Patch-based active learning (PTAL) for spectral-spatial classification on hyperspectral data

Active learning (AL) has been shown to be a useful approach to improving the efficiency of the classification process for remote-sensing imagery. Current AL methods are essentially based on pixel-wise classification. In this paper, a new patch-based active learning (PTAL) framework is proposed for spectral-spatial classification on hyperspectral remote-sensing data. The method consists of two major steps. In the initialization stage, the original hyperspectral images are partitioned into overlapping patches. Then, for each patch, the spectral and spatial information as well as the label are extracted. A small set of patches is randomly selected from the data set for annotation, then a patch-based support vector machine (PTSVM) classifier is initially trained with these patches. In the second stage (close-loop stage of query and retraining), the trained PTSVM classifier is combined with one of three query methods, which are margin sampling (MS), entropy query-by-bagging (EQB), and multi-class level uncertainty (MCLU), and is subsequently employed to query the most informative samples from the candidate pool comprising the rest of the patches from the data set. The query selection cycle enables the PTSVM model to select the most informative queries for human annotation. Then, these informative queries are added to the training set. This process runs iteratively until a stopping criterion is met. Finally, the trained PTSVM is employed to patch classification. In order to compare this to pixel-based active learning (PXAL) models, the prediction label of a patch by PTSVM is transformed into a pixel-wise label of a pixel predictor to get the classification maps. Experimental results show better performance of the proposed PTAL methods on classification accuracy and computational time on three different hyperspectral data sets as compared with PXAL methods.     

Active learning for hyperspectral image classification using sparse code histogram and graph-based spatial refinement

In order to address the challenge of hyperspectral image (HSI) classification with very limited labelled samples, active learning (AL) has become a hot research issue in recent years. Although lots of AL approaches have been proposed in the literature, most of them concentrate on how to select the most informative samples, while ignore the significance of the input feature. We believe that the input feature and the query selection are both crucial for constituting an efficient AL algorithm. In this article, we propose a new discriminative feature, sparse code histogram (SCH), to conduct the AL procedure. SCH exhibits a much stronger distinguishability than several other widely used features, and thus a better AL performance could be expected. With this novel input feature, a probabilistic classifier, multinomial logistic regression, is trained to obtain the class probability of each sample. Considering that the class probability is usually biased due to the limited labelled samples, a graph-based spatial refinement is proposed to refine the class probability by exploiting the contextual information. Based on the refined class probability, informative samples are selected for manual labelling and classifier retraining. Such a process is iterated until a stopping criterion is met. Experimental results demonstrate that the proposed method could usually achieve above 90% classification accuracy with only two iterations, which significantly outperforms several state-of-the-art approaches.     

Gene subset selection using an iterative approach based on genetic algorithms

Microarray data are expected to be useful for cancer classification. However, the process of gene selection for the classification contains a major problem due to properties of the data such as the small number of samples compared with the huge number of genes (higher-dimensional data), irrelevant genes, and noisy data. Hence, this article aims to select a near-optimal (small) subset of informative genes that is most relevant for the cancer classification. To achieve this aim, an iterative approach based on genetic algorithms has been proposed. Experimental results show that the performance of the proposed approach is superior to other previous related work, as well as to four methods tried in this work. In addition, a list of informative genes in the best gene subsets is also presented for biological usage.     

基于邻域熵的高光谱波段选择算法

为了减少高光谱图像数据中的冗余信息,优化计算效率,并提升图像数据后续应用的有效性,提出一种基于邻域熵(NE)的高光谱波段选择算法.首先,为了高效计算样本的邻域子集,采用了局部敏感哈希(LSH)作为近似最近邻的搜索策略;然后,引入了NE理论来度量波段和类之间的互信息(MI),并把最小化特征集合与类变量之间的条件熵作为选取...     

Unsupervised hyperspectral feature selection based on fuzzy c-means and grey wolf optimizer

Hyperspectral image (HSI) with hundreds of narrow and consecutive spectral bands provides substantial information to discriminate various land-covers. However, the existence of redundant features/bands not only gives rise to increasing of computation time but also interferes the classification result of hyperspectral images. Obviously, it is a very challenging problem how to select an effective feature subset from original bands to reduce the dimensionality of the hyperspectral dataset. In this study, a novel unsupervised feature selection method is suggested to remove the redundant features of HSI by feature subspace decomposition and optimization of feature combination. Feature subset decomposition is achieved by the fuzzy c-means (FCM) algorithm. The optimal feature selection is based on the optimization process of grey wolf optimizer (GWO) algorithm and maximum entropy (ME) principle. To evaluate the effectiveness of the proposed method, experiments are conducted on three well-known hyperspectral datasets, Indian Pines, Pavia University, and Salinas. Six state-of-the-art feature selection methods are used to compare with the proposed method. Experimental results successfully confirm the superior performance of our proposal with respect to three classification accuracy indices overall accuracy (OA), average accuracy (AA) and kappa coefficient (κ).     

A novel logistic multi-class supervised classification model based on multi-fractal spectrum parameters for hyperspectral data

A novel logistic multi-class supervised classification model based on multi-fractal spectrum parameters is proposed to avoid the error that is caused by the difference between the real data distribution and the hypothetic Gaussian distribution and avoid the computational burden working in the logistic regression classification directly for hyperspectral data. The multi-fractal spectra and parameters are calculated firstly with training samples along the spectral dimension of hyperspectral data. Secondly, the logistic regression model is employed in our work because the logistic regression classification model is a distribution-free nonlinear model which is based on the conditional probability without the Gaussian distribution assumption of the random variables, and the obtained multi-fractal parameters are applied to establish the multi-class logistic regression classification model. Finally, the Newton–Raphson method is applied to estimate the model parameters via the maximum likelihood algorithm. The classification results of the proposed model are compared with the logistic regression classification model based on an adaptive bands selection method by using the Airborne Visible/Infrared Imaging Spectrometer and airborne Push Hyperspectral Imager data. The results illuminate that the proposed approach achieves better accuracy with lower computational cost simultaneously.     

基于自适应主动学习与联合双边滤波的高光谱图像分类

在高光谱图像分类中,选择合适的样本作为训练样本对分类器进行训练非常重要。将样本的不确定性与代表性相结合,通过自适应主动学习方法来完成样本的选择。用核K均值聚类来获取具有代表性的样本,用最优标号和次优标号的概率差值与两者比值的加权和来度量不确定性。此外,为了提高分类的准确率,利用联合双边滤波来获取高光谱图像的空间信息,并将其融入分类过程中。最后,提出一种融合自适应主动学习与联合双边滤波的空谱结合高光谱图像分类方法,并通过实验验证了所提方法的优越性。     

Feature selection for hyperspectral data based on recursive support vector machines

In this article, a feature selection algorithm for hyperspectral data based on a recursive support vector machine (R‐SVM) is proposed. The new algorithm follows the scheme of a state‐of‐the‐art feature selection algorithm, SVM recursive feature elimination or SVM‐RFE, and uses a new ranking criterion derived from the R‐SVM. Multiple SVMs are used to address the multiclass problem. The algorithm is applied to Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data to select the most informative bands and the resulting subsets of the bands are compared with SVM‐RFE using the accuracy of classification as the evaluation of the effectiveness of the feature selection. The experimental results for an agricultural case study indicate that the feature subset generated by the newly proposed algorithm is generally competitive with SVM‐RFE in terms of classification accuracy and is more robust in the presence of noise.     

A label propagation method using spatial-spectral consistency for hyperspectral image classification

In this article, a label propagation approach with automatic seed selection is developed for hyperspectral image classification. The core idea is to combine pixel-wise classification results with spatial information described by a data graph. Using only the support vector machine (SVM) classifier on spectral features to tackle the hyperspectral classification task will produce results with a salt-and-pepper appearance. To overcome this limitation, the spatial information is incorporated by label propagation. The performance of label propagation is dependent on two points: the seeds and the connection graph. Generally, a limited number of labelled samples are available, which are considered as seeds in label propagation. However, the limited seeds will result in bad label propagation. Therefore, pseudo-seeds are automatically selected in local windows. Specifically, the pixels whose initial labels according to SVM are consistent with their most spatial neighbours are selected as seeds. Through seed selection, the number of seeds is greatly increased. Then, the label information of the selected seeds is propagated to their spatial neighbours using a data graph which is constructed according to the local structures in the image. Through seed selection and label propagation on the graph, the problem of salt-and-pepper appearance is solved elegantly – the noisy labels are highly suppressed and most of the structures are preserved. Competitive experimental results on a variety of hyperspectral data sets demonstrate the effectiveness of the proposed method.     

Semi-supervised classification framework of hyperspectral images based on the fusion evidence entropy

Increasing attention is being paid to the classification of ground objects using hyperspectral spectrometer images. A key challenge of most hyperspectral classifications is the cost of training samples. It is difficult to acquire enough effective marked label sets using classification model frameworks. In this paper, a semi-supervised classification framework of hyperspectral images is proposed to better solve problems associated with hyperspectral image classification. The proposed method is based on an iteration process, making full use of the small amount of labeled data in a sample set. In addition, a new unlabeled data trainer in the self-training semi-supervised learning framework is explored and implemented by estimating the fusion evidence entropy of unlabeled samples using the minimum trust evaluation and maximum uncertainty. Finally, we employ different machine learning classification methods to compare the classification performance of different hyperspectral images. The experimental results indicate that the proposed approach outperforms traditional state-of-the-art methods in terms of low classification errors and better classification charts using few labeled samples.     

一种新的高光谱遥感图像降维方法

高光谱遥感图像的高数据维给图像进一步处理带来了困难，为了解决这一问题，提出了自适应波段选择(ABS)的降维方法。该方法充分考虑了高光谱图像的空间相关性和谱间相关性，通过计算各个波段的指数来选择信息量大并且与其他波段相关性小的波段。对各波段相应的指数重新排列之后，有两种方法来选择最终波段：一种是选择波段指数比设定指数大的波段，另一种方法是选择波段指数排在前n个的所有波段。为了验证ABS方法的有效性，对降维后的高光谱图像进行了贝叶斯监督分类，分类结果表明自适应波段选择的方法能够选择出信息丰富的波段，分类精度与使用原始波段相比提高10．4％，计算复杂度大大降低。     

On the use of small training sets for neural network-based characterization of mixed pixels in remotely sensed hyperspectral images

In this work, neural network-based models involved in hyperspectral image spectra separation are considered. Focus is on how to select the most highly informative samples for effectively training the neural architecture. This issue is addressed here by several new algorithms for intelligent selection of training samples: (1) a border-training algorithm (BTA) which selects training samples located in the vicinity of the hyperplanes that can optimally separate the classes; (2) a mixed-signature algorithm (MSA) which selects the most spectrally mixed pixels in the hyperspectral data as training samples; and (3) a morphological-erosion algorithm (MEA) which incorporates spatial information (via mathematical morphology concepts) to select spectrally mixed training samples located in spatially homogeneous regions. These algorithms, along with other standard techniques based on orthogonal projections and a simple Maximin-distance algorithm, are used to train a multi-layer perceptron (MLP), selected in this work as a representative neural architecture for spectral mixture analysis. Experimental results are provided using both a database of nonlinear mixed spectra with absolute ground truth and a set of real hyperspectral images, collected at different altitudes by the digital airborne imaging spectrometer (DAIS 7915) and reflective optics system imaging spectrometer (ROSIS) operating simultaneously at multiple spatial resolutions.     

基于信噪比估计的波段选择与高光谱异常检测

有效的波段选择方法可以极大地提高高光谱图像处理速度的同时改善处理效果。为了自动判断低信噪比波段,提出了一种基于小波变换的图像信噪比估计(SNR estimation,SNRE)方法,利用小波变换后对角方向上的高频成分估计噪声方差并计算信噪比。将该方法分别结合基于方差和相关系数(V_COR)的最优索引指数、最大信息量(MI)、高阶矩(偏度或峰度)结合信息散度(K3_KL)等3种基于信息量的波段选择方法后选择波段。将这些改进后的波段选择方法应用于高光谱异常检测。实验结果表明SNRE预选波段结合MI和K3_KL选择波段用于异常检测能进一步提高检测精度。     

An automatic classification of text documents based on correlative association of words

Training speed of the classifier without degrading its predictive capability is an important concern in text classification. Feature selection plays a key role in this context. It selects a subset of most informative words (terms) from the set of all words. The correlative association of words towards the classes increases an incertitude for the words to represent a class. The representative words of a class are either of positive or negative nature. The standard feature selection methods, viz. Mutual Information (MI), Information Gain (IG), Discriminating Feature Selection (DFS) and Chi Square (CHI), do not consider positive and negative nature of the words that affects the performance of the classifiers. To address this issue, this paper presents a novel feature selection method named Correlative Association Score (CAS). It combines the strength, mutual information, and strong association of the words to determine their positive and negative nature for a class. CAS selects a few (k) informative words from the set of all words (m). These informative words generate a set of N-grams of length 1-3. Finally, the standard Apriori algorithm ensembles the power of CAS and CHI to select the top most, b informative N-grams, where b is a number set by an empirical evaluation. Multinomial Naive Bayes (MNB) and Linear Support Vector Machine (LSVM) classifiers evaluate the performance of the selected N-Grams. Four standard text data sets, viz. Webkb, 20Newsgroup, Ohsumed10, and Ohsumed23 are used for experimental analysis. Two standard performance measures named Macro_F1 and Micro_F1 show a significant improvement in the results using proposed CAS method.     

Semisupervised self-learning granular neural networks for remote sensing image classification

Accuracy of a pattern classification model mostly depends on ample number of training samples, which is the major bottleneck for classifying land cover of remote sensing images. Further, the unbalance scenario typically encountered in hyperspectral remote sensing images, i.e., limited number of training samples with more dimensions, makes the decision-making process cumbersome. Under such inevitable constraints, the article aims to develop an improved classification model using semisupervised self-learning granular neural networks (GNNs) for remote sensing images. The proposed semisupervised method has adopted a new strategy for selecting the potential candidate samples from the unlabeled dataset and used GNN as the base classifier. We have considered GNN because of its transparent architecture that leads to improved performance with less computational complexity compared to the conventional neural networks. Performance of the model is further enhanced with fuzzy granulation of features using class belonging information and selection of granulated features using neighborhood rough sets (NRS). The proposed model thus takes the mutual advantages of GNN architecture, fuzzy granulation with class belonging information, NRS-based feature selection and the most important, improved semisupervised self-learning approach. Performance of the model is compared with other similar methods and verified in terms of different performance measurement indexes, using two multispectral and two hyperspectral remote sensing images.     

Modified CC-LR algorithm with three diverse feature sets for motor imagery tasks classification in EEG based brain–computer interface

Motor imagery (MI) tasks classification provides an important basis for designing brain–computer interface (BCI) systems. If the MI tasks are reliably distinguished through identifying typical patterns in electroencephalography (EEG) data, a motor disabled people could communicate with a device by composing sequences of these mental states. In our earlier study, we developed a cross-correlation based logistic regression (CC-LR) algorithm for the classification of MI tasks for BCI applications, but its performance was not satisfactory. This study develops a modified version of the CC-LR algorithm exploring a suitable feature set that can improve the performance. The modified CC-LR algorithm uses the C3 electrode channel (in the international 10–20 system) as a reference channel for the cross-correlation (CC) technique and applies three diverse feature sets separately, as the input to the logistic regression (LR) classifier. The present algorithm investigates which feature set is the best to characterize the distribution of MI tasks based EEG data. This study also provides an insight into how to select a reference channel for the CC technique with EEG signals considering the anatomical structure of the human brain. The proposed algorithm is compared with eight of the most recently reported well-known methods including the BCI III Winner algorithm. The findings of this study indicate that the modified CC-LR algorithm has potential to improve the identification performance of MI tasks in BCI systems. The results demonstrate that the proposed technique provides a classification improvement over the existing methods tested.     

Hyperspectral band selection with objective image quality assessment

Hyperspectral sensors often collect hundreds of bands at a time, so hyperspectral images can accurately characterize different land-cover types with abundant spectral information. However, these spectral bands also contain redundant information that needs to be removed. Band selection is one of the most widely used methods to remove noised or redundant bands. Because labelled samples are difficult to collect, most band selection methods adopt unsupervised ways to select diverse and representative bands. Still, noised bands are often selected because they usually have low correlation with other bands. In this article, objective image quality assessment is introduced to indicate the quality of every band, and combined with the redundancy measure, a new unsupervised band selection method is proposed. Three real hyperspectral images are used to demonstrate the effectiveness of the proposed algorithm.     

Multi-feature extraction and selection in writer-independent off-line signature verification

Some of the fundamental problems faced in the design of signature verification (SV) systems include the potentially large number of input features and users, the limited number of reference signatures for training, the high intra-personal variability among signatures, and the lack of forgeries as counterexamples. In this paper, a new approach for feature selection is proposed for writer-independent (WI) off-line SV. First, one or more preexisting techniques are employed to extract features at different scales. Multiple feature extraction increases the diversity of information produced from signature images, allowing to produce signature representations that mitigate intra-personal variability. Dichotomy transformation is then applied in the resulting feature space to allow for WI classification. This alleviates the challenges of designing off-line SV systems with a limited number of reference signatures from a large number of users. Finally, boosting feature selection is used to design low-cost classifiers that automatically select relevant features while training. Using this global WI feature selection approach allows to explore and select from large feature sets based on knowledge of a population of users. Experiments performed with real-world SV data comprised of random, simple, and skilled forgeries indicate that the proposed approach provides a high level of performance when extended shadow code and directional probability density function features are extracted at multiple scales. Comparing simulation results to those of off-line SV systems found in literature confirms the viability of the new approach, even when few reference signatures are available. Moreover, it provides an efficient framework for designing a wide range of biometric systems from limited samples with few or no counterexamples, but where new training samples emerge during operations.     

基于空间邻域信息的高光谱遥感影像半监督协同训练

针对tri_training协同训练算法在小样本的高光谱遥感影像半监督分类过程中,存在增选样本的误标记问题,提出一种基于空间邻域信息的半监督协同训练分类算法tri_training_SNI(tri_training based on Spatial Neighborhood Information)。首先利用分类器度量方法不一致度量和新提出的不一致精度度量从MLR(Multinomial Logistic Regression)、KNN(k-Nearest Neighbor)、ELM(Extreme Learning Machine)和RF(Random Forest)4个分类器中选择3分类性能差异性最大的3个分类器;然后在样本选择过程中,采用选择出来的3个分类器,在两个分类器分类结果相同的基础上,加入初始训练样本的8邻域信息进行未标记样本的二次筛选和标签的确定,提高了半监督学习的样本选择精度。通过对AVIRIS和ROSIS两景高光谱遥感影像进行分类实验,结果表明与传统的tri_training协同算法相比,该算法在分类精度方面有明显提高。