Comparing supervised and unsupervised multiresolution segmentation approaches for extracting buildings from very high resolution imagery

Although multiresolution segmentation (MRS) is a powerful technique for dealing with very high resolution imagery, some of the image objects that it generates do not match the geometries of the target objects, which reduces the classification accuracy. MRS can, however, be guided to produce results that approach the desired object geometry using either supervised or unsupervised approaches. Although some studies have suggested that a supervised approach is preferable, there has been no comparative evaluation of these two approaches. Therefore, in this study, we have compared supervised and unsupervised approaches to MRS. One supervised and two unsupervised segmentation methods were tested on three areas using QuickBird and WorldView-2 satellite imagery. The results were assessed using both segmentation evaluation methods and an accuracy assessment of the resulting building classifications. Thus, differences in the geometries of the image objects and in the potential to achieve satisfactory thematic accuracies were evaluated. The two approaches yielded remarkably similar classification results, with overall accuracies ranging from 82% to 86%. The performance of one of the unsupervised methods was unexpectedly similar to that of the supervised method; they identified almost identical scale parameters as being optimal for segmenting buildings, resulting in very similar geometries for the resulting image objects. The second unsupervised method produced very different image objects from the supervised method, but their classification accuracies were still very similar. The latter result was unexpected because, contrary to previously published findings, it suggests a high degree of independence between the segmentation results and classification accuracy. The results of this study have two important implications. The first is that object-based image analysis can be automated without sacrificing classification accuracy, and the second is that the previously accepted idea that classification is dependent on segmentation is challenged by our unexpected results, casting doubt on the value of pursuing ‘optimal segmentation’. Our results rather suggest that as long as under-segmentation remains at acceptable levels, imperfections in segmentation can be ruled out, so that a high level of classification accuracy can still be achieved.     

Comparing object-based and pixel-based classifications for mapping savannas

The development of robust object-based classification methods suitable for medium to high resolution satellite imagery provides a valid alternative to ‘traditional’ pixel-based methods. This paper compares the results of an object-based classification to a supervised per-pixel classification for mapping land cover in the tropical north of the Northern Territory of Australia. The object-based approach involved segmentation of image data into objects at multiple scale levels. Objects were assigned classes using training objects and the Nearest Neighbour supervised and fuzzy classification algorithm. The supervised pixel-based classification involved the selection of training areas and a classification using the maximum likelihood classifier algorithm. Site-specific accuracy assessment using confusion matrices of both classifications were undertaken based on 256 reference sites. A comparison of the results shows a statistically significant higher overall accuracy of the object-based classification over the pixel-based classification. The incorporation of a digital elevation model (DEM) layer and associated class rules into the object-based classification produced slightly higher accuracies overall and for certain classes; however this was not statistically significant over the object-based using spectral information solely. The results indicate object-based analysis has good potential for extracting land cover information from satellite imagery captured over spatially heterogeneous land covers of tropical Australia.     

典型遥感数据分类方法的适用性分析——以遥感图像场景分类为例

目前普遍采用的分类器通常都是针对单一或小量任务而设计的,在小数据量的处理中能取得比较满意的结果。但对于海量遥感数据的处理,其在处理时效和分类精度方面还有待研究。本文以遥感图像场景分类任务为例,着重对遥感数据分类问题中几种典型分类方法的适用性进行比较研究,包括K近邻(KNN)、随机森林(RF),支持向量机(SVM)和稀疏表达分类器(SRC)等。分别从参数敏感性,训练样本数据量,待分类样本数据量和样本特征维数对分类器性能的影响等几个方面进行比较分析。实验结果表明:(1)KNN,RF和L0-SRC方法相比RBF-SVM,Linear-SVM和L1-SRC,受参数影响的程度更弱;(2)待分类样本固定的情况下,随着训练样本数目的增加,SRC类型分类方法的分类性能最佳,SVM类型方法次之,然后是RF和KNN,在总体分类时间上呈现出L0-SRCL1-SRCRFRBF-SVM/Linear-SVMKNN/L0-SRC-Batch的趋势;(3)训练样本固定的情况下,所有分类方法的分类精度几乎都不受待分类样本数目变化的影响,RBF-SVM方法性能最佳,其次是L1-SRC,然后是Linear-SVM,最后是RF和L0-SRC/L0-SRC-Batch,在总体分类时间上,L1-SRC和L0-SRC相比其他分类方法最为耗时;(4)样本特征维数的变化不仅影响分类器的运行效率,同时也影响其分类精度,其中SRC和KNN分类器器无需较高的特征维数即可获得较好的分类结果,SVM对高维特征具有较强的包容性和学习能力,RF分类器对特征维数增加则表现得并不敏感,特征维数的增加并不能对其分类精度的提升带来更多的贡献。总的来说,在大数据量的遥感数据分类任务中,现有分类方法具有良好的适用性,但是对于分类器的选择应当基于各自的特点和优势,结合实际应用的特点进行权衡和选择,选择参数敏感性较小,分类总体时间消耗低但分类精度相对较高的分类方法。     

Learning with transductive SVM for semisupervised pixel classification of remote sensing imagery

Land cover classification using remotely sensed data requires robust classification methods for the accurate mapping of complex land cover area of different categories. In this regard, support vector machines (SVMs) have recently received increasing attention. However, small number of training samples remains a bottleneck to design suitable supervised classifiers. On the other hand, adequate number of unlabeled data is available in remote sensing images which can be employed as additional source of information about margins. To fully leverage all of the precious unlabeled data, integration of filtering in a transductive SVM is proposed.Using two labeled image datasets of small size and two large unlabeled image datasets, the effectiveness of the proposed method is explored. Experimental results show that the proposed technique achieves average overall accuracies of around 4.5–7.8%, 0.8–2.6% and 0.9–2.2% more than the standard inductive SVM (ISVM), progressive transductive SVM (PTSVM) and low density separation (LDS) classifiers, respectively on larger domains in case of labeled datasets. Using image datasets, visual interpretation from the classified images as well as the segmentation quality reveal that the proposed method can efficiently filter informative data from the unlabeled samples.     

Segmentation-based classification of hyperspectral imagery using projected and correlation clustering techniques

A partitional clustering-based segmentation is used to carry out supervised classification for hyperspectral images. The main contribution of this study lies in the use of projected and correlation partitional clustering techniques to perform image segmentation. These types of clustering techniques have the capability to concurrently perform clustering and feature/band reduction, and are also able to identify different sets of relevant features for different clusters. Using these clustering techniques segmentation map is obtained, which is combined with the pixel-level support vector machines (SVM) classification result, using majority voting. Experiments are conducted over two hyperspectral images. Combination of pixel-level classification result with the segmentation maps leads to significant improvement of accuracies in both the images. Additionally, it is also observed that, classified maps obtained using SVM combined with projected and correlation clustering techniques results in higher accuracies as compared to classified maps obtained from SVM combined with other partitional clustering techniques.     

GF-2影像面向对象典型城区地物提取方法

国产高分遥感影像信息丰富,提供了精准的地物空间细节,深入研究高分数据处理及其提取城区地类目标信息的方法具有重要意义。本文以国产高分二号（GF-2）遥感影像为数据源,利用规则集的面向对象分类方法,通过ESP尺度分析工具选取得出最优分割尺度,建立各类地物的特征体系及分类规则,最终提取出研究区典型城区地物信息,并将之与传统基于像元的SVM监督分类结果作比较。结果表明：规则集的面向对象分类总体精度为92.23%,Kappa系数为0.9,比SVM监督分类有大幅度提高。对高分二号等高分辨率影像,面向对象的分类方法精度更高,图示效果更好,是城区地物提取的有效方法。     

Evaluation of different machine learning methods for land cover mapping of a Mediterranean area using multi-seasonal Landsat images and Digital Terrain Models

Land cover monitoring using digital Earth data requires robust classification methods that allow the accurate mapping of complex land cover categories. This paper discusses the crucial issues related to the application of different up-to-date machine learning classifiers: classification trees (CT), artificial neural networks (ANN), support vector machines (SVM) and random forest (RF). The analysis of the statistical significance of the differences between the performance of these algorithms, as well as sensitivity to data set size reduction and noise were also analysed. Landsat-5 Thematic Mapper data captured in European spring and summer were used with auxiliary variables derived from a digital terrain model to classify 14 different land cover categories in south Spain. Overall, statistically similar accuracies of over 91% were obtained for ANN, SVM and RF. However, the findings of this study show differences in the accuracy of the classifiers, being RF the most accurate classifier with a very simple parameterization. SVM, followed by RF, was the most robust classifier to noise and data reduction. Significant differences in their performances were only reached for thresholds of noise and data reduction greater than 20% (noise, SVM) and 25% (noise, RF), and 80% (reduction, SVM) and 50% (reduction, RF), respectively.     

Effects of Training Samples and Classifiers on Classification of Landsat-8 Imagery

In this study, we used Landsat-8 imagery to test object- and pixel-based image classification approaches in an urban fringe area. For object-based classification, we applied four machine learning classifiers: decision tree (DT), naive Bayes (NB), random trees (RT), and support vector machine (SVM). For pixel-based classification, we utilized the maximum likelihood classifier (MLC). Specifically, we explored the influence of repeated sampling on classification results with different training sample sizes. We found that (1) except the overall accuracy of NB, those of the other four classifiers increased as the training sample size increased; (2) repeated sampling had a significant effect on classification accuracy, especially for the DT and NB classifiers; and (3) SVM achieved the best classification accuracy. In addition, the performance of the object-based classifiers was superior to that of the pixel-based classifier. The results of this study can provide guidance on the training sample size and classifier selection.     

Water-Body types identification in urban areas from radarsat-2 fully polarimetric SAR data

This paper presents a novel method for supervised water-body extraction and water-body types identification from Radarsat-2 fully polarimetric (FP) synthetic aperture radar (SAR) data in complex urban areas. First, supervised water-body extraction using the Wishart classifier is performed, and the false alarms that are formed in built-up areas are removed using morphological processing methods and spatial contextual information. Then, the support vector machine (SVM), the classification and regression tree (CART), TreeBagger (TB), and random forest (RF) classifiers are introduced for water-body types (rivers, lakes, ponds) identification. In SAR images, certain other objects that are misclassified as water are also considered in water-body types identification. Several shape and polarimetric features of each candidate water-body are used for identification. Radarsat-2 PolSAR data that were acquired over Suzhou city and Dongguan city in China are used to validate the effectiveness of the proposed method, and the experimental results are evaluated at both the object and pixel levels. We compared the water-body types classification results using only shape features and the combination of shape and polarimetric features, the experimental results show that the polarimetric features can eliminate the misclassifications from certain other objects like roads to water areas, and the increasement of classification accuracy embodies at both the object and pixel levels. The experimental results show that the proposed methods can achieve satisfactory accuracies at the object level [89.4% (Suzhou), 95.53% (Dongguan)] and the pixel level [96.22% (Suzhou), 97.95% (Dongguan)] for water-body types classification, respectively.     

高光谱影像的引导滤波多尺度特征提取

为了解决高光谱遥感影像分类中单一尺度特征无法有效表达地物类间差异和区分地物边界的不足,提高影像分类精度和改善分类目视解译效果,提出了采用引导滤波提取多尺度的空间特征的方法。首先,利用主成分分析对高光谱影像进行降维,移除噪声并突出主要特征;然后,将第1主成分作为引导影像,将包含信息量最多的若干主成分分别作为输入影像,应用依次增加的滤波半径分别进行引导滤波处理提取多个尺度的特征,获得影像不同尺度的结构信息;最后,将多尺度特征输入分类器中进行影像监督分类。采用仿真数据和帕维亚大学(Pavia University)、帕维亚城区(Pavia Centre)等3幅高光谱实验数据,提取了基于引导滤波的多尺度特征、多尺度形态特征和多尺度纹理特征,输入到支持向量机、随机森林和K近邻分类器中,进行了实验。实验结果表明:采用支持向量机分类Pavia University数据,相对于采用多尺度形态特征的分类结果,引导滤波特征的总体精度提高了6.5%;Pavia Centre和Salinas两幅影像最高分类精度均由引导滤波特征实现,分别达到98.51%和98.39%。实验证实基于引导滤波提取的多尺度特征能有效地描述地物结构,进而获得更高的分类精度和改善目视解译效果。     

Enhanced land use/cover classification of heterogeneous tropical landscapes using support vector machines and textural homogeneity

Land use/cover classification is a key research field in remote sensing and land change science as thematic maps derived from remotely sensed data have become the basis for analyzing many socio-ecological issues. However, land use/cover classification remains a difficult task and it is especially challenging in heterogeneous tropical landscapes where nonetheless such maps are of great importance. The present study aims at establishing an efficient classification approach to accurately map all broad land use/cover classes in a large, heterogeneous tropical area, as a basis for further studies (e.g., land use/cover change, deforestation and forest degradation). Specifically, we first compare the performance of parametric (maximum likelihood), non-parametric (k-nearest neighbor and four different support vector machines – SVM), and hybrid (unsupervised–supervised) classifiers, using hard and soft (fuzzy) accuracy assessments. We then assess, using the maximum likelihood algorithm, what textural indices from the gray-level co-occurrence matrix lead to greater classification improvements at the spatial resolution of Landsat imagery (30 m), and rank them accordingly. Finally, we use the textural index that provides the most accurate classification results to evaluate whether its usefulness varies significantly with the classifier used. We classified imagery corresponding to dry and wet seasons and found that SVM classifiers outperformed all the rest. We also found that the use of some textural indices, but particularly homogeneity and entropy, can significantly improve classifications. We focused on the use of the homogeneity index, which has so far been neglected in land use/cover classification efforts, and found that this index along with reflectance bands significantly increased the overall accuracy of all the classifiers, but particularly of SVM. We observed that improvements in producer's and user's accuracies through the inclusion of homogeneity were different depending on land use/cover classes. Early-growth/degraded forests, pastures, grasslands and savanna were the classes most improved, especially with the SVM radial basis function and SVM sigmoid classifiers, though with both classifiers all land use/cover classes were mapped with producer's and user's accuracies of ∼90%. Our classification approach seems very well suited to accurately map land use/cover of heterogeneous landscapes, thus having great potential to contribute to climate change mitigation schemes, conservation initiatives, and the design of management plans and rural development policies.     

An effective ensemble classification framework using random forests and a correlation based feature selection technique

Accurate classification of heterogeneous land surfaces with homogeneous land cover classes is a challenging task as satellite images are characterized by a large number of features in the spectral and spatial domains. The identifying relevance of a feature or feature set is an important task for designing an effective classification scheme. Here, an ensemble of random forests (RF) classifiers is realized on the basis of relevance of features. Correlation‐based Feature Selection (CFS) was utilized to assess the relevance of a subset of features by studying the individual predictive ability of each feature along with the degree of redundancy between them. Predictability of RF was greatly improved by random selection of the relevant features in each of the splits. An investigation was carried out on different types of images from the Landsat Enhanced Thematic Mapper Plus (Landsat ETM+) and QuickBird sensors. It has been observed that the performance of the RF classifier was significantly improved while using the optimal set of relevant features compared with a few of the most advanced supervised classifiers such as maximum likelihood classifier (MLC), Navie Bayes, multi‐layer perception (MLP), support vector machine (SVM) and bagging.     

基于误差分析的组合分类器研究

提出了一种基于误差分析的组合分类器,通过结合两种监督分类方法,提出的算法分别估计了两种监督分类方法在计算过程中的误差,给出了规则输出的置信区间,再根据置信区间的大小对两种分类方法的输出结果进行加权平均,从而得到更精确的规则输出.利用该方法对遥感图像进行分类实验,在不同训练样本分布与不同训练样本数量的情况下,比较新的组合分类器与单一分类器的精度.结果表明新的组合分类器能够取得比单一的分类器更高的分类精度.结果还显示出,两个分类器的独立性越强,组合分类器的效果越好.另外一个实验比较了新的组合分类器与和式规则组合分类器的分类精度,结果仍显示出了新方法的优越性.     

Optimizing multi-resolution segmentation scale using empirical methods: Exploring the sensitivity of the supervised discrepancy measure Euclidean distance 2 (ED2)

Multiresolution segmentation (MRS) has proven to be one of the most successful image segmentation algorithms in the geographic object-based image analysis (GEOBIA) framework. This algorithm is relatively complex and user-dependent; scale, shape, and compactness are the main parameters available to users for controlling the algorithm. Plurality of segmentation results is common because each parameter may take a range of values within its parameter space or different combinations of values among parameters. Finding optimal parameter values through a trial-and-error process is commonly practiced at the expense of time and labor, thus, several alternative supervised and unsupervised methods for supervised automatic parameter setting have been proposed and tested. In the case of supervised empirical assessments, discrepancy measures are employed for computing measures of dissimilarity between a reference polygon and an image object candidate. Evidently the reliability of the optimal-parameter prediction heavily relies on the sensitivity of the segmentation quality metric. The idea behind pursuing optimal parameter setting is that, for instance, a given scale setting provides image object candidates different from the other scale setting; thus, by design the supervised quality metric should capture this difference. In this exploratory study, we selected the Euclidean distance 2 (ED2) metric, a recently proposed supervised metric, whose main design goal is to optimize the geometrical discrepancy (potential segmentation error (PSE)) and arithmetic discrepancy between image objects and reference polygons (number-of segmentation ratio (NSR)) in two dimensional Euclidean space, as a candidate to investigate the validity and efficacy of empirical discrepancy measures for finding the optimal scale parameter setting of the MRS algorithm. We chose test image scenes from four different space-borne sensors with varying spatial resolutions and scene contents and systematically segmented them using the MRS algorithm at a series of parameter settings. The discriminative capacity of the ED2 metric across different scales groups was tested using non-parametric statistical methods. Our results showed that the ED2 metric significantly discriminates the quality of image object candidates at smaller scale values but it loses the sensitivity at larger scale values. This questions the meaningfulness of the ED2 metric in the MRS algorithm’s parameter optimization. Our contention is that the ED2 metric provides some notion of the optimal scale parameter at the expense of time. In this respect, especially in operational-level image processing, it is worth to re-think the trade-off between execution time of the processor-intensive MRS algorithm at series of parameter settings targeting a less-sensitive quality metric and an expert-lead trial-and-error approach.     

TanDEM-X multiparametric data features in sea ice classification over the Baltic sea

In this study, we assess the potential of X-band Interferometric Synthetic Aperture Radar imagery for automated classification of sea ice over the Baltic Sea. A bistatic SAR scene acquired by the TanDEM-X mission over the Bothnian Bay in March of 2012 was used in the analysis. Backscatter intensity, interferometric coherence magnitude, and interferometric phase have been used as informative features in several classification experiments. Various combinations of classification features were evaluated using Maximum likelihood (ML), Random Forests (RF) and Support Vector Machine (SVM) classifiers to achieve the best possible discrimination between open water and several sea ice types (undeformed ice, ridged ice, moderately deformed ice, brash ice, thick level ice, and new ice). Adding interferometric phase and coherence-magnitude to backscatter-intensity resulted in improved overall classification per- formance compared to using only backscatter-intensity. The RF algorithm appeared to be slightly superior to SVM and ML due to higher overall accuracies, however, at the expense of somewhat longer processing time. The best overall accuracy (OA) for three methodologies were achieved using combination of all tested features were 71.56, 72.93, and 72.91％ for ML, RF and SVM classifiers, respectively. Compared to OAs of 62.28, 66.51, and 63.05％ using only backscatter intensity, this indicates strong benefit of SAR interferometry in discriminating different types of sea ice. In contrast to several earlier studies, we were particularly able to successfully discriminate open water and new ice classes.     

Object-based image analysis of suburban landscapes using Landsat-8 imagery

Many experiments of object-based image analysis have been conducted in remote sensing classification. However, they commonly used high-resolution imagery and rarely focused on suburban area. In this research, with the Landsat-8 imagery, classification of a suburban area via the object-based approach is achieved using four classifiers, including decision tree (DT), support vector machine (SVM), random trees (RT), and naive Bayes (NB). We performed feature selection at different sizes of segmentation scale and evaluated the effects of segmentation and tuning parameters within each classifier on classification accuracy. The results showed that the influence of shape on overall accuracy was greater than that of compactness, and a relatively low value of shape should be set with increasing scale size. For DT, the optimal maximum depth usually varied from 5 to 8. For SVM, the optimal gamma was less than or equal to 10^?2, and its optimal C was greater than or equal to 10². For RT, the optimal active variables was less than or equal to 4, and the optimal maximum tree number was greater than or equal to 30. Furthermore, although there was no statistically significant difference between some classification results produced using different classifiers, SVM has a slightly better performance.     

Comparative performance of convolutional neural network,weighted and conventional support vector machine and random forest for classifying tree species using hyperspectral and photogrammetric data

ABSTRACT

The classification of tree species can significantly benefit from high spatial and spectral information acquired by unmanned aerial vehicles (UAVs) associated with advanced classification methods. This study investigated the following topics concerning the classification of 16 tree species in two subtropical forest fragments of Southern Brazil: i) the potential integration of UAV-borne hyperspectral images with 3D information derived from their photogrammetric point cloud (PPC); ii) the performance of two machine learning methods (support vector machine – SVM and random forest – RF) when employing different datasets at a pixel and individual tree crown (ITC) levels; iii) the potential of two methods for dealing with the imbalanced sample set problem: a new weighted SVM (wSVM) approach, which attributes different weights to each sample and class, and a deep learning classifier (convolutional neural network – CNN), associated with a previous step to balance the sample set; and finally, iv) the potential of this last classifier for tree species classification as compared to the above mentioned machine learning methods. Results showed that the inclusion of the PPC features to the hyperspectral data provided a great accuracy increase in tree species classification results when conventional machine learning methods were applied, between 13 and 17% depending on the classifier and the study area characteristics. When using the PPC features and the canopy height model (CHM), associated with the majority vote (MV) rule, the SVM, wSVM and RF classifiers reached accuracies similar to the CNN, which outperformed these classifiers for both areas when considering the pixel-based classifications (overall accuracy of 84.4% in Area 1, and 74.95% in Area 2). The CNN was between 22% and 26% more accurate than the SVM and RF when only the hyperspectral bands were employed. The wSVM provided a slight increase in accuracy not only for some lesser represented classes, but also some major classes in Area 2. While conventional machine learning methods are faster, they demonstrated to be less stable to changes in datasets, depending on prior segmentation and hand-engineered features to reach similar accuracies to those attained by the CNN. To date, CNNs have been barely explored for the classification of tree species, and CNN-based classifications in the literature have not dealt with hyperspectral data specifically focusing on tropical environments. This paper thus presents innovative strategies for classifying tree species in subtropical forest areas at a refined legend level, integrating UAV-borne 2D hyperspectral and 3D photogrammetric data and relying on both deep and conventional machine learning approaches.     

Evaluation of data fusion and image segmentation in earth observation based rapid mapping workflows

This paper is an exploratory study, which aimed to discover the synergies of data fusion and image segmentation in the context of EO-based rapid mapping workflows. Our approach pillared on the geographic object-based image analysis (GEOBIA) focusing on multiscale, internally-displaced persons’ (IDP) camp information extraction from very high spatial resolution (VHSR) images. We applied twelve pansharpening algorithms to two subsets of a GeoEye-1 image scene that was taken over a former war-induced ephemeral settlement in Sri Lanka. A multidimensional assessment was employed to benchmark pansharpening algorithms with respect to their spectral and spatial fidelity. The multiresolution segmentation (MRS) algorithm of the eCognition Developer software served as the key algorithm in the segmentation process. The first study site was used for comparing segmentation results produced from the twelve fused products at a series of scale, shape, and compactness settings of the MRS algorithm. The segmentation quality and optimum parameter settings of the MRS algorithm were estimated by using empirical discrepancy measures. Non-parametric statistical tests were used to compare the quality of image object candidates, which were derived from the twelve pansharpened products. A wall-to-wall classification was performed based on a support vector machine (SVM) classifier to classify image objects candidates of the fused images. The second site simulated a more realistic crisis information extraction scenario where the domain expertise is crucial in segmentation and classification. We compared segmentation and classification results of the original images (non-fused) and twelve fused images to understand the efficacy of data fusion. We have shown that the GEOBIA has the ability to create meaningful image objects during the segmentation process by compensating the fused image’s spectral distortions with the high-frequency information content that has been injected during fusion. Our findings further questioned the necessity of the data fusion step in rapid mapping context. Bypassing time-intensive data fusion helps to actuate EO-based rapid mapping workflows. We, however, emphasize the fact that data fusion is not limited to VHSR image data but expands over many different combinations of multi-date, multi-sensor EO-data. Thus, further research is needed to understand the synergies of data fusion and image segmentation with respect to multi-date, multi-sensor fusion scenarios and extrapolate our findings to other remote sensing application domains beyond EO-based crisis information retrieval.     

结合多分类器的遥感数据专题分类方法研究

采用标准的多分类器结合方法进行遥感图像的分类研究。首先介绍了标准的多分类器结合的算法,然后以Landsat-TM多光谱遥感数据的土地覆被分类为例,分别给出了抽象级上相同训练特征的多分类器结合、抽象级上不同训练特征的多分类器结合和测量级上的多分类器结合进行土地覆被分类的方法,并进行了实例研究。参与分类器结合的单个分类器包括最大似然分类器,最小距离分类器,马氏距离分类器,K-NN分类器,多层感知器神经网络分类器。分类器的分类精度用总体精度、用户精度、生产者精度、kappa系数和条件kappa系数评价。结果表明,每一种多分类器结合的分类方法都能够比较显著地提高总体分类精度。文章最后对不同多分类器结合方式的优缺点进行了分析。     

An intelligent geospatial processing unit for image classification based on geographic vector agents (GVAs)

Spatial modeling methods usually use pixels and image objects as fundamental processing units to address real‐world objects, geo‐objects, in image space. To do this, both pixel‐based and object‐based approaches typically employ a linear two‐staged workflow of segmentation and classification. Pixel‐based methods segment a classified image to address geo‐objects in image space. In contrast, object‐based approaches classify a segmented image to identify geo‐objects from raster datasets. These methods lack the ability to simultaneously integrate the geometry and theme of geo‐objects in image space. This article explores Geographical Vector Agents (GVAs) as an automated and intelligent processing unit to directly address real‐world objects in the process of remote sensing image classification. The GVA is a distinct type of geographic automata characterized by elastic geometry, dynamic internal structure, neighborhoods and their respective rules. We test this concept by modeling a set of objects on a subset IKONOS image and LiDAR DSM datasets without the setting parameters (e.g. scale, shape information), usually applied in conventional Geographic Object‐Based Image Analysis (GEOBIA) approaches. The results show that the GVA approach achieves more than 3.5% improvement for correctness, 2% improvement for quality, although no significant improvement for completeness to GEOBIA, thus demonstrating the competitive performance of GVAs classification.