Hyperspectral satellite data analysis for pure pixels extraction and evaluation of advanced classifier algorithms for LULC classification

The study was carried out for Indian capital city Delhi using Hyperion sensor onboard EO-1 satellite of NASA. After MODTRAN-4 based atmospheric correction, MNF, PPI and n-D visualizer were applied and endmembers of 11 LCLU classes were derived which were employed in classification of LULC. To incur better classification accuracy, a comparative study was also carried out to evaluate the potential of three classifier algorithms namely Random Forest (RF), Support Vector Machines (SVM) and Spectral Angle Mapper (SAM). The results of this study reemphasize the utility of satellite borne hyperspectral data to extract endmembers and also to delineate the potential of random forest as expert classifier to assess land cover with higher classification accuracy that outperformed the SVM by 19% and SAM by 27% in overall accuracy. This research work contributes positively to the issue of land cover classification through exploration of hyperspectral endmembers. The comparison of classification algorithms’ performance is valuable for decision makers to choose better classifier for more accurate information extraction.     

Machine Learning Feature Selection Methods for Landslide Susceptibility Mapping

This paper explores the use of adaptive support vector machines, random forests and AdaBoost for landslide susceptibility mapping in three separated regions of Canton Vaud, Switzerland, based on a set of geological, hydrological and morphological features. The feature selection properties of the three algorithms are studied to analyze the relevance of features in controlling the spatial distribution of landslides. The elimination of irrelevant features gives simpler, lower dimensional models while keeping the classification performance high. An object-based sampling procedure is considered to reduce the spatial autocorrelation of data and to estimate more reliably generalization skills when applying the model to predict the occurrence of new unknown landslides. The accuracy of the models, the relevance of features and the quality of landslide susceptibility maps were found to be high in the regions characterized by shallow landslides and low in the ones with deep-seated landslides. Despite providing similar skill, random forests and AdaBoost were found to be more efficient in performing feature selection than adaptive support vector machines. The results of this study reveal the strengths of the classification algorithms, but evidence: (1) the need for relying on more than one method for the identification of relevant variables; (2) the weakness of the adaptive scaling algorithm when used with landslide data; and (3) the lack of additional features which characterize the spatial distribution of deep-seated landslides.     

塔里木河下游土地利用覆被MISR多角度遥感制图

通过对塔里木河下游MISR卫星多角度观测数据的不同组合构建多角度数据集,探索多角度观测与传统垂直观测对土地利用覆被遥感制图效果的影响,分别使用SVM(支持向量机)与传统的MLC(最大似然分类法)作为分类器,对分类后得到的混淆矩阵进行分析。结论证实:无论是使用传统的MLC还是SVM作为分类器,多角度观测都取得比垂直观测更高的总体分类精度;MISR近红外波段虽然分辨率较低,但依然含有丰富的信息,对地表覆被的分类有重要影响;无论使用哪一数据集,SVM法都能获得更高的分类精度;不同相机对分类结果的影响各不相同,其中C、D相机的作用更重要。     

基于极限学习机的GF-2影像分类

遥感图像分类是提取图像有效信息过程中重要的一部分,为了探寻最优的分类方法,许多机器学习算法逐步应用于遥感分类中。极限学习机（extreme learning machine,ELM）以其高效、快速和良好的泛化性能在模式识别领域得到广泛应用。本文采用训练速度快、运算量小的极限学习机算法与支持向量机（support vector machines,SVM）算法和最大似然法进行分类对比,对高分辨率遥感图像进行分类,分析极限学习机算法对于遥感图像分类的准确度等性能。选取吉林省长春市部分区域的GF-2遥感数据,将融合后的影像设置为原始数据,利用3种方法进行分类。研究结果表明,极限学习机算法分类图像总体分类精度达到85%以上,kappa系数达到0.718,与其他分类方法相比分类准确度较高,且极限学习机运行时间比支持向量机运行时间约短2 480 s,约为支持向量机运行时间的1/8,因此具有良好的性能和实用价值。     

引导滤波联合局部判别嵌入的高光谱影像分类

高光谱遥感影像分类是高光谱遥感影像处理和应用的重要组成部分。然而,高光谱遥感影像具有波段数量较多和空间分辨率较高等特点,给分类任务带来一定的挑战。为了提高分类精度,充分利用影像的空间信息和像素间的局部信息,提出一种引导滤波联合局部判别嵌入的高光谱影像分类方法。首先,对高光谱遥感影像进行归一化,利用主成分分析方法实现特征提取,将提取的第一主成分影像作为引导图像;其次,采用引导滤波分别提取各波段影像的空间特征;然后,将提取的空间影像特征进行叠加,通过局部Fisher判别分析完成低维嵌入;最后,将得到的低维嵌入特征输入支持向量机分类器得到分类结果。采用Indian Pines和Pavia University两幅高光谱影像进行实验的结果表明：在分别从各类地物中随机选取10%和100个样本作为训练样本的情况下,其总体分类精度分别提高到98.28%和99.45%;对比其他相关方法,该方法能够获取更高的分类精度。该方法在低维嵌入的同时,有效利用了影像的空间信息,改善了分类效果。     

Automated Kerogen Classification in Microscope Images of Dispersed Kerogen Preparation

We develop the classification part of a system that analyses transmitted light microscope images of dispersed kerogen preparation. The system automatically extracts kerogen pieces from the image and labels each piece as either inertinite or vitrinite. The image pre-processing analysis consists of background removal, identification of kerogen material, object segmentation, object extraction (individual images of pieces of kerogen) and feature calculation for each object. An expert palynologist was asked to label the objects into categories inertinite and vitrinite, which provided the ground truth for the classification experiment. Ten state-of-the-art classifiers and classifier ensembles were compared: Naïve Bayes, decision tree, nearest neighbour, the logistic classifier, multilayered perceptron (MLP), support vector machines (SVM), AdaBoost, Bagging, LogitBoost and Random Forest. The logistic classifier was singled out as the most accurate classifier, with an accuracy greater than 90. Using a 10 times 10-fold cross-validation provided within the Weka software, we found that the logistic classifier was significantly better than five classifiers (p<0.05) and indistinguishable from the other four classifiers. The initial set of 32 features was subsequently reduced to 6 features without compromising the classification accuracy. A further evaluation of the system alerted us to the possible sensitivity of the classification to the ground truth that might vary from one human expert to another. The analysis also revealed that the logistic classifier made most of the correct classifications with a high certainty.     

四种常用的全球1km土地覆盖数据中国区域的精度评价

精确的全球及区域土地覆盖数据是陆地表层过程研究的重要基础.对数据质量的了解与数据本身同等重要,定量的数据精度评估不仅是对数据本身包含信息的丰富,而且对有助于发现问题,从而促进土地覆盖制图方法的发展.基于一个新的分类系统(森林、灌木草地、农田、裸地、城市、湿地和水体),以中国1∶10万土地利用数据为参考数据,从类型面积一致性、空间一致性两方面对4种全球土地覆盖数据集在中国区域的分类精度进行了评价,包括美国地质调查局为国际地圈-生物圈计划的全球土地覆盖数据集(IGBPDISCover);美国马里兰大学的全球土地覆盖数据集(UMd);欧盟联合研究中心(JRC)空间应用研究所(SAI)的2000年全球土地覆盖数据产品(GLC2000);MODIS 2000年的土地覆盖数据产品(MOD12Q1).并对4种土地覆盖产品误差的空间和类型分布进行了分析.结果表明:在4种土地覆盖分类产品中,GLC2000和MODIS土地覆盖数据有更高的整体分类精度,IGBP数据的整体分类精度次之,但是3种数据在局部都存在明显的分类错误;UMd的分类精度整体最低.通过对4种数据分类精度的空间和类型分布规律的分析,认为4种数据都不能很好的满足陆地表层过程模拟的需要.建议发展土地覆盖类型决策融合方法,将现存多源土地覆盖分类信息融合起来,制备更高精度的中国土地覆盖分类图.     

基于随机森林模型的城市不透水面提取研究 ——以呼和浩特市为例

城市不透水面信息对于城市生态环境动态演化过程研究具有重要意义。以Landsat 8遥感影像为数据源,以呼和浩特市为实证区域,进行了随机森林模型应用于城市不透水面的提取研究,并与目前应用广泛的支持向量机模型进行了对比分析。研究表明：在不同的抽样比例训练样本条件下,随机森林模型对于城市不透水面的提取精度均优于支持向量机的提取精度;对于随机森林模型和支持向量机模型,70%的训练样本比例均为最佳训练样本抽样比例。在该抽样比例下,随机森林模型提取城市不透水面的总体分类精度为93.29%,Kappa系数为0.9051,支持向量机模型的总体分类精度为91.26%,Kappa系数为0.8757;随机森林模型对于城市裸土的识别度较高,能更好地将城市裸土和不透水面进行区分,而支持向量机模型对于城市裸土、不透水面和绿地的区分能力均弱于随机森林模型。综合而言,随机森林模型对城市不透水面的提取精度优于支持向量机模型,随机森林模型可以有效应用于城市不透水面提取领域,进一步丰富了城市不透水面提取方法体系构成。     

Appraisal of land use/land cover of mangrove forest ecosystem using support vector machine

Human activities in many parts of the world have greatly changed the natural land cover. This study has been conducted on Pichavaram forest, south east coast of India, famous for its unique mangrove bio-diversity. The main objectives of this study were focused on monitoring land cover changes particularly for the mangrove forest in the Pichavaram area using multi-temporal Landsat images captured in the 1991, 2000, and 2009. The land use/land cover (LULC) estimation was done by a unique hybrid classification approach consisting of unsupervised and support vector machine (SVM)-based supervised classification. Once the vegetation and non-vegetation classes were separated, training site-based classification technology i.e., SVM-based supervised classification technique was used. The agricultural area, forest/plantation, degraded mangrove and mangrove forest layers were separated from the vegetation layer. Mud flat, sand/beach, swamp, sea water/sea, aquaculture pond, and fallow land were separated from non-vegetation layer. Water logged areas were delineated from the area initially considered under swamp and sea water-drowned areas. In this study, the object-based post-classification comparison method was employed for detecting changes. In order to evaluate the performance, an accuracy assessment was carried out using the randomly stratified sampling method, assuring distribution in a rational pattern so that a specific number of observations were assigned to each category on the classified image. The Kappa accuracy of SVM classified image was highest (94.53 %) for the 2000 image and about 94.14 and 89.45 % for the 2009 and 1991 images, respectively. The results indicated that the increased anthropogenic activities in Pichavaram have caused an irreversible loss of forest vegetation. These findings can be used both as a strategic planning tool to address the broad-scale mangrove ecosystem conservation projects and also as a tactical guide to help managers in designing effective restoration measures.     

An adaptive framework for spectral-spatial classification based on a combination of pixel-based and object-based scenarios

Remotely sensed image analysis using spectral-spatial information plays a key role in modern remote sensing applications. This article presents a new semi-automatic framework for spectral-spatial classification of hyperspectral images. The proposed framework benefits from a combination of pixel-based and object-based classification scenarios in which the main parameters are adaptively tuned. In order to reduce the complexity of the method, an unsupervised band selection technique is used as well. Meanwhile, the wavelet thresholding is applied in order to smooth the selected bands. The classification results after applying the proposed method to well-known standard hyperspectral datasets are better than those of the most of the other state-of-the-art approaches. As an example, the overall classification accuracy achieved by applying the proposed semi-automatic spectral-spatial classification framework to the Salinas dataset is more than 99% for 10% training samples per class. Moreover, the vital parameters are adaptively set in our approach.     

Fine classification of construction land using high-resolution remote sensing images: a case study in planning restricted zone of nuclear power plant

Detailed construction land information plays a significant role in monitoring planning restricted zone of nuclear power plant and ecological environment protection. This study focuses on developing fine classifying method of construction land in planning restricted zone of nuclear power plant using high spatial resolution GF(GaoFen)-1 remote sensing images. The object-oriented classification method is used in this study; the important process of which is image segmentation and classification. Multi-scale segmentation method, rule-based decision tree, and the nearest neighbor classifier are used in classifying construction land classes, i.e., road, industrial, and residential. An optimal segmentation scale is crucial to image segmentation in object-oriented classification. Instead of laborious trial-and-error experiments for optimal image segmentation, the change rates of the local variance in the homogeneous region are calculated to get the optimal segmentation scales. Multi-level classification strategy is used in the following classification. Rule-based decision tree is used to classify road and water, vegetation and non-vegetation, and industrial and residential. And the nearest neighbor classifier is used to classify cropland and forest within the vegetation land use type. The accuracy assessment result shows that the overall accuracy is 89.67% and Kappa coefficient is 0.85 for object-oriented classification, which is much higher than pixel-based maximum likelihood classifier (overall accuracy is 79.17% and Kappa coefficient is 0.74) and support vector machine classifier (overall accuracy is 74.16% and Kappa coefficient is 0.68).     

吉林西部多时相遥感数据分类方案的构建及应用

为了深化遥感监测方法在生态环境调查中的应用,本文以吉林西部为试验区,设计了一种多时相遥感数据分类方案。该方案以物候信息为主,结合地物特征变量(植被、水体和土地信息)构建的多维特征空间数据集用于土地覆被分类。该遥感分类方案提取了9种地表覆被类型,结果表明:地表植被季节变化信息和土地利用信息的引入能明显改善土地覆被的分类精度;与基于原始波段的分类方案相比,多时相遥感数据分类方案的分类精度最好,总体分类精度为95.50%,Kappa系数为95.04%。     

Mediterranean forest mapping using hyper-spectral satellite imagery

Mediterranean forests are characterized by spatiotemporal heterogeneity that is associated with Mediterranean climate, floristic biodiversity and topographic variability. Satellite remote sensing can be an effective tool for characterizing and monitoring forest vegetation distribution within these fragmented Mediterranean landscapes. The heterogeneity of Mediterranean vegetation, however, often exceeds the resolution typical of most satellite sensors. Hyper-spectral remote sensing technology demonstrates the capacity for accurate vegetation identification. The objective of this research is to determine to what extent forest types can be discriminated using different image analysis techniques and spectral band combinations of Hyperion satellite imagery. This research mapped forest types using a pixel-based Spectral Angle Mapper (SAM), nearest neighbour and membership function classifiers of the object-oriented classification. Hyperion classification was done after reducing Hyperion data using nine selected band combinations. Results indicate that the selection of band combination while reducing the Hyperion dataset improves classification results for both the overall and the individual forest type accuracy, in particular for the selected optimum Hyperion band combination. One shortcoming is that the performance of the best selected band combination was superior in terms of both overall and individual forest type accuracy when applying the membership classifier of the object-oriented method compared to SAM and nearest neighbour classifiers. However, all techniques seemed to suffer from a number of problems, such as spectral similarity among forest types, overall low energy response of the Hyperion sensor, Hyperion medium spatial resolution and spatiotemporal and spectral heterogeneity of the Mediterranean ecosystem at multiple scales.     

Relationship in metastasis of coal fire and land use/cover using thermal imagery and support vector machine classifier

The objective of the present study was to delineate temporal and spatial changes in the coal fire and land use/cover within Bastacolla area of Jharia coal field. Studying this variation helped to decipher interconnection among the dynamics of the coal fire and concomitant changes in land use/cover. The detection of coal fires during a span of 14 years along with transitioning land use/cover was cost-effective and enabled planning for management of coal resources and environment. Landsat series of satellite data of 2002, 2009, 2013, and 2016 were processed for generating land surface temperature profiles vis-a-vis classified land use/cover of the study area. A single cut-off temperature was derived for mapping of coal fires using land surface temperature profile from 2002 to 2016. The satellite images were classified using support vector machines, and for depicting land use/cover change, post-classification change detection was done. Classification accuracy obtained was excellent with kappa coefficient ranging from 0.897 for classified image of 2002 to 0.799 for classified image of 2016. Results revealed that coal fires had shifted to the central west part of the area. Furthermore, pockets of coal fire from northern and eastern part of the study area have diminished. OB dumps and coal quarry/coal dump may be attributed towards the spatial change in coal fire while; OB dumps showed connotation with the highest temperature zones. Ground verifications for temperature profiles and coal fires were carried out using thermal camera which enunciated good agreement with results.     

GIS-based evolutionary optimized Gradient Boosted Decision Trees for forest fire susceptibility mapping

Rampant pasture burning has lead to various forest fires taking their toll over the health of many forests. Nanda Devi Biosphere Reserve, located in the northern part of India, witnessed a majority of these incidents in the recent past, though, it remains comprehensively untouched from research studies. The scale of these wildfires has led to an immense requirement of preventive measures to be taken for recuperating from such events. This requires for an in-depth analysis of the study area, its history of wildfires and their causes. These efforts would assist in laying a blueprint for a contingency plan in the event of a wildfire. This work proposes an evolutionary optimized gradient boosted decision trees for preparing wildfire susceptibility maps for the study area that would aid in the government’s forest preservation and disaster management activities. The study took 18 ignition factors of elevation, slope, aspect, plan curvature, topographic position index, topographic water index, normalized difference vegetation index, soil texture, temperature, rainfall, aridity index, potential evapotranspiration, relative humidity, wind speed, land cover and distance from roads, rivers and habitations into consideration. The study revealed that approximately 1432.025 km² of area was very highly susceptible to forest fires while 1202.356 km² was highly susceptible to forest fires. The proposed model was compared against various machine learning models such as random forest, neural networks and support vector machines, and it outperformed them by achieving an overall accuracy of 95.5%. The proposed model demonstrated good prospects for application in the field of hazard susceptibility mappings.     

Mapping flood susceptibility in an arid region of southern Iraq using ensemble machine learning classifiers: a comparative study

This study examined the efficacy of three machine ensemble classifiers, namely, random forest, rotation forest and AdaBoost, in assessing flood susceptibility in an arid region of southern Iraq. A dataset was created from flooded and non-flooded areas to train and validate the ensemble classifiers using a binary classification scheme (1—flood, 0—non-flood). The prepared dataset was then partitioned into two sets with a 70/30 ratio: 70% (2478 pixels) for training and 30% (1062 pixels) for testing. A total of 10 influential flood factors were selected and prepared based on data availability and a literature review. The selected factors were surface elevation, slope, plain curvature, topographic wetness index, stream power index, distance to rivers, drainage density, lithology, soil and land use/land cover. The information gain ratio was first utilised to explore the predictive abilities of the factors. The predictive performances of the three ensemble models were compared using six statistical measures: sensitivity, specificity, accuracy, kappa, root mean square error and area under the operating characteristics curve. The results revealed that the AdaBoost classifier was the best in terms of the statistical measures, followed by the random forest and rotation forest models. A flood susceptibility map was prepared based on the result of each classifier and classified into five zones: very low, low, moderate, high and very high. For the model with the best performance, i.e., the AdaBoost model, these zones were distributed over an area of 6002 km² (44%) for the very low–low zone, 2477 km² (18%) for the moderate zone and 5048 km² (40%) for the high–very high zones. This study proved the high capabilities of ensemble machine learning classifiers to decipher flood susceptibility zones in an arid region.     

植被覆盖度遥感估算研究进展

植被覆盖度是刻画地表植被覆盖的重要参数,在全球变化研究、地表过程模拟和水文生态模型中发挥着重要作用.遥感能够反映不同空间尺度的植被覆盖信息及其变化趋势,是获取区域及全球植被覆盖度参数的一个重要手段.综合分析了用于植被覆盖度估算的遥感数据源,包括高光谱数据、多光谱数据、微波数据和激光雷达数据.而且分析了各种常用的植被覆盖度遥感估算方法及其优缺点,并评价了现有基于遥感数据的植被覆盖度产品及存在问题.最后,针对目前研究中存在的问题,讨论了植被覆盖度遥感估算研究的发展趋势,指出高时空分辨率长时间序列的全球植被覆盖度数据集、多源遥感数据融合和同化技术是未来植被覆盖度遥感估算研究的主要方向.     

Evaluating the impact of classification algorithms and spatial resolution on the accuracy of land cover mapping in a mountain environment in Pakistan

Satellite images of various spatial resolutions and different image classification techniques have been utilized for land cover (LC) mapping at local and regional scale studies. Mapping capabilities and achievable accuracies of LC classification in a mountain environment are, however, influenced by the spatial resolution of the utilized images and applied classification techniques. Hence, developing and characterizing regionally optimized methods are essential for the planning and monitoring of natural resources. In this study, the potential of four non-parametric image classification techniques, i.e., k-nearest neighbor (k-NN), support vector machine (SVM), random forest (RF), and neural network (NN) on the accuracy of LC classification was evaluated in the Hindu Kush mountains ranges of northern Pakistan. Moreover, we have assessed the impact of the spatial resolution of the utilized satellite imagery, i.e., SPOT-5 with 2.5 m and Landsat-8 with 30 m on the accuracy of the derived LC classification. For the classification of LC based on SPOT-5, we have achieved the highest overall classification accuracy (OCA) = 89% with kappa coefficient (KC = 0.86) using SVM followed by k-NN, RF, and NN. However, for LC classification derived from Landsat-8 imagery, we achieved the highest OCA = 71% with KC = 0.59 using RF and SVM followed by k-NN and NN. The higher accuracy derived from SPOT-5 versus Landsat-8 indicated that the results of LC classification based on SPOT-5 are more accurate and reliable than Landsat-8. The findings of the present study will be useful for the classification and mapping task of LC in a mountain environment using SPOT-5 and Landsat-8 at local and regional scale studies.     

基于ETM影像数据融合的土地覆盖分类研究

在ERDAS软件支持下，对ETM遥感影像数据的TM1-TM5，TM7与其全色波段TM8进行融合，采用主成分、乘积法、Brovey转换三种融合方法，重采样方法分别为邻域法、立方卷积法及双线性内插法。采用相同的训练样本区及最大似然法分类方法，对融合产生的9幅影像及未融合影像进行土地覆盖分类，通过对分类影像的Producers Accuracy，Users Accuracy，Kappa三者的精度数据和地物波谱信息的对比分析，在总体上，上述的影像融合方法对提高土地覆盖分类的精度不明显，但就某些地物类型来说，还是值得采用的；三种融合方法和三种重采样方式它们之问相比较而言，乘积法融合法和立方卷积重采样法相对较为可取。     

基于遥感数据的全球及区域土地覆盖制图--现状、战略和趋势

土地利用／土地覆盖变化研究是近年来全球变化研究的焦点之一。全球和区域尺度的土地覆盖特征对全球环境状况的评估、模拟未来全球环境的情景有重要的作用。2000年在Internat ionalJournalofRemoteSensing杂志上出版了题为"GlobalandRegionalLandCoverCharacterizat ion from Remotely Sensed Data"的专辑。在此基础上,介绍、总结了国际上利用遥感影像进行全球和区域等大尺度土地覆盖研究的新进展。分别从数据源与制图的时空尺度、制图方法（数据预处理、分类、精度评估）等方面进行了介绍,并对现今的两个全球土地覆盖数据库进行了比较分析。