基于自动子空间划分的高光谱数据特征提取

针对遥感高光谱图像数据量大、维数高的特点,提出了一种自动子空间划分方法用于高光谱图像数据量减小处理。该方法主要包括3个处理步骤:数据空间划分,子空间主成分分析和基于类别可分性准则的特征选择。该方法充分利用了高光谱图像各波段数据之间的局部相关性,将整个数据划分为若干个具有较强相关性的独立子空间,然后在子空间内利用主成分分析进行特征提取,根据各类地物间的类别可分性选择有效特征,最后利用地物分类来验证该方法的有效性。实验结果表明,该方法能够有效地实现高光谱图像数据维数减小和特征提取,同现有的自适应子空间分解方法和分段主成分变换方法相比,该方法所提取的特征用于分类时能获得较好的分类精度。利用该方法进行处理,当高光谱数据维数降低了90%时,9类地物分类实验的总体分类精度可以达到80.2%。     

A pairwise decision tree framework for hyperspectral classification

A novel pairwise decision tree (PDT) framework is proposed for hyperspectral classification, where no partitions and clustering are needed and the original C‐class problem is divided into a set of two‐class problems. The top of the tree includes all original classes. Each internal node consists of either a set of class pairs or a set of class pairs and a single class. The pairs are selected by the proposed sequential forward selection (SFS) or sequential backward selection (SBS) algorithms. The current node is divided into next‐stage nodes by excluding either class of each selected pair. In the classification, an unlabelled pixel is recursively classified into the next node, by excluding the less similar class of each node pair until the classification result is obtained. Compared to the single‐stage classifier approach, the pairwise classifier framework and the binary hierarchical classifier (BHC), experiments on an Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data set for a nine‐class problem demonstrated the effectiveness of the proposed framework.     

Cost-sensitive and modular land-cover classification based on posterior probability estimates

Many types of nonlinear classifiers have been proposed to automatically generate land-cover maps from satellite images. Some are based on the estimation of posterior class probabilities, whereas others estimate the decision boundary directly. In this paper, we propose a modular design able to focus the learning process on the decision boundary by using posterior probability estimates. To do so, we use a self-configuring architecture that incorporates specialized modules to deal with conflicting classes, and we apply a learning algorithm that focuses learning on the posterior probability regions that are critical for the performance of the decision problem stated by the user-defined misclassification costs. Moreover, we show that by filtering the posterior probability map, the impulsive noise, which is a common effect in automatic land-cover classification, can be significantly reduced. Experimental results show the effectiveness of the proposed solutions on real multi- and hyperspectral images, versus other typical approaches, that are not based on probability estimates, such as Support Vector Machines.     

A belief classification rule for imprecise data

The classification of imprecise data is a difficult task in general because the different classes can partially overlap. Moreover, the available attributes used for the classification are often insufficient to make a precise discrimination of the objects in the overlapping zones. A credal partition (classification) based on belief functions has already been proposed in the literature for data clustering. It allows the objects to belong (with different masses of belief) not only to the specific classes, but also to the sets of classes called meta-classes which correspond to the disjunction of several specific classes. In this paper, we propose a new belief classification rule (BCR) for the credal classification of uncertain and imprecise data. This new BCR approach reduces the misclassification errors of the objects difficult to classify by the conventional methods thanks to the introduction of the meta-classes. The objects too far from the others are considered as outliers. The basic belief assignment (bba) of an object is computed from the Mahalanobis distance between the object and the center of each specific class. The credal classification of the object is finally obtained by the combination of these bba’s associated with the different classes. This approach offers a relatively low computational burden. Several experiments using both artificial and real data sets are presented at the end of this paper to evaluate and compare the performances of this BCR method with respect to other classification methods.     

Gray Wolf Optimizer for hyperspectral band selection

In this paper, we propose a new optimization-based framework to reduce the dimensionality of hyperspectral images. One of the most problems in hyperspectral image classification is the Hughes phenomenon caused by the irrelevant spectral bands and the high correlation between the adjacent bands. The problematic is how to find the relevant bands to classify the pixels of hyperspectral image without reducing the classification accuracy rate. We propose to reformulate the problem of band selection as a combinatorial problem by modeling an objective function based on class separability measures and the accuracy rate. We use the Gray Wolf Optimizer, which is a new meta-heuristic algorithm more efficient than Practical Swarm Optimization, Gravitational Search Algorithm, Differential Evolution, Evolutionary Programming and Evolution Strategy. The experimentations are performed on three widely used benchmark hyperspectral datasets. Comparisons with the state-of-the-art approaches are also conducted. The analysis of the results proves that the proposed approach can effectively investigate the spectral band selection problem and provides a high classification accuracy rate by using a few samples for training.     

Task decomposition and module combination based on class relations: a modular neural network for pattern classification.

We propose a method for decomposing pattern classification problems based on the class relations among training data. By using this method, we can divide a K-class classification problem into a series of ((2)(K)) two-class problems. These two-class problems are to discriminate class C(i) from class C(j) for i=1, ..., K and j=i+1, while the existence of the training data belonging to the other K-2 classes is ignored. If the two-class problem of discriminating class C(i) from class C(j) is still hard to be learned, we can further break down it into a set of two-class subproblems as small as we expect. Since each of the two-class problems can be treated as a completely separate classification problem with the proposed learning framework, all of the two-class problems can be learned in parallel. We also propose two module combination principles which give practical guidelines in integrating individual trained network modules. After learning of each of the two-class problems with a network module, we can easily integrate all of the trained modules into a min-max modular (M(3)) network according to the module combination principles and obtain a solution to the original problem. Consequently, a large-scale and complex K-class classification problem can be solved effortlessly and efficiently by learning a series of smaller and simpler two-class problems in parallel.     

价值样本选取的不均衡分类

基于传统模型的实际分类问题,不均衡分类是一个常见的挑战问题。由于传统分类器较难学习少数类数据集内部的本质结构,导致更多地偏向于多数类,从而使少数类样本被误分为多数类样本。与此同时,样本集中的冗余数据和噪音数据也会对分类器造成困扰。为有效处理上述问题,提出一种新的不均衡分类框架SSIC,该框架充分考虑数据统计特性,自适应从大小类中选取有价值样本,并结合代价敏感学习构建不均衡数据分类器。首先,SSIC通过组合部分多数类实例和所有少数类实例来构造几个平衡的数据子集。在每个子集上,SSIC充分利用数据的特征来提取可区分的高级特征并自适应地选择重要样本,从而可以去除冗余噪声数据。其次,SSIC通过在每个样本上自动分配适当的权重来引入一种代价敏感的支持向量机(SVM),以便将少数类视为与多数类相等。     

Joint spectral-spatial hyperspectral image classification based on hierarchical subspace switch ensemble learning algorithm

In this paper, a novel spectral-spatial hyperspectral image classification method has been proposed by designing hierarchical subspace switch ensemble learning algorithm. First, the hyperspectral images are processed by fast bilateral filtering to get the spatial features. The spectral features and spatial features are combined to form the initial feature set. Second, Hierarchical instance learning based on iterative means clustering method is designed to obtain hierarchical instance space. Third, random subspace method (RSM) is used for sampling the features and samples, thereby forming multiple sub sample set. After that, semi-supervised learning (S²L) is applied to choose test samples for improving classification performance without touching the class labels. Then, micro noise linear dimension reduction (mNLDR) is used for dimension reduction. Afterwards, ensemble multiple kernels SVM(EMK_SVM) are used for stable classification results. Finally, final classification results are obtained by combining classification results with voting strategy. Experimental results on real hyperspectral scenes demonstrate that the proposed method can effectively improve the classification performance apparently.     

高光谱图像分类的融合分层深度网络联合稀疏表示算法

在高光谱图像分类领域中每个像素的局部邻域一旦包含来自不同类别的样本,联合稀疏表示将受邻域内字典原子与测试样本之间同谱异类的影响,严重降低分类性能.根据高光谱图像的特点,文中提出融合分层深度网络的联合稀疏表示算法.在光谱和空间特征学习之间交替提取判别性光谱信息和空间信息,构建兼具空谱特征的学习字典,用于联合稀疏表示.在分类过程中将学习字典与测试样本间的相关系数与分类误差融合并决策.在两个高光谱遥感数据集上的实验验证文中算法的有效性.     

Active Rare Class Discovery and Classification Using Dirichlet Processes

Classification is used to solve countless problems. Many real world computer vision problems, such as visual surveillance, contain uninteresting but common classes alongside interesting but rare classes. The rare classes are often unknown, and need to be discovered whilst training a classifier. Given a data set active learning selects the members within it to be labelled for the purpose of constructing a classifier, optimising the choice to get the best classifier for the least amount of effort. We propose an active learning method for scenarios with unknown, rare classes, where the problems of classification and rare class discovery need to be tackled jointly. By assuming a non-parametric prior on the data the goals of new class discovery and classification refinement are automatically balanced, without any tunable parameters. The ability to work with any specific classifier is maintained, so it may be used with the technique most appropriate for the problem at hand. Results are provided for a large variety of problems, demonstrating superior performance.     

Credal classification rule for uncertain data based on belief functions

In this paper we present a new credal classification rule (CCR) based on belief functions to deal with the uncertain data. CCR allows the objects to belong (with different masses of belief) not only to the specific classes, but also to the sets of classes called meta-classes which correspond to the disjunction of several specific classes. Each specific class is characterized by a class center (i.e. prototype), and consists of all the objects that are sufficiently close to the center. The belief of the assignment of a given object to classify with a specific class is determined from the Mahalanobis distance between the object and the center of the corresponding class. The meta-classes are used to capture the imprecision in the classification of the objects when they are difficult to correctly classify because of the poor quality of available attributes. The selection of meta-classes depends on the application and the context, and a measure of the degree of indistinguishability between classes is introduced. In this new CCR approach, the objects assigned to a meta-class should be close to the center of this meta-class having similar distances to all the involved specific classes? centers, and the objects too far from the others will be considered as outliers (noise). CCR provides robust credal classification results with a relatively low computational burden. Several experiments using both artificial and real data sets are presented at the end of this paper to evaluate and compare the performances of this CCR method with respect to other classification methods.     

Combining linear discriminant functions with neural networks for supervised learning

A novel supervised learning method is proposed by combining linear discriminant functions with neural networks. The proposed method results in a tree-structured hybrid architecture. Due to constructive learning, the binary tree hierarchical architecture is automatically generated by a controlled growing process for a specific supervised learning task. Unlike the classic decision tree, the linear discriminant functions are merely employed in the intermediate level of the tree for heuristically partitioning a large and complicated task into several smaller and simpler subtasks in the proposed method. These subtasks are dealt with by component neural networks at the leaves of the tree accordingly. For constructive learning, growing and credit-assignment algorithms are developed to serve for the hybrid architecture. The proposed architecture provides an efficient way to apply existing neural networks (e.g. multi-layered perceptron) for solving a large scale problem. We have already applied the proposed method to a universal approximation problem and several benchmark classification problems in order to evaluate its performance. Simulation results have shown that the proposed method yields better results and faster training in comparison with the multilayered perceptron.     

Search-based structured prediction

We present Searn, an algorithm for integrating search and learning to solve complex structured prediction problems such as those that occur in natural language, speech, computational biology, and vision. Searn is a meta-algorithm that transforms these complex problems into simple classification problems to which any binary classifier may be applied. Unlike current algorithms for structured learning that require decomposition of both the loss function and the feature functions over the predicted structure, Searn is able to learn prediction functions for any loss function and any class of features. Moreover, Searn comes with a strong, natural theoretical guarantee: good performance on the derived classification problems implies good performance on the structured prediction problem.     

Incorporating prior knowledge into learning by dividing training data

In most large-scale real-world pattern classification problems, there is always some explicit information besides given training data, namely prior knowledge, with which the training data are organized. In this paper, we proposed a framework for incorporating this kind of prior knowledge into the training of min-max modular (M³) classifier to improve learning performance. In order to evaluate the proposed method, we perform experiments on a large-scale Japanese patent classification problem and consider two kinds of prior knowledge included in patent documents: patent’s publishing date and the hierarchical structure of patent classification system. In the experiments, traditional support vector machine (SVM) and M³-SVM without prior knowledge are adopted as baseline classifiers. Experimental results demonstrate that the proposed method is superior to the baseline classifiers in terms of training cost and generalization accuracy. Moreover, M³-SVM with prior knowledge is found to be much more robust than traditional support vector machine to noisy dated patent samples, which is crucial for incremental learning.     

Automatic cervical cell segmentation and classification in Pap smears

Cervical cancer is one of the leading causes of cancer death in females worldwide. The disease can be cured if the patient is diagnosed in the pre-cancerous lesion stage or earlier. A common physical examination technique widely used in the screening is Papanicolaou test or Pap test. In this research, a method for automatic cervical cancer cell segmentation and classification is proposed. A single-cell image is segmented into nucleus, cytoplasm, and background, using the fuzzy C-means (FCM) clustering technique. Four cell classes in the ERUDIT and LCH datasets, i.e., normal, low grade squamous intraepithelial lesion (LSIL), high grade squamous intraepithelial lesion (HSIL), and squamous cell carcinoma (SCC), are considered. The 2-class problem can be achieved by grouping the last 3 classes as one abnormal class. Whereas, the Herlev dataset consists of 7 cell classes, i.e., superficial squamous, intermediate squamous, columnar, mild dysplasia, moderate dysplasia, severe dysplasia, and carcinoma in situ. These 7 classes can also be grouped to form a 2-class problem. These 3 datasets were tested on 5 classifiers including Bayesian classifier, linear discriminant analysis (LDA), K-nearest neighbor (KNN), artificial neural networks (ANN), and support vector machine (SVM). For the ERUDIT dataset, ANN with 5 nucleus-based features yielded the accuracies of 96.20% and 97.83% on the 4-class and 2-class problems, respectively. For the Herlev dataset, ANN with 9 cell-based features yielded the accuracies of 93.78% and 99.27% for the 7-class and 2-class problems, respectively. For the LCH dataset, ANN with 9 cell-based features yielded the accuracies of 95.00% and 97.00% for the 4-class and 2-class problems, respectively. The segmentation and classification performances of the proposed method were compared with that of the hard C-means clustering and watershed technique. The results show that the proposed automatic approach yields very good performance and is better than its counterparts.     

An all-pair quantum SVM approach for big data multiclass classification

In this paper, we discuss a quantum approach for the all-pair multiclass classification problem. In an all-pair approach, there is one binary classification problem for each pair of classes, and so there are k(k???1)/2 classifiers for a k-class classification problem. As compared to the classical multiclass support vector machine that can be implemented with polynomial run time complexity, our approach exhibits exponential speedup due to quantum computing. The quantum all-pair algorithm can also be used with other classification algorithms, and a speedup gain can be achieved as compared to their classical counterparts.     

基于LFDA和GA-ELM的高光谱图像地物识别方法研究

高光谱图像的高维特性和波段间的高相关性,导致高光谱图像地物识别问题研究中,面临着数据量大、信息冗余的问题,降低了高光谱图像的分类识别精度。针对以上问题,提出了基于局部保留降维（Local Fisher Discriminant Analysis,LFDA）结合遗传算法（Genetic Algorithm, GA ）优化极限学习机（Extreme Learning Machine, ELM）的高光谱图像分类方法。首先,采用LFDA对高光谱图像数据进行降维处理,消除信息冗余并保留局部邻域内主要特征;然后用GA优化ELM,对降维处理后的特征样本进行分类,提高高光谱图像的分类识别精度。将该方法应用于Salinas和Pavia University高光谱图像的地物识别问题研究,分类精度分别达到了98.56%和97.11%,由此验证了该方法的有效性。     

Discriminant ECOC: a heuristic method for application dependent design of error correcting output codes

We present a heuristic method for learning error correcting output codes matrices based on a hierarchical partition of the class space that maximizes a discriminative criterion. To achieve this goal, the optimal codeword separation is sacrificed in favor of a maximum class discrimination in the partitions. The creation of the hierarchical partition set is performed using a binary tree. As a result, a compact matrix with high discrimination power is obtained. Our method is validated using the UCI database and applied to a real problem, the classification of traffic sign images.     

On combining multiscale deep learning features for the classification of hyperspectral remote sensing imagery

In recent years, satellite imagery has greatly improved in both spatial and spectral resolution. One of the major unsolved problems in highly developed remote sensing imagery is the manual selection and combination of appropriate features according to spectral and spatial properties. Deep learning framework can learn global and robust features from the training data set automatically, and it has achieved state-of-the-art classification accuracies over different image classification tasks. In this study, a technique is proposed which attempts to classify hyperspectral imagery by incorporating deep learning features. Firstly, deep learning features are extracted by multiscale convolutional auto-encoder. Then, based on the learned deep learning features, a logistic regression classifier is trained for classification. Finally, parameters of deep learning framework are analysed and the potential development is introduced. Experiments are conducted on the well-known Pavia data set which is acquired by the reflective optics system imaging spectrometer sensor. It is found that the deep learning-based method provides a more accurate classification result than the traditional ones.     

Learn class hierarchy using convolutional neural networks

A large amount of research on Convolutional Neural Networks (CNN) has focused on flat Classification in the multi-class domain. In the real world, many problems are naturally expressed as hierarchical classification problems, in which the classes to be predicted are organized in a hierarchy of classes. In this paper, we propose a new architecture for hierarchical classification, introducing a stack of deep linear layers using cross-entropy loss functions combined to a center loss function. The proposed architecture can extend any neural network model and simultaneously optimizes loss functions to discover local hierarchical class relationships and a loss function to discover global information from the whole class hierarchy while penalizing class hierarchy violations. We experimentally show that our hierarchical classifier presents advantages to the traditional classification approaches finding application in computer vision tasks. The same approach can also be applied to some CNN for text classification.