行了实例应用。结果表明,该方法可以有效地解决装备不常用备件的多类分类问题。