用于支持向量机拒识区域的加权k近邻法

为解决1-v-r和1-v-1支持向量机中存在的拒识区域问题,提出一种加权k近邻法。该方法计算落入拒识区域中的样本,即拒识样本到所有训练样本的距离,选择最近的k个样本为拒识样本的类别投票,并根据距离大小进行加权,得票多的类即拒识样本的所属类。实验结果表明,加权k近邻法实现了零拒识,提高了传统多分类支持向量机的分类性能。     

嵌入非对称拒识代价的二元分类算法

针对传统分类算法隐含的假设(相信并且接受每个样本的分类结果)在医疗/故障诊断和欺诈/入侵检测等领域中并不适用的问题,提出嵌入非对称拒识代价的二元分类问题,并对其进行简化.在此基础上设计出基于支持向量机(SVM)的代价敏感分类算法(CSVM-CRC).该算法包括训练 SVM 分类器、计算后验概率、估计分类可靠性和确定最优拒识阈值4个步骤.基于10个 Benchmark 数据集的实验研究表明, CSVM-CRC 算法能够有效降低平均代价.     

马氏距离法在支持向量机拒识区域中的应用

为克服传统多分类支持向量机中存在的拒识区域,提出一种基于马氏距离的拒识区域解决方案.该方法首先计算落入拒识区域中的样本点到每类样本集的马氏距离,然后选择较小的马氏距离对应的类为样本的所属类.标准数据实验结果表明,马氏距离法在实验数据上消除拒识区域,有效提高了算法的分类性能和泛化能力.     

基于欧氏距离的支持向量机拒识区域解决方案

为克服传统多分类支持向量机中存在的拒识区域问题,提高算法的分类性能和泛化能力,提出一种基于欧氏距离的拒识区域解决方案。该方法直接计算落入拒识区域中的样本点到每类中心的欧氏距离,然后选择较小的欧氏距离对应的类为样本的所属类。基于标准数据集的实验结果表明,欧氏距离法实现了零拒识,有效提高了算法的分类性能和泛化能力。     

文档识别中误切分字符拒识问题的研究

自动文档识别中字切分算法如果仅仅依靠大小位置等度量信息,很容易产生误切分图像块,需要字符分类器给出一定的反馈才能准确切分,为此提出了一个新的拒识算法,目标是尽可能准确地拒识非法字符。该文分析了基于距离的分类器的置信度和广义置信度,在此基础上改进了常用的广义置信度映射函数,并设计了一个基于样本学习的拒识规则,提高了拒识算法的适应性。在中日韩三种文档样本上的实验表明,该文算法明显改善了系统性能,对于较低质量的印刷文本识别具有一定的普遍意义。     

前向FP神经网络实现控制系统的拒识模式

在现代控制系统中,处理误识与拒识的关系至关重要,因为在现代化控制系统中前者造成的危害将比后者大得多.通过将新型FP(ForwardPropagation)神经网络应用于实现控制系统的拒识模式,该FP神经网络通过对由控制系统控制指令码和状态信息码构成样本集的学习,不仅可以实现系统中每个控制指令码和状态信息码的吸引半径按需设定,且每个信息码具有最大吸引半径,控制系统自动具有最优的纠错和容错能力,对于系统中控制向量吸引半径之外的信号给以100%的拒识.从而提高系统信号识别的正确率,保证控制系统安全可靠的正常运行.     

基于置信度的非特定人语音识别拒识算法的研究

语音拒识技术是实现一个实用语音识别系统的关键。提出了一种新颖的基于置信度的非特定人语音识别拒识算法，该算法同时考虑了备选假设模型和多候选的信息，适用于拒识不正确的识别结果和词表外(OOV)语音。在一个非特定人英语命令词识别系统中做了一些相关的实验来评估这个算法的性能。实验结果表明，该算法可以有效地去除识别不可靠的语音，提高语音识别的整体性能。     

低分辨雷达目标分类的最小代价拒判算法

为解决低分辨雷达目标自动识别中,干扰目标、虚假目标的存在以及不同类别目标样本集混叠的问题,提出了一种基于最小代价的拒判K近邻识别算法.该算法根据雷达识别系统最小代价的原则,利用Fisher判别函数,确定拒判门限.设计了基于两类拒判域的K近邻识别算法,第一类拒判根据训练样本集特征值的波动范围,对干扰目标和虚假目标进行拒判;第二类拒判根据测试样本与最近邻、次近邻的距离差,实现混叠区域的目标样本拒判.算法先对测试样本进行拒判分析,再利用K近邻算法识别分类.实验结果表明,基于以上算法的低分辨雷达目标识别系统具有较好的鲁棒性和识别性能.     

嵌入拒识的极限学习机在基因表达数据分类中的应用

机器学习技术已被广泛地应用在计算机辅助诊断中,以辅助专家进行医学诊断,但是几乎所有的分类器,都是默认的接收计算机的分类结果,而这种默认的结果在很多情况下会引起较大的误差。如根据诊断者的临床数据判断此人为病人的概率为50%,为健康人的概率也为50%,这时无论计算机将此人分类为健康人还是病人都会面临50%的错误概率。因此,提出嵌入拒识的极限学习机,不仅充分利用了极限学习机快速的学习能力、良好的泛化性能,而且通过嵌入拒识选项,"拒识"可靠性较低的样本来克服分类正确率较低的问题,使得自动分类过程更加可靠。实验结果表明:所提方法在降低误分类率方面是非常有效的。     

基于主动学习的模式类别挖掘模型

在实际应用问题中,由于客观世界物质的多样性、模糊性和复杂性,经常会遇到大量未知样本类别信息的数据挖掘问题,而传统方法往往都依赖于已知样本类别信息才能对数据进行有效挖掘,对于未知模式类别信息的多类数据目前还没有有效的处理方法.针对未知类别信息的多类样本挖掘问题,提出了一种基于主动学习的模式类别挖掘模型(pattern class mining model based on active learning, PM_AL)来解决未知类别信息的模式类别挖掘问题.该模型通过衡量已得到的模式类别与未标记样本间的关系,引入样本差异度的方法来抽取最有价值样本,通过主动学习方式以较小的标记代价快速挖掘无标记样本所蕴含的可能模式类别,从而有助于将无类别标记的多分类问题转化成有类别标记的多分类问题.实验结果表明,PM_AL算法能够以较小的标记代价处理无类别信息的模式类别挖掘问题.     

Symmetric deformable image registration via optimization of information theoretic measures

The use of information theoretic measures (ITMs) has been steadily growing in image processing, bioinformatics, and pattern classification. Although the ITMs have been extensively used in rigid and affine registration of multi-modal images, their computation and accuracy are critical issues in deformable image registration. Three important aspects of using ITMs in multi-modal deformable image registration are considered in this paper: computation, inverse consistency, and accuracy; a symmetric formulation of the deformable image registration problem through the computation of derivatives and resampling on both source and target images, and sufficient criteria for inverse consistency are presented for the purpose of achieving more accurate registration. The techniques of estimating ITMs are examined and analytical derivatives are derived for carrying out the optimization in a computationally efficient manner. ITMs based on Shannon’s and Renyi’s definitions are considered and compared. The obtained evaluation results via registration functions, and controlled deformable registration of multi-modal digital brain phantom and in vivo magnetic resonance brain images show the improved accuracy and efficiency of the developed formulation. The results also indicate that despite the recent favorable studies towards the use of ITMs based on Renyi’s definitions, these measures are seen not to provide improvements in this type of deformable registration as compared to ITMs based on Shannon’s definitions.     

Approaches for the production and evaluation of fuzzy land cover classifications from remotely-sensed data

Remote sensing is an attractive source of data for land cover mapping applications. Mapping is generally achieved through the application of a conventional statistical classification, which allocates each image pixel to a land cover class. Such approaches are inappropriate for mixed pixels, which contain two or more land cover classes, and a fuzzy classification approach is required. When pixels may have multiple and partial class membership measures of the strength of class membership may be output and, if strongly related to the land cover composition, mapped to represent such fuzzy land cover. This type of representation can be derived by softening the output of a conventional ‘hard’ classification or using a fuzzy classification. The accuracy of the representation provided by a fuzzy classification is, however, difficult to evaluate. Conventional measures of classification accuracy cannot be used as they are appropriate only for ‘hard’ classifications. The accuracy of a classification may, however, be indicated by the way in which the strength of class membership is partitioned between the classes and how closely this represents the partitioning of class membership on the ground. In this paper two measures of the closeness of the land cover representation derived from a classification to that on the ground were used to evaluate a set of fuzzy classifications. The latter were based on measures of the strength of class membership output from classifications by a discriminant analysis, artificial neural network and fuzzy c-means classifiers. The results show the importance of recognising and accommodating for the fuzziness of the land cover on the ground. The accuracy assessment methods used were applicable to pure and mixed pixels and enabled the identification of the most accurate land cover representation derived. The results showed that the fuzzy representations were more accurate than the ‘hard’ classifications. Moreover, the outputs derived from the artificial neural network and the fuzzy c-means algorithm in particular were strongly related to the land cover on the ground and provided the most accurate land cover representations. The ability to appropriately represent fuzzy land cover and evaluate the accuracy of the representation should facilitate the use of remote sensing as a source of land cover data.     

Decision-making method using a visual approach for cluster analysis problems; indicative classification algorithms and grouping scope

Abstract: Currently, classifying samples into a fixed number of clusters (i.e. supervised cluster analysis) as well as unsupervised cluster analysis are limited in their ability to support 'cross-algorithms' analysis. It is well known that each cluster analysis algorithm yields different results (i.e. a different classification); even running the same algorithm with two different similarity measures commonly yields different results. Researchers usually choose the preferred algorithm and similarity measure according to analysis objectives and data set features, but they have neither a formal method nor tool that supports comparisons and evaluations of the different classifications that result from the diverse algorithms. Current research development and prototype decisions support a methodology based upon formal quantitative measures and a visual approach, enabling presentation, comparison and evaluation of multiple classification suggestions resulting from diverse algorithms. This methodology and tool were used in two basic scenarios: (I) a classification problem in which a 'true result' is known, using the Fisher iris data set; (II) a classification problem in which there is no 'true result' to compare with. In this case, we used a small data set from a user profile study (a study that tries to relate users to a set of stereotypes based on sociological aspects and interests). In each scenario, ten diverse algorithms were executed. The suggested methodology and decision support system produced a cross-algorithms presentation; all ten resultant classifications are presented together in a 'Tetris-like' format. Each column represents a specific classification algorithm, each line represents a specific sample, and formal quantitative measures analyse the 'Tetris blocks', arranging them according to their best structures, i.e. best classification.     

A theoretical and experimental investigation of graph theoretical measures for land development in satellite imagery

Today's commercial satellite images enable experts to classify region types in great detail. In previous work, we considered discriminating rural and urban regions. However, a more detailed classification is required for many purposes. These fine classifications assist government agencies in many ways including urban planning, transportation management, and rescue operations. In a step toward the automation of the fine classification process, this paper explores graph theoretical measures over grayscale images. The graphs are constructed by assigning photometric straight-line segments to vertices, while graph edges encode their spatial relationships. We then introduce a set of measures based on various properties of the graph. These measures are nearly monotonic (positively correlated) with increasing structure (organization) in the image. Thus, increased cultural activity and land development are indicated by increases in these measures - without explicit extraction of road networks, buildings, residences, etc. These latter, time consuming (and still only partially automated) tasks can be restricted only to "promising" image regions, according to our measures. In some applications our measures may suffice. We present a theoretical basis for the measures followed by extensive experimental results in which the measures are first compared to manual evaluations of land development. We then present and test a method to focus on, and (pre)extract, suburban-style residential areas. These are of particular importance in many applications, and are especially difficult to extract. In this work, we consider commercial IKONOS data. These images are orthorectified to provide a fixed resolution of 1 meter per pixel on the ground. They are, therefore, metric in the sense that ground distance is fixed in scale to pixel distance. Our data set is large and diverse, including sea and coastline, rural, forest, residential, industrial, and urban areas.     

Estimating the effect of crop classification error on evapotranspiration derived from remote sensing in the lower Colorado River basin, USA

In the U.S. Bureau of Reclamation's Lower Colorado River Accounting System (LCRAS), crop classifications derived from remote sensing are used to calculate regional estimates of crop evapotranspiration for water monitoring and management activities on the lower Colorado River basin. The LCRAS accuracy assessment was designed to quantify the impact of crop classification error on annual total crop evapotranspiration (ETc), as calculated from the Penman-Monteith method using the map crop classification as input. The accuracy assessment data were also used to generate a sample-based estimate of total ETc using the crop type identified by direct ground observation of each sample field. A stratified random sampling design was implemented using field size as the stratification variable. The stratified design did not markedly improve precision for the accuracy assessment objective, but it was highly effective for the objective of estimating ETc derived from the ground-observed crop types. The sampling design and analysis methodology developed for LCRAS demonstrates the utility of a multi-purpose approach that satisfies the accuracy assessment objectives, but also allows for rigorous, sample-based estimates of other collective properties of a region (e.g., total ETc in this study). We discuss key elements of this multi-purpose sampling strategy and the planning process used to implement such a strategy.     

Riparian trees genera identification based on leaf-on/leaf-off airborne laser scanner data and machine learning classifiers in northern France

ABSTRACT

Riparian forests are valuable environments delivering multiples ecological services. Because they face both natural and anthropogenic constraints, riparian forests need to be accurately mapped in terms of genera/species diversity. Previous studies have shown that the Airborne Laser Scanner (ALS) data have the potential to classify trees in different contexts. However, an assessment of important features and classification results for broadleaved deciduous riparian forests mapping using ALS remains to be achieved. The objective of this study was to estimate which features derived from ALS data were important for describing trees genera from a riparian deciduous forest, and provide results of classifications using two Machine Learning algorithms. The procedure was applied to 191 trees distributed in eight genera located along the Sélune river in Normandy, northern France. ALS data from two surveys, in the summer and winter, were used. From these data, trees crowns were extracted and global morphology and internal structure features were computed from the 3D points clouds. Five datasets were established, containing for each one an increasing number of genera. This was implemented in order to assess the level of discrimination between trees genera. The most discriminant features were selected using a stepwise Quadratic Discriminant Analysis (sQDA) and Random Forest, allowing the number of features to be reduced from 144 to 3–9, depending on the datasets. The sQDA-selected features highlighted the fact that, with an increasing number of genera in the datasets, internal structure became more discriminant. The selected features were used as variables for classification using Support Vector Machine (SVM) and Random Forest (RF) algorithms. Additionally, Random Forest classifications were conducted using all features computed, without selection. The best classification performances showed that using the sQDA-selected features with SVM produced accuracy ranging from 83.15% when using three genera (Oak, Alder and Poplar). A similar result was obtained using RF and all features available for classification. The latter also achieved the best classification performances when using seven and eight genera. The results highlight that ML algorithms are suitable methods to map riparian trees.     

Region-based error concealment of right-view frames for stereoscopic video transmission

Transmission of stereoscopic video stream through error-prone wireless channels results in the loss of blocks. In this paper, we propose a region-based error concealment method that exploits encoding modes, inter-view and intra-view correlations of stereoscopic video sequences. Based on the statistical analysis of the encoding modes of the surrounding macroblocks (MBs) and of the spatially corresponding MB in the neighboring frame, the lost MBs are classified into three types: (1) smooth MBs, (2) regular motion MBs and (3) irregular motion MBs. Following the classification, corresponding operations, including direct replacement, quarter-pixel motion and disparity compensation, and bilateral error concealment with adjustable weights, are applied to reconstruct the smooth MBs, regular motion MBs and irregular motion MBs, respectively. Experimental results show that the proposed method outperforms other existing error concealment methods for stereoscopic video transmission in terms of both objective and subjective evaluations.     

多特征层次化答案质量评价方法研究

社交媒体中的问答对可以为自动问答系统提供答案,但有些答案的质量不高,因此答案质量评价方法具有研究价值。已有的评价方法没有考虑问题类别特征,对不同类型的问题采用统一的评价方法。因此提出了一个层次分类模型。首先分析问题类型;然后提取文本、非文本、语言翻译性、答案中的链接数4类特征,依据特征分类影响力随问题类型不同而不同这一客观现象,采用逻辑回归算法对各类型问题的答案质量进行评价,取得了较好的实验效果;最后分析了影响各类问题答案质量的主要特征。     

An evaluation of fuzzy and texture-based classification approaches for mapping regenerating tropical forest classes from Landsat-TM data

Two classification approaches were investigated for the mapping of tropical forests from Landsat-TM data of a region north of Manaus in the Brazilian state of Amazonas. These incorporated textural information and made use of fuzzy approaches to classification. In eleven class classifications the texture-based classifiers (based on a Markov random field model) consistently provided higher classification accuracies than conventional per-pixel maximum likelihood and minimum distance classifications, indicating that they are more able to characterize accurately several regenerating forest classes. Measures of the strength of class memberships derived from three classification algorithms (based on the probability density function, a posteriori probability and the Mahalanobis distance) could be used to derive fuzzy image classifications and be used in post-classification processing. The latter, involving either the summation of class memberships over a local neighbourhood or the application of homogeneity measures, were found to increase classification accuracy by some 10 per cent in comparison with a conventional maximum likelihood classification, a result of comparable accuracy to that derived from the texture-based classifications.     

Representing classifier confidence in the safety critical domain: an illustration from mortality prediction in trauma cases

This work proposes a novel approach to assessing confidence measures for software classification systems in demanding applications such as those in the safety critical domain. Our focus is the Bayesian framework for developing a model-averaged probabilistic classifier implemented using Markov chain Monte Carlo (MCMC) and where appropriate its reversible jump variant (RJ-MCMC). Within this context we suggest a new technique, building on the reject region idea, to identify areas in feature space that are associated with “unsure” classification predictions. We term such areas “uncertainty envelopes” and they are defined in terms of the full characteristics of the posterior predictive density in different regions of the feature space. We argue this is more informative than use of a traditional reject region which considers only point estimates of predictive probabilities. Results from the method we propose are illustrated on synthetic data and also usefully applied to real life safety critical systems involving medical trauma data.