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.     

A comparative assessment between object and pixel-based classification approaches for land use/land cover mapping using SPOT 5 imagery

Abstract

Land use/land cover (LULC) classification with high accuracy is necessary, especially in eco-environment research, urban planning, vegetation condition study and soil management. Over the last decade a number of classification algorithms have been developed for the analysis of remotely sensed data. The most notable algorithms are the object-oriented K-Nearest Neighbour (K-NN), Support Vector Machines (SVMs) and the Decision Trees (DTs) amongst many others. In this study, LULC types of Selangor area were analyzed on the basis of the classification results acquired using the pixel-based and object-based image analysis approaches. SPOT 5 satellite images with four spectral bands from 2003 and 2010 were used to carry out the image classification and ground truth data were collected from Google Earth and field trips. In pixel-based image analysis, a supervised classification was performed using the DT classifier. On the other hand, object-oriented (K-NN) image analysis was evaluated using standard nearest neighbour as classifier. Subsequently SVM object-based classification was performed. Five LULC categories were extracted and the results were compared between them. The overall classification accuracies for 2003 and 2010 showed that the object-oriented (K-NN) (90.5% and 91%) performed better results than the pixel-based DT (68.6% and 68.4%) and object-based SVM (80.6% and 78.15%). In general, the object-oriented (K-NN) performed better than both DTs and SVMs. The obtained LULC classification maps can be used to improve various applications such as change detection, urban design, environmental management and zooning.     

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.     

Fusion of RADARSAT-2 and multispectral optical remote sensing data for LULC extraction in a tropical agricultural area

In this study, we investigated the performance of different fusion and classification techniques for land cover mapping in Hilir Perak, Peninsula Malaysia using RADAR and Landsat-8 images in a predominantly agricultural area. The fusion methods used are Brovey Transform, Wavelet Transform, Ehlers and Layer Stacking and their results classified into seven different land cover classes which include (1) pixel-based classifiers (spectral angle mapper (SAM), maximum likelihood (ML), support vector machine (SVM)) and (2) Object-based (rule-based and standard nearest neighbour (NN)) classifiers. The result shows that pixel-based classification achieved maximum accuracy of the optical data classification using SVM in Landsat-8 with 74.96% accuracy compared to SAM and ML. For multisource data classification, the highest overall accuracy recorded for layer stacking (SVM) was 79.78%, Ehlers fusion (SVM) with 45.57%, Brovey fusion (SVM) with 63.70% and Wavelet fusion (SVM) 61.16%. And for object-based classifiers, the overall classification accuracy is 95.35% for rule-based and 76.33% for NN classifier, respectively. Based on the analysis of their performances, object-based and the rule-based classifiers produced the best classification accuracy from the fused images.     

ALOS PALSAR and Hyperion Data Fusion for Land Use Land Cover Feature Extraction

Detailed and enhanced land use land cover (LULC) feature extraction is possible by merging the information extracted from two different sensors of different capability. In this study different pixel level image fusion algorithms (PCA, Brovey, Multiplicative, Wavelet and combination of PCA & IHS) are used for integrating the derived information like texture, roughness, polarization from microwave data and high spectral information from hyperspectral data. Span image which is total intensity image generated from Advanced Land observing Satellite-Phase array L-band SAR (ALOS-PALSAR) quad polarization data and EO-1 Hyperion data (242 spectral bands) were used for fusion. Overall PCA fused images had shown better result than other fusion techniques used in this study. However, Brovey fusion method was found good for differentiating urban features. Classification using support vector machines was conducted for classifying Hyperion, ALOS PALSAR and fused images. It was observed that overall classification accuracy and kappa coefficient with PCA fused images was relatively better than other fusion techniques as it was able to discriminate various LULC features more clearly.     

Object- and pixel-based classifications of macroalgae farming area with high spatial resolution imagery

Macroalgae plays an important role in coastal ecosystems. The accurate delineation of macroalgae areas is important for environmental management. This study compared the pixel- and object-based methods using Gaofen satellite no. 2 image to explore an efficient classification approach. Expert system rules and nearest neighbour classifier were adopted for object-based classification, whereas maximum likelihood classifier was implemented in the pixel-based approach. Normalized difference vegetation index, normalized difference water index, mean value of the blue band and geometric characteristics were selected as features to distinguish macroalgae farms by considering the spectral and spatial characteristics. Results show that the object-based method achieved a higher overall accuracy and kappa coefficient than the pixel-based method. Moreover, the object-based approach displayed superiority in identifying Porphyra class. These findings suggest that the object-based method can delineate macroalgae farming areas efficiently and be applied in the future to monitor the macroalgae farms with high spatial resolution imagery.     

Quantifying deforestation in the Brazilian Amazon using Advanced Land Observing Satellite Phased Array L-band Synthetic Aperture Radar (ALOS PALSAR) and Shuttle Imaging Radar (SIR)-C data

The study examined the capability of dual-polarization SAR data for forest cover mapping and change assessment in the Brazilian Amazon Forest regions. Shuttle Imaging Radar (SIR)-C and Advanced Land Observing Satellite Phased Array L-band Synthetic Aperture Radar (ALOS PALSAR) data were analysed to map and quantify deforestation. The images were classified using hybrid classifier, where each land cover was grouped in various spectral sub-classes interpreted on the imagery and later merged together to generate the desired land cover classes. The classification accuracy for forest was reasonably high (>90%). The technique applied in this study can be extended for operational mapping and monitoring of deforestation in the tropics, particularly for those regions which are often covered by cloud.     

Dimensionality Reduction and Classification of Hyperspectral Images Using Object-Based Image Analysis

Object-based image analysis (OBIA) has attained great importance for the delineation of landscape features, particularly with the accessibility to satellite images with high spatial resolution acquired by recent sensors. Statistical parametric classifiers have become ineffective mainly due to their assumption of normal distribution, vast increase in the dimensions of the data and availability of limited ground sample data. Despite pixel-based approaches, OBIA takes semantic information of extracted image objects into consideration, and thus provides more comprehensive image analysis. In this study, Indian Pines hyperspectral data set, which was recorded by the AVIRIS hyperspectral sensor, was used to analyse the effects of high dimensional data with limited ground reference data. To avoid the dimensionality curse, principal component analysis (PCA) and feature selection based on Jeffries–Matusita (JM) distance were utilized. First 19 principal components representing 98.5% of the image were selected using the PCA technique whilst 30 spectral bands of the image were determined using JM distance. Nearest neighbour (NN) and random forest (RF) classifiers were employed to test the performances of pixel- and object-based classification using conventional accuracy metrics. It was found that object-based approach outperformed the traditional pixel-based approach for all cases (up to 18% improvement). Also, the RF classifier produced significantly more accurate results (up to 10%) than the NN classifier.     

Evaluating SAR polarization modes at L-band for forest classification purposes in Eastern Amazon,Brazil

Single, interferometric dual, and quad-polarization mode data were evaluated for the characterization and classification of seven land use classes in an area with shifting cultivation practices located in the Eastern Amazon (Brazil). The Advanced Land-Observing Satellite (ALOS) Phased Array L-band Synthetic Aperture Radar (PALSAR) data were acquired during a six month interval. A clear-sky Landsat-5/TM image acquired at the same period was used as additional ground reference and as ancillary input data in the classification scheme. We evaluated backscattering intensity, polarimetric features, interferometric coherence and texture parameters for classification purposes using support vector machines (SVM) and feature selection. Results showed that the forest classes were characterized by low temporal backscattering intensity variability, low coherence and high entropy. Quad polarization mode performed better than dual and single polarizations but overall accuracies remain low and were affected by precipitation events on the date and prior SAR date acquisition. Misclassifications were reduced by integrating Landsat data and an overall accuracy of 85% was attained. The integration of Landsat to both quad and dual polarization modes showed similarity at the 5% significance level. SVM was not affected by SAR dimensionality and feature selection technique reveals that co-polarized channels as well as SAR derived parameters such as Alpha-Entropy decomposition were important ranked features after Landsat’ near-infrared and green bands. We show that in absence of Landsat data, polarimetric features extracted from quad-polarization L-band increase classification accuracies when compared to single and dual polarization alone. We argue that the joint analysis of SAR and their derived parameters with optical data performs even better and thus encourage the further development of joint techniques under the Reducing Emissions from Deforestation and Degradation (REDD) mechanism.     

Landsat 8 vs. Landsat 5: A comparison based on urban and peri-urban land cover mapping

An image dataset from the Landsat OLI spaceborne sensor is compared with the Landsat TM in order to evaluate the excellence of the new imagery in urban landcover classification. Widely known pixel-based and object-based image analysis methods have been implemented in this work like Maximum Likelihood, Support Vector Machine, k-Nearest Neighbor, Feature Analyst and Sub-pixel. Classification results from Landsat OLI provide more accurate results comparing to the Landsat TM. Object-based classifications produced a more uniform result, but suffer from the absorption of small rare classes into large homogenous areas, as a consequence of the segmentation, merging and the spatial parameters in the spatial resolution (30 m) of Landsat images. Based exclusively on the overall accuracy reports, the SVM pixel-based classification from Landsat 8 proved to be the most accurate for the purpose of mapping urban land cover, using medium spatial resolution imagery.     

Object-based classification of hyperspectral data using Random Forest algorithm

This paper presents a new framework for object-based classification of high-resolution hyperspectral data. This multi-step framework is based on multi-resolution segmentation (MRS) and Random Forest classifier (RFC) algorithms. The first step is to determine of weights of the input features while using the object-based approach with MRS to processing such images. Given the high number of input features, an automatic method is needed for estimation of this parameter. Moreover, we used the Variable Importance (VI), one of the outputs of the RFC, to determine the importance of each image band. Then, based on this parameter and other required parameters, the image is segmented into some homogenous regions. Finally, the RFC is carried out based on the characteristics of segments for converting them into meaningful objects. The proposed method, as well as, the conventional pixel-based RFC and Support Vector Machine (SVM) method was applied to three different hyperspectral data-sets with various spectral and spatial characteristics. These data were acquired by the HyMap, the Airborne Prism Experiment (APEX), and the Compact Airborne Spectrographic Imager (CASI) hyperspectral sensors. The experimental results show that the proposed method is more consistent for land cover mapping in various areas. The overall classification accuracy (OA), obtained by the proposed method was 95.48, 86.57, and 84.29% for the HyMap, the APEX, and the CASI data-sets, respectively. Moreover, this method showed better efficiency in comparison to the spectral-based classifications because the OAs of the proposed method was 5.67 and 3.75% higher than the conventional RFC and SVM classifiers, respectively.     

Per-pixel and object-oriented classification methods for mapping urban land cover extraction using SPOT 5 imagery

To have sustainable management and proper decision-making, timely acquisition and analysis of surface features are necessary. Traditional pixel-based analysis is the popular way to extract different categories, but it is not comparable by the achievements that can be achieved through the object-based method that uses the additional characteristics of features in the process of classification. In this paper, three types of classification were used to classify SPOT 5 satellite image in mapping land cover; Support vector machine (SVM) pixel-based, SVM object-based and Decision Tree (DT) pixel-based classification. Normalised Difference Vegetation Index and the brightness value of two infrared bands (NIR and SWIR) were used in manually developed DT classification. The classification of the SVM (pixel based) was generated using the selected groups of pixels that represent the selected features. In addition, the SVM (object based) was implemented by using radial-based function kernel. The classified features were oil palm, rubber, urban area, soil, water and other vegetation. The study found that the overall classification of the DT was the lowest at 69.87% while those of SVM (pixel based) and SVM (object based) were 76.67 and 81.25%, respectively.     

Spectral angle mapper and object-based classification combined with hyperspectral remote sensing imagery for obtaining land use/cover mapping in a Mediterranean region

In this study, we test the potential of two different classification algorithms, namely the spectral angle mapper (SAM) and object-based classifier for mapping the land use/cover characteristics using a Hyperion imagery. We chose a study region that represents a typical Mediterranean setting in terms of landscape structure, composition and heterogeneous land cover classes. Accuracy assessment of the land cover classes was performed based on the error matrix statistics. Validation points were derived from visual interpretation of multispectral high resolution QuickBird-2 satellite imagery. Results from both the classifiers yielded more than 70% classification accuracy. However, the object-based classification clearly outperformed the SAM by 7.91% overall accuracy (OA) and a relatively high kappa coefficient. Similar results were observed in the classification of the individual classes. Our results highlight the potential of hyperspectral remote sensing data as well as object-based classification approach for mapping heterogeneous land use/cover in a typical Mediterranean setting.     

人工蜂群算法优化的SVM遥感影像分类

SVM分类器的参数设定对分类精度有着显著的影响,针对现有人工智能算法优化参数易陷入局部最优的现状,提出了一种基于人工蜂群算法改进SVM参数的遥感分类方法(ABC-SVM)。该方法模仿蜜蜂采蜜的行为,以训练样本的交叉验证精度代表蜜源的丰富程度,通过蜂群的分工协作搜索出最优蜜源(即SVM分类器最优参数),最终利用参数优化后的SVM分类器实现遥感影像的分类。本文先后比较了3种人工智能算法(包括人工蜂群算法优化的SVM(ABC-SVM)、遗传算法GA(Genetic Algorithm)优化的SVM(GA-SVM)、粒子群算法PSO(Practical Swarm Optimization)优化的SVM(PSO-SVM))在UCI标准数据集上的分类精度和效率,以及3种人工智能算法优化的SVM算法与未经优化参数的SVM算法在遥感影像上分类的差异。结果显示:(1)在利用UCI数据集测试3种人工智能算法优化的SVM算法的结果中,ABC-SVM显示出更高的分类精度、更高的适应度和更快的收敛速度;(2)在利用遥感影像验证4种分类算法精度的结果中,人工智能算法优化后的SVM比未经参数优化的SVM算法的分类精度更高;其中,ABC-SVM分类精度最高,分别比遗传算法、粒子群算法的结果高1.67%、1.50%。     

VHSR图像基于分割对象分类器性能评价

对比研究了平行六面体、最近邻分类法、最大似然法、神经网络等经典分类算法以及近年来新发展的支持向量机分类算法在基于分割对象的高分辨率遥感图像分类中的性能,详细分析了不同内积核函数对于支持向量机分类的影响。对两个试验区进行试验的结果表明,支持向量机分类算法分类精度得到明显改善,同时分类结果受参数、样本选择等影响较小,稳定性好。     

高分辨率遥感影像5种面向对象分类方法对比研究

针对主流的面向对象分类方法在遥感影像处理中的使用范围不明确的问题,以e-Cognition软件平台为基础,处理标准数据集,综合考虑视觉效果、总体精度和用户精度3方面,系统地比较分析了主流的面向对象分类方法在高分辨率影像中的分类效果和精度分析。试验结果表明：使用不同的分类方法均存在混分现象且混分对象不完全一样。在处理同一标准数据集时,隶属度函数分类方法的精度最高但分类速度最慢,Bayes的分类效果最差但操作简单,支持向量机（SVM）、决策树（DT）、随机森林（RF）的分类速度均较快且都有较高的精度,其中SVM分类方法在区分相似性高的对象方面具有明显优势。在选择高分影像分类方法时,要充分考虑分类影像的特征选择从而选择合适的分类方法。     

基于双频多极化数据的农作物后向散射特性模拟分析及类型识别

以北京昌平地区为研究区域,获取了2007年该试验区C波段ENVISAT/ASAR数据和L波段ALOS/PALSAR数据,并提取了地物的后向散射系数。首先,利用MIMICS模型对该地区的春玉米、夏玉米和果木的后向散射特性进行模拟和分析;然后,将模拟结果同雷达实际观测数据进行对比;最后,利用不同作物之间的后向散射系数数值大小关系,建立分类二叉树,很好地区分了春玉米和夏玉米,总分类精度达86.66%。研究结果表明:双频多极化雷达数据能够提供有利于作物类型识别的多方面信息,对农作物遥感具有较大的优势和潜力。     

A Comparison of Multisensor Integration Methods for Land Cover Classification in the Brazilian Amazon

Many data fusion methods are available, but it is poorly understood which fusion method is suitable for integrating Landsat Thematic Mapper (TM) and radar data for land cover classification. This research explores the integration of Landsat TM and radar images (i.e., ALOS PALSAR L-band and RADARSAT-2 C-band) for land cover classification in a moist tropical region of the Brazilian Amazon. Different data fusion methods—principal component analysis (PCA), wavelet-merging technique (Wavelet), high-pass filter resolution-merging (HPF), and normalized multiplication (NMM)—were explored. Land cover classification was conducted with maximum likelihood classification based on different scenarios. This research indicates that individual radar data yield much poorer land cover classifications than TM data, and PALSAR L-band data perform relatively better than RADARSAT-2 C-band data. Compared to the TM data, the Wavelet multisensor fusion improved overall classification by 3.3%-5.7%, HPF performed similarly, but PCA and NMM reduced overall classification accuracy by 5.1%-6.1% and 7.6%-12.7%, respectively. Different polarization options, such as HH and HV, work similarly when used in data fusion. This research underscores the importance of selecting a suitable data fusion method that can preserve spectral fidelity while improving spatial resolution.     

The use of logistic model tree (LMT) for pixel- and object-based classifications using high-resolution WorldView-2 imagery

Logistic model tree (LMT), a new method integrating standard decision tree (DT) induction and linear logistic regression algorithm in a single tree, have been recently proposed as an alternative to DT-based learning algorithms. In this study, the LMT was applied in the context of pixel- and object-based classifications using high-resolution WorldView-2 imagery, and its performance was compared with C4.5, random forest and Adaboost. Results of the study showed that the LMT generally produced more accurate classification results than the other methods for both pixel- and object-based classifications. The improvement in classification accuracy reached to 3% in pixel-based and 5% in object-based classifications. It was also estimated that the LMT algorithm produced the most accurate results considering the allocation and overall disagreement errors. Based on the Wilcoxon’s Signed-Ranks tests, the performance differences between the LMT and the other methods were statistically significant for both pixel- and object-based image classifications.     

Mediterranean forest species mapping using classification of Hyperion imagery

Regional operational forest species mapping is an active research topic that aims to provide the systematic and updatable information necessary for understanding and monitoring the rapidly changing forest environment. In this study, we investigated the potential of satellite hyperspectral imagery in regional forest species mapping by employing a pixel-based and an object-based nearest neighbour classifier in two different Mediterranean study areas. The overall thematic accuracy of the produced maps was assessed using reference data collected in the field and ranged between 0.72 and 0.83. No approach was found to be superior for the study areas. The McNemar test showed no statistically significant difference at the 95% confidence level in the classification accuracies achieved by the two approaches. Both pixel- and object-based approaches provide useful maps, suggesting that regional forest species mapping from space has much potential.