On the classification of remote sensing high spatial resolution image data

In this paper is investigated a methodology implementing an object-based approach to digital image classification using spectral and spatial attributes in a multiple-stage classifier structured as a binary tree. It is a well-established fact that object-based image classification is particularly appropriate when dealing with high spatial resolution image data. Following this approach, the image is initially segmented into objects that carry informational value. Next, spectral and spatial attributes are extracted from every object in the scene, and implemented into a classifier to produce a thematic map. As the combined number of spectral and spatial variables may become large compared to the number of available training samples, a reduction in the data dimensionality may be required whenever parametric classifiers are used, in order to mitigate the effects of the Hughes phenomenon. To this end the sequential feature selection (SFS) procedure is applied in a multiple-stage classifier structured as a binary tree. The advantage of a binary tree classifier lies in the fact that only one pair of classes is considered at each stage (node), allowing for an optimal selection of features. This proposed approach was tested using Quickbird image data covering an urban scene. The results are compared against results yielded by the traditional single-stage Gaussian maximum likelihood classifier. The results suggest the proposed methodology is adequate in the classification of high spatial resolution image data.     

基于SLIC和主动学习的高光谱遥感图像分类方法

在主动学习的基础上,提出一种基于SLIC的高光谱遥感图像主动分类方法。首先提取图像纹理特征并与光谱特征融合,使用PCA对新数据进行降维,取前三个主成分构成假彩色图像,然后使用SLIC处理该图像获得超像素;接着随机抽取定量超像素作为初始训练样本,样本光谱信息为超像素样本中所有像素点的光谱信息均值,样本标签为超像素中出现次数最多的类别;然后通过主动学习得到SVM分类器;最后使用分类器对超像素分类得到其类别,并将超像素类别赋予其包含的像素点,从而达到高光谱遥感图像分类的目的。实验表明：该方法明显降低了主动学习过程的时间消耗,有效地提高了分类效果,其OA,AA和Kappa值显著优于未使用SLIC的主动学习方法。     

GPU-based fast hyperspectral image classification using joint sparse representation with spectral consistency constraint

Due to the fact that neighboring hyperspectral pixels often belong to the same class with high probability, spatial correlation between pixels has been widely used in hyperspectral image classification. In this paper, a novel joint sparse representation classifier with spectral consistency constraint (JSRC-SCC) is proposed. Specifically, to efficiently exploit contextual structure information, a local adaptive weighted average value is reallocated as the central pixel of a window through spatial filtering, and then, representation coefficients are estimated by the joint sparse representation model, which is imposed by the spectral consistency constraint under \(\textit{l}_1\)-minimization. For the purpose of fast classification, graphics processing units are adopted to accelerate this model. Experimental results on two classical hyperspectral image data sets demonstrate the proposed method can not only produce satisfying classification performance, but also shorten the computational time significantly.     

A Multiscale Approach to Automatic Medical Image Segmentation Using Self-Organizing Map

In this paper,a new medical image classification scheme is proposed using selforganizing map(SOM)combined with multiscale technique.It addresses the problem of the handling of edge pixels in the traditional multiscale SOM classifiers.First,to solve the difficulty in manual selection of edge pixels,a multiscale edge detection algorithm based on wavelet transform is proposed.Edge pixels detected are then selected into the training set as a new class and a multiscale SOM classifier is trained using this training set.In this new scheme,the SOM classifier can perform both the classification on the entire image and the edge detection simultaneously.On the other hand,the misclassification of the traditional multiscale SOM classifier in regions near edges is graeatly reduced and the correct classification is improved at the same time.     

Edge-preserving-based collaborative representation for spectral-spatial classification

Recently, the nearest regularized subspace (NRS) classifier and its spectral–spatial versions such as joint collaborative representation (JCR) and weighted JCR (WJCR) have gained an increasing interest in the hyperspectral image classification. JCR and WJCR average each pixel with its neighbours in a spatial neighbourhood window. The use of spatial information as averaging of pixels in a local window may degrade the classification accuracy in the neighbourhood of discontinuities and class boundaries. We propose the edge-preserving-based collaborative representation (EPCR) classifier in this article, which overcomes this problem by using the edge image estimated by the original full-band hyperspectral image. The estimated edge image is used for calculation of the weights of neighbours and also the final residuals in the collaborative representation classifier. The advantage of multiscale spatial window is also assessed in this work. Moreover, the kernelized versions of NRS and its improved versions are developed in this article. Our experimental results on several popular hyperspectral images indicate that EPCR and its kernelized version are superior to some state-of-the-art classification methods.     

Per-pixel vs. object-based classification of urban land cover extraction using high spatial resolution imagery

In using traditional digital classification algorithms, a researcher typically encounters serious issues in identifying urban land cover classes employing high resolution data. A normal approach is to use spectral information alone and ignore spatial information and a group of pixels that need to be considered together as an object. We used QuickBird image data over a central region in the city of Phoenix, Arizona to examine if an object-based classifier can accurately identify urban classes. To demonstrate if spectral information alone is practical in urban classification, we used spectra of the selected classes from randomly selected points to examine if they can be effectively discriminated. The overall accuracy based on spectral information alone reached only about 63.33%. We employed five different classification procedures with the object-based paradigm that separates spatially and spectrally similar pixels at different scales. The classifiers to assign land covers to segmented objects used in the study include membership functions and the nearest neighbor classifier. The object-based classifier achieved a high overall accuracy (90.40%), whereas the most commonly used decision rule, namely maximum likelihood classifier, produced a lower overall accuracy (67.60%). This study demonstrates that the object-based classifier is a significantly better approach than the classical per-pixel classifiers. Further, this study reviews application of different parameters for segmentation and classification, combined use of composite and original bands, selection of different scale levels, and choice of classifiers. Strengths and weaknesses of the object-based prototype are presented and we provide suggestions to avoid or minimize uncertainties and limitations associated with the approach.     

Mapping mixed vegetation communities in salt marshes using airborne spectral data

The aim of this study is to evaluate a new neural network classifier using spectrally sampled image data to map mixed halophytic vegetation in tidal environments. The work is based on the concept of vegetation communities, mixtures of several species, characteristic of salt marshes. The study site is the Venice lagoon, and the material available is a spectrally sampled Compact Airborne Spectral Imager (CASI) image, in conjunction with ground truth for precise characterization of vegetation communities. Detailed observations of vegetation species and of their fractional abundance were collected for 36 Regions Of Interest (ROI): such field polygons are used for classification training and accuracy assessment. To select the most significant spectral channels, the Spectral Reconstruction method was applied to the image data: a set of 6 bands was selected as optimal for classification, out of the 15 available. The spatial heterogeneity of salt-marsh vegetation is significant and even at the spatial resolution of the airborne CASI image data, mixed pixels are observed. The Vegetation Community based Neural Network Classifier (VCNNC) is introduced to cope with a situation where no pure pixels exist, and was applied to the set of 6 selected bands. Both quantitative and qualitative comparisons of classification results of VCNNC with those of conventional Neural Network Classifier (NNC), trained and assessed on exactly the same data sets, shows that VCNNC's accuracy is substantially higher (≈ 91%) than that of NNC (≈ 84%), while the Kappa coefficient is 0.87 for VCNNC and 0.75 for the NNC method.     

Some issues in the classification of DAIS hyperspectral data

Classification accuracy depends on a number of factors, of which the nature of the training samples, the number of bands used, the number of classes to be identified relative to the spatial resolution of the image and the properties of the classifier are the most important. This paper evaluates the effects of these factors on classification accuracy using a test area in La Mancha, Spain. High spectral and spatial resolution DAIS data were used to compare the performance of four classification procedures (maximum likelihood, neural network, support vector machines and decision tree). There was no evidence to support the view that classification accuracy inevitably declines as the data dimensionality increases. The support vector machine classifier performed well with all test data sets. The use of the orthogonal MNF transform resulted in a decline in classification accuracy. However, the decision‐tree approach to feature selection worked well. Small increases in classifier accuracy may be obtained using more sophisticated techniques, but it is suggested here that greater attention should be given to the collection of training and test data that represent the range of land surface variability at the spatial scale of the image.     

Land cover classification in rugged areas using simulated moderateresolution remote sensor data and an artificial neural network

Rugged land cover classification accuracies produced by an artificial neural network (ANN) using simulated moderate-resolution remote sensor data exceed overall accuracies produced using the maximum likelihood rule (MLR). Land cover in spatially-complex areas and at broad spatial scales may be difficult to monitor due to ambiguities in spectral reflectance information produced from cloud-related and topographic effects, or from sampling constraints. Such ambiguities may produce inconsistent estimates of changes in vegetation status, surface energy balance, run-off yields, or other land cover characteristics. By use of a 'back-classification' protocol, which uses the same pixels for testing as for training the classifier, tests of ANN versus MLR-based classifiers demonstrated the ANNbased classifier equalled or exceeded classification accuracies produced by the MLR-based classifier in five of six land cover classes evaluated.     

基于伪标签深度学习的高光谱影像半监督分类

研究了一种仅利用少量标记点训练深度卷积神经网络并对高光谱影像进行分类的方法。以图像分割获得的同质区增加训练样本数目；借助这些增加的样本训练初始分类器并预测所有未知点的初始类别；将每一初始类别聚集为适当的类簇，以类簇号作为伪标签对深度卷积网进行预训练；最后利用经过同质区增加的训练样本精调预训练深度卷积网。实验结果证明新方法可以在仅用少量实际标记样本的情况下成功地训练深度卷积网，对高光谱数据进行有效分类。     

Segmentation of remotely-sensed images by a split-and-merge process+

Abstract

This paper describes the application of an image segmentation technique to remotely-sensed terrain images used for environmental monitoring. The segmentation is a preprocessing operation which is applied prior to image classification in order to improve classification accuracy from that achievable by classifying pixels individually on the basis of their spectral signatures. The method uses a split-and-merge technique to segment images into regions of homogeneous tone and texture wherever this is possible. The split-and-merge technique employs a hierarchical quadtree data structure. Texture is measured using easily computed grey value difference statistics. The homogeneity criteria employed in region merging are dependent on local statistics. The segmented image is classified using a region classifier for regions and the normal per-pixel classifier for single pixels in areas of inhomogeneity. The technique is illustrated by example classifications of aerial Multispectral Scanner data from two test sites. A quantitative analysis of the performance shows that an increased classification accuracy is achieved.     

3D卷积自编码器高光谱图像分类模型

目的 高光谱图像分类是遥感领域的基础问题，高光谱图像同时包含丰富的光谱信息和空间信息，传统模型难以充分利用两种信息之间的关联性，而以卷积神经网络为主的有监督深度学习模型需要大量标注数据，但标注数据难度大且成本高。针对现有模型的不足，本文提出了一种无监督范式下的高光谱图像空谱融合方法，建立了3D卷积自编码器（3D convolutional auto-encoder，3D-CAE）高光谱图像分类模型。方法 3D卷积自编码器由编码器、解码器和分类器构成。将高光谱数据预处理后，输入到编码器中进行无监督特征提取，得到一组特征图。编码器的网络结构为3个卷积块构成的3D卷积神经网络，卷积块中加入批归一化技术防止过拟合。解码器为逆向的编码器，将提取到的特征图重构为原始数据，用均方误差函数作为损失函数判断重构误差并使用Adam算法进行参数优化。分类器由3层全连接层组成，用于判别编码器提取到的特征。以3D-CNN （three dimensional convolutional neural network）为自编码器的主干网络可以充分利用高光谱图像的空间信息和光谱信息，做到空谱融合。以端到端的方式对模型进行训练可以省去复杂的特征工程和数据预处理，模型的鲁棒性和稳定性更强。结果 在Indian Pines、Salinas、Pavia University和Botswana等4个数据集上与7种传统单特征方法及深度学习方法进行了比较，本文方法均取得最优结果，总体分类精度分别为0.948 7、0.986 6、0.986 2和0.964 9。对比实验结果表明了空谱融合和无监督学习对于高光谱遥感图像分类的有效性。结论 本文模型充分利用了高光谱图像的光谱特征和空间特征，可以做到无监督特征提取，无需大量标注数据的同时分类精度高，是一种有效的高光谱图像分类方法。     

Spectral reflectance of coral reef benthos and substrate assemblages on Heron Reef,Australia

Studies investigating the spectral reflectance of coral reef benthos and substrates have focused on the measurement of pure endmembers, where the entire field of view (FOV) of a spectrometer is focused on a single benthos or substrate type. At the spatial scales of the current satellite sensors, the heterogeneity of coral reefs even at a sub-metre scale means that many individual image pixels will be made up of a mixture of benthos and substrate types. If pure endmember spectra are used as training data for image classification, there is a spatial discrepancy, because many pixels will have a mixed endmember spectral reflectance signature. This study investigated the spectral reflectance of coral reef benthos and substrates at a spatial scale directly linked to the pixel size of high spatial resolution imaging systems, by incorporating multiple benthos and substrate types into the spectrometer FOV in situ. A total of 334 spectral reflectance signatures were measured of 19 assemblages of the coral reef benthos and substrate types. The spectra were analysed for separability using first derivative values, and a discrimination decision tree was designed to identify the assemblages. Using the decision tree, it was possible to identify 15 assemblages with a mean overall classification accuracy of 62.6%.     

Using Dempster–Shafer to incorporate knowledge into satellite image classification

Remote sensing imaging techniques make use of data derived from high resolution satellite sensors. Image classification identifies and organises pixels of similar spatial distribution or similar statistical characteristics into the same spectral class (theme). Contextual data can be incorporated, or ‘fused’, with spectral data to improve the accuracy of classification algorithms. In this paper we use Dempster–Shafer’s theory of evidence to achieve this data fusion. Incorporating a Knowledge Base of evidence within the classification process represents a new direction for the development of reliable systems for image classification and the interpretation of remotely sensed data.     

高光谱图像小样本分类的卷积神经网络方法

目的 与传统分类方法相比,基于深度学习的高光谱图像分类方法能够提取出高光谱图像更深层次的特征。针对现有深度学习的分类方法网络结构简单、特征提取不够充分的问题,提出一种堆叠像元空间变换信息的数据扩充方法,用于解决训练样本不足的问题,并提出一种基于不同尺度的双通道3维卷积神经网络的高光谱图像分类模型,来提取高光谱图像的本质空谱特征。方法 通过对高光谱图像的每一像元及其邻域像元进行旋转、行列变换等操作,丰富中心像元的潜在空间信息,达到数据集扩充的作用。将扩充之后的像素块输入到不同尺度的双通道3维卷积神经网络学习训练集的深层特征,实现更高精度的分类。结果 5次重复实验后取平均的结果表明,在随机选取了10%训练样本并通过8倍数据扩充的情况下,Indian Pines数据集实现了98.34%的总体分类精度,Pavia University数据集总体分类精度达到99.63%,同时对比了不同算法的运行时间,在保证分类精度的前提下,本文算法的运行时间短于对比算法,保证了分类模型的稳定性、高效性。结论 本文提出的基于双通道卷积神经网络的高光谱图像分类模型,既解决了训练样本不足的问题,又综合了高光谱图像的光谱特征和空间特征,提高了高光谱图像的分类精度。     

谐波能量谱特征向量的高光谱影像Bayes分类

摘要：对于高光谱影像存在高维非线性、数据冗余多、纯训练样本难以提取等不足,本文引入频率域空间的谐波分析（Harmonic Analysis,HA）理论并提出了一种高光谱影像的HA-Bayes监督分类方法。该方法在保持高光谱数据空-谱特性不变的情况下,从光谱维角度分析不同分解层的影像光谱谐波特征,将高光谱影像变换成由谐波能量谱组成的频率域特征矢量信息。通过建立谐波能量谱特征向量的先验知识,实现Bayes准则下谐波能量谱特征矢量信息判别与分类,最终实现高光谱影像分类。将此方法应用到ROSIS高光谱影像分类时获得的分类总体精度达85.5%,Kappa系数也达到了0.812。进一步实验也证明频率域的谐波分析在高光谱遥感影像特征提取与分类方面具有更好的优势和潜力。     

基于SSAE深度学习特征表示的高光谱遥感图像分类方法

针对遥感图像中高光谱数据的分类问题,提出一种基于堆叠稀疏自动编码器(SSAE)深度学习特征表示的高光谱遥感图像分类方法。首先,将光谱数据样本进行预处理和归一化。然后,将其输入到SSAE中进行特征表示学习,并通过网格搜索来获得最优网络参数,以此获得有效的特征表示。最后通过支持向量机(SVM)分类器对输入图像特征进行分类,最终实现遥感图像中像素的分类。在两个标准数据集上的实验结果表明,该方法能够实现准确的高光谱地物分类。     

混合编码差分进化粒子群算法及多示例学习的高光谱影像降维与分类

目的 高光谱遥感影像由于其巨大的波段数直接导致信息的高冗余和数据处理的复杂,这不仅带来庞大的计算量,而且会损害分类精度。因此,在对高光谱影像进行处理、分析之前进行降维变得非常必要。分类作为一种重要的获取信息的手段,现有的基于像素点和图斑对象特征辨识地物种类的方法在强噪声干扰训练样本条件下精度偏低,在对象的基础上,将光谱和空间特征相似的对象合并成比其还要大的集合,再按照各个集合的光谱和空间特征进行分类,则不容易受到噪声等因素的干扰。方法 提出混合编码差分进化粒子群算法的双种群搜索策略进行降维,基于支持向量机的多示例学习算法作为分类方法,构建封装型降维与分类模型。结果 采用AVIRIS影像进行实验,本文算法相比其他相近的分类方法能获得更高的分类精度,达到96.03%,比其他相近方法中最优的像元级的混合编码的分类方法精度高出0.62%。结论 在针对强干扰的训练样本条件下,本文算法在降维过程中充分发挥混合编码差分进化算法的优势,分类中训练样本中的噪声可以看做多示例学习中训练包"歧义性"的特定表现形式,有效提高了分类的精度。     

3D密集全卷积的高光谱地物分类算法研究

针对高光谱遥感图像训练样本较少、光谱维度较高、空间特征与频谱特征存在差异性而导致高光谱地物分类的特征提取不合理、分类精度不稳定和训练时间长等问题,提出了基于3D密集全卷积（3D-DSFCN）的高光谱图像（HSI）分类算法。算法通过密集模块中的3D卷积核分别提取光谱特征和空间特征,采用特征映射模块替换传统网络中的池化层和全连接层,最后通过softmax分类器进行分类。实验结果表明,基于3D-DSFCN的HSI分类方法提高了地物分类的准确率、增强了低频标签的分类稳定性。     

结合注意力和改进样本选取方法的少样本高光谱分类孪生网络

针对高光谱图像(hyperspectral image)样本人工标记困难导致的样本数量不足的问题, 本文提出了一个结合注意力和空间邻域的少样本孪生网络算法. 它首先对高光谱图像进行PCA预处理, 实现数据降维; 其次, 对模型训练样本采用间隔采样和边缘采样的方式进行选取, 以有效减少冗余信息; 之后, Siamese network以大小不同的patch形式进行两两结合, 构建出样本对作为训练集进行训练, 不仅实现了数据增强的效果, 还能在提取光谱信息特征的同时, 充分提取目标像素光谱信息以及其周围邻域空间信息; 最后, 添加光谱维度的注意力模块以及空间维度的相似度度量模块, 分别对光谱信息和空间邻域信息进行权重分布, 以达到提升分类性能的目的. 实验结果表明, 本文提出的方法在部分公开数据集上对比常用方法取得了较好的实验效果.