结合随机森林面向对象的森林资源分类

针对森林资源分类研究较少且缺少相对简单有效的方法的情况,提出一种结合面向对象和随机森林的森林资源分类方法。面向对象分割技术可减少“椒盐效应”,随机森林分类算法具有高准确度、抗噪声能力强、性能稳定等优势。鉴于此,通过调整面向对象的分割参数,构造最优特征空间及估算随机森林中决策树的数量等,构建了最优的面向对象随机森林分类模型。另外,选择了SVM算法作对比试验。试验结果显示,本文分类算法的总体精度达到83.34%,Kappa系数为0.7892,明显高于SVM,能够有效提高森林资源分类的精度。     

Fusion of HJ1B and ALOS PALSAR data for land cover classification using machine learning methods

Image classification from remote sensing is becoming increasingly urgent for monitoring environmental changes. Exploring effective algorithms to increase classification accuracy is critical. This paper explores the use of multispectral HJ1B and ALOS (Advanced Land Observing Satellite) PALSAR L-band (Phased Array type L-band Synthetic Aperture Radar) for land cover classification using learning-based algorithms. Pixel-based and object-based image analysis approaches for classifying HJ1B data and the HJ1B and ALOS/PALSAR fused-images were compared using two machine learning algorithms, support vector machine (SVM) and random forest (RF), to test which algorithm can achieve the best classification accuracy in arid and semiarid regions. The overall accuracies of the pixel-based (Fused data: 79.0%; HJ1B data: 81.46%) and object-based classifications (Fused data: 80.0%; HJ1B data: 76.9%) were relatively close when using the SVM classifier. The pixel-based classification achieved a high overall accuracy (85.5%) using the RF algorithm for classifying the fused data, whereas the RF classifier using the object-based image analysis produced a lower overall accuracy (70.2%). The study demonstrates that the pixel-based classification utilized fewer variables and performed relatively better than the object-based classification using HJ1B imagery and the fused data. Generally, the integration of the HJ1B and ALOS/PALSAR imagery can improve the overall accuracy of 5.7% using the pixel-based image analysis and RF classifier.     

Mapping multiple tree species classes using a hierarchical procedure with optimized node variables and thresholds based on high spatial resolution satellite data

ABSTRACT

Tree species distribution mapping using remotely sensed data has long been an important research area. However, previous studies have rarely established a comprehensive and efficient classification procedure to obtain an accurate result. This study proposes a hierarchical classification procedure with optimized node variables and thresholds to classify tree species based on high spatial resolution satellite imagery. A classification tree structure consisting of parent and leaf nodes was designed based on user experience and visual interpretation. Spectral, textural, and topographic variables were extracted based on pre-segmented images. The random forest algorithm was used to select variables by ranking the impact of all variables. An iterating approach was used to optimize variables and thresholds in each loop by comprehensively considering the test accuracy and selected variables. The threshold range for each selected variable was determined by a statistical method considering the mean and standard deviation for two subnode types at each parent node. Classification of tree species was implemented using the optimized variables and thresholds. The results show that (1) the proposed procedure can accurately map the tree species distribution, with an overall accuracy of over 86% for both training and test stages; (2) critical variables for each class can be identified using this proposed procedure, and optimal variables of most tree plantation nodes are spectra related; (3) the overall forest classification accuracy using the proposed method is more accurate than that using the random forest (RF) and classification and regression tree (CART). The proposed approach provides results with 3.21% and 7.56% higher overall land cover classification accuracy and 4.68% and 10.28% higher overall forest classification accuracy than RF and CART, respectively.     

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.     

Estimation of aboveground biomass of Robinia pseudoacacia forest in the Yellow River Delta based on UAV and Backpack LiDAR point clouds

Forest plantations are an important source of terrestrial carbon sequestration. The forest of Robinia pseudoacacia in the Yellow River Delta (YRD) is the largest artificial ecological protection forest in China. However, more than half of the forest has appeared different degrees of dieback and even death since the 1990s. Timely and accurate estimation of the forest aboveground biomass (AGB) is a basis for studying the carbon cycle of forests. Light Detecting and Ranging (LiDAR) has been proved to be one of the most powerful methods for forest biomass estimation. However, because of an irregular and overlapping shape of the broadleaved forest canopy in a growing season, it is difficult to segment individual trees and estimate the tree biomass from airborne LiDAR data. In this study, a new method was proposed to solve this problem of individual tree detection in the Robinia pseudoacacia forest based on a combination of the Unmanned Aerial Vehicle-Light Detecting and Ranging (UAV-LiDAR) with the Backpack-LiDAR. The proposed method mainly consists of following steps: (i) at a plot level, trees in the UAV-LiDAR data were detected by seed points obtained by an individual tree segmentation (ITS) method from the Backpack-LiDAR data; (ii) height and diameter at breast height (DBH) of an individual tree would be extracted from UAV and Backpack LiDAR data, respectively; (iii) the individual tree AGB would be calculated through an allometric equation and the forest AGB at the plot level was accumulated; and (iv) the plot-level forest AGB was taken as a dependent variable, and various metrics extracted from UAV-LiDAR point cloud data as independent variables to estimate forest AGB distribution in the study area by using both multiple linear regression (MLR) and random forest (RF) models. The results demonstrate that: (1) the seed points extracted from Backpack-LiDAR could significantly improve the overall accuracy of individual tree detection (F = 0.99), and thus increase the forest AGB estimation accuracy; (2) compared with MLR model, the RF model led to a higher estimation accuracy (p < 0.05); and (3) LiDAR intensity information selected by both MLR and RF models and laser penetration rate (LP) played an important role in estimating healthy forest AGB.     

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.     

GEE环境下联合Sentinel主被动遥感数据的国家公园土地覆盖分类

国家公园的土地覆盖分类对于掌握自然资源现状、查明存在的生态安全威胁并快速应对具有基础性数据支撑作用。基于谷歌地球引擎（Google Earth Engine,GEE）平台,结合哨兵（Sentinel）主被动遥感数据及其导出的光谱指数、纹理特征和地形特征,分别采用基于像元的随机森林（random forest,RF）算法和面向对象的简单非迭代聚类（simple noniterative clustering,SNIC）+RF算法实现了钱江源国家公园异质性景观的土地覆盖（耕地、森林、草地、水体、人造地表和裸地）分类。地面实验表明,在多种输入数据组合中,基于像元和面向对象方法分类获得的最高总体精度分别为92.37%和93.98%。合成孔径雷达（synthetic aperture radar,SAR）数据的纳入能够提高基于像元方法的分类精度,但在面向对象方法中未能体现精度提升效果。通过SNIC+RF算法生成的土地覆盖分类图完整性更好,所需特征数量较少,并且算法能够在GEE环境下快速执行,适用于国家公园管理实践。     

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.     

The effect of atmospheric and topographic correction methods on land cover classification accuracy

Mapping of vegetation in mountain areas based on remote sensing is obstructed by atmospheric and topographic distortions. A variety of atmospheric and topographic correction methods has been proposed to minimize atmospheric and topographic effects and should in principle lead to a better land cover classification. Only a limited number of atmospheric and topographic combinations has been tested and the effect on class accuracy and on different illumination conditions is not yet researched extensively. The purpose of this study was to evaluate the effect of coupled correction methods on land cover classification accuracy. Therefore, all combinations of three atmospheric (no atmospheric correction, dark object subtraction and correction based on transmittance functions) and five topographic corrections (no topographic correction, band ratioing, cosine correction, pixel-based Minnaert and pixel-based C-correction) were applied on two acquisitions (2009 and 2010) of a Landsat image in the Romanian Carpathian mountains. The accuracies of the fifteen resulting land cover maps were evaluated statistically based on two validation sets: a random validation set and a validation subset containing pixels present in the difference area between the uncorrected classification and one of the fourteen corrected classifications. New insights into the differences in classification accuracy were obtained. First, results showed that all corrected images resulted in higher overall classification accuracies than the uncorrected images. The highest accuracy for the full validation set was achieved after combination of an atmospheric correction based on transmittance functions and a pixel-based Minnaert topographic correction. Secondly, class accuracies of especially the coniferous and mixed forest classes were enhanced after correction. There was only a minor improvement for the other land cover classes (broadleaved forest, bare soil, grass and water). This was explained by the position of different land cover types in the landscape. Finally, coupled correction methods showed most efficient on weakly illuminated slopes. After correction, accuracies in the low illumination zone (cos β ≤ 0.65) were improved more than in the moderate and high illumination zones. Considering all results, best overall classification results were achieved after combination of the transmittance function correction with pixel-based Minnaert or pixel-based C-topographic correction. Furthermore, results of this bi-temporal study indicated that the topographic component had a higher influence on classification accuracy than the atmospheric component and that it is worthwhile to invest in both atmospheric and topographic corrections in a multi-temporal study.     

基于GF-1号影像的南方水稻种植信息提取

针对南方复杂地区水稻遥感信息提取研究中机器自动学习分类研究较少、分类精度不高的问题,以福建省三明市建宁县溪口镇为研究区,基于GF-1号卫星影像,采用面向对象的随机森林遥感分类算法对研究区内水稻田信息进行提取。首先通过优化面向对象分割参数和随机森林分类模型参数,提取并调用了影像中的多种特征;再对光谱特征、植被指数特征、纹理特征、几何特征进行特征空间优选;最后通过设置4种特征优选试验进行对比,得到最优分类模型。实验结果显示,基于特征空间优选的面向对象随机森林分类算法的水稻提取精度高达90%,分类总体精度可达87%,Kappa系数为0.85;与其他试验结果相比,漏分和误分现象较少,实现了南方地区水稻信息高精度自动识别。该方法计算特征少、实现简便,对于国产高分卫星影像在南方复杂地区作物自动提取中的应用具有参考性。     

Measures of spatio-temporal accuracy for time series land cover data

Remote sensing is a useful tool for monitoring changes in land cover over time. The accuracy of such time-series analyses has hitherto only been assessed using confusion matrices. The matrix allows global measures of user, producer and overall accuracies to be generated, but lacks consideration of any spatial aspects of accuracy. It is well known that land cover errors are typically spatially auto-correlated and can have a distinct spatial distribution. As yet little work has considered the temporal dimension and investigated the persistence or errors in both geographic and temporal dimensions. Spatio-temporal errors can have a profound impact on both change detection and on environmental monitoring and modelling activities using land cover data. This study investigated methods for describing the spatio-temporal characteristics of classification accuracy. Annual thematic maps were created using a random forest classification of MODIS data over the Jakarta metropolitan areas for the period of 2001–2013. A logistic geographically weighted model was used to estimate annual spatial measures of user, producer and overall accuracies. A principal component analysis was then used to extract summaries of the multi-temporal accuracy. The results showed how the spatial distribution of user and producer accuracy varied over space and time, and overall spatial variance was confirmed by the principal component analysis. The results indicated that areas of homogeneous land cover were mapped with relatively high accuracy and low variability, and areas of mixed land cover with the opposite characteristics. A multi-temporal spatial approach to accuracy is shown to provide more informative measures of accuracy, allowing map producers and users to evaluate time series thematic maps more comprehensively than a standard confusion matrix approach. The need to identify suitable properties for a temporal kernel are discussed.     

面向对象的旱区植被遥感精细分类研究

针对旱区植被分类尺度过大、种群无法准确提取的问题,该文提出了面向对象的CFS-RF分类模型,即利用CFS算法对先验样本数据集进行特征优选,结合随机森林构建分类规则,完成分类过程。以新疆阿勒泰为研究区,利用GF-2数据,通过CFS、ReliefF两种不同特征选择方法和J48、SVM、RF 3种分类算法构造出6种面向对象分类方案来实现小尺度植被种群提取。结果表明,经过特征选择,上述分类方案的精度和效率均得到了提升。其中,CFS-RF算法最优,总体精度达到92.41%,Kappa系数为0.90,更适用于旱区植被遥感精细分类。     

Adjustment of geodetic measurements with mixed multiplicative and additive random errors

Adjustment has been based on the assumption that random errors of measurements are added to functional models. In geodetic practice, we know that accuracy formulae of modern geodetic measurements often consist of two parts: one proportional to the measured quantity and the other constant. From the statistical point of view, such measurements are of mixed multiplicative and additive random errors. However, almost no adjustment has been developed to strictly address geodetic data contaminated by mixed multiplicative and additive random errors from the statistical point of view. We systematically develop adjustment methods for geodetic data contaminated with multiplicative and additive errors. More precisely, we discuss the ordinary least squares (LS) and weighted LS methods and extend the bias-corrected weighted LS method of Xu and Shimada (Commun Stat B29:83–96, 2000) to the case of mixed multiplicative and additive random errors. The first order approximation of accuracy for all these three methods is derived. We derive the biases of weighted LS estimates. The three methods are then demonstrated and compared with a synthetic example of surface interpolation. The bias-corrected weighted LS estimate is unbiased up to the second order approximation and is of the best accuracy. Although the LS method can warrant an unbiased estimate for a linear model with multiplicative and additive errors, it is less accurate and always produces a very poor estimate of the variance of unit weight.     

GF-6影像应用于林地与非林地识别的潜力分析

为研究我国首颗携带红边波段的高分六影像（GF-6）在林地与非林地上的识别贡献,本文选择复杂林地类型的安徽省黄山市作为研究区,采用特征优选（RFE）与随机森林（RF）相结合的方法开展了林地与非林地识别潜力研究。首先根据实地调查、Google Earth影像及林地"一张图"样本数据构建了样本库;然后基于DEM、多时相光谱特征、植被指数、红边指数等特征开展分类,并比较不同模型精度及不同变量的重要度。结果表明：GF-6红边信息对林地非林地识别较为重要,引入红边信息可将总体分类精度提升2%,其他新增波段及地形特征对林地与非林地识别贡献并不明显;多时相数据的运用相比单时相数据可整体提高林地类型的分类精度2.93%~4.1%,单时相分类结果6月最好,9月次之,12月最差;特征优选可以有效减少数据输入维数（46到15）,并取得最高分类精度,在不牺牲精度的同时保证了运算数据量的减少且明确了不同变量的贡献,具有较强的应用意义。     

超体素随机森林与LSTM神经网络联合优化的室内点云高精度分类方法

针对现有三维点云数据分割分类方法存在分类目标内部不一致的问题,提出一种超体素随机森林与长短期记忆神经网络（long short-term memory,LSTM）联合优化的室内点云高精度分类方法。该方法根据超体素结构具备内部特征一致性的特点,对原始点云进行超体素划分,并以超体素为基本单元进行多元特征计算,搭建室内点云超体素随机森林分类模型,实现点云数据的粗分类。在此基础上,引入LSTM对粗分类的超体素邻域连接关系进行神经网络模型训练与预测,实现超体素粗分类结果的优化。基于开放数据集对所提分类方法进行有效性和精度验证,结果显示,该方法在公开数据集中对13类要素的分类精度可达到83.2%;与经典的深度学习框架相比,该方法在小样本训练时可以达到更优的分类精度。     

基于随机森林算法的城市不透水面信息提取——以长春市为例

为了快速、准确地掌握不透水面的空间分布及满足动态变化信息现实需求,本文基于多分类器集成学习的思想,引入随机森林算法,以Landsat8影像为数据源,长春市为实验区,选取光谱特征、纹理测度、空间变换后的独立分量等25个特征变量进行分类研究,根据OOB误差进行重要性分析并试验得出最优的分类模型,实现高精度不透水面信息的提取,最后与传统参数分类法进行比较。结果表明：随机森林算法的总体精度可以达到94%,高出最大似然分类法5.9%,支持向量机算法0.77%,Kappa系数为0.914 3,均方根误差为0.104 3,不透水面的提取精度达95.54%,可以精确地得出所需信息,为城市建设与规划提供有效的专题数据。     

高光谱遥感影像多级联森林深度网络分类算法

高光谱遥感技术在环境监测、应急保障、精细地物提取等方面有着广泛的应用,随着高分五号高光谱数据的正式发布,高光谱遥感技术将发挥更重要的作用。遥感影像分类作为高光谱遥感影像信息处理的重要部分,已成为当前研究重点。本文针对传统多级联森林深度学习中模型复杂、无法利用基分类器差异信息、对类间差异较小的样本无法正确区分等不足,提出了一种改进的多级联森林深度学习模型,在模型框架中,分别采用了随机森林和旋转森林作为基分类器,并引入逻辑回归分类器作为判别器用于训练层扩展。相较于传统的深度神经网络,改进的多级联森林深度网络超参数较少且能够自适应确定训练层,更方便进行模型优化。实验采用了高分五号数据集及两个公开的高光谱数据集(Indian Pines数据集及Pavia University数据集)进行精度评定,同时选择了传统分类器支持向量机、深度置信网等模型作为对比分析。实验结果表明,改进的多级联森林深度学习模型能有效地进行高光谱遥感影像分类,且较传统的分类方法精度有所提升。     

In-season crop classification using elements of the Kennaugh matrix derived from polarimetric RADARSAT-2 SAR data

Accurate spatio-temporal classification of crops is of prime importance for in-season crop monitoring. Synthetic Aperture Radar (SAR) data provides diverse physical information about crop morphology. In the present work, we propose a day-wise and a time-series approach for crop classification using full-polarimetric SAR data. In this context, the 4 × 4 real Kennaugh matrix representation of a full-polarimetric SAR data is utilized, which can provide valuable information about various morphological and dielectric attributes of a scatterer. The elements of the Kennaugh matrix are used as the parameters for the classification of crop types using the random forest and the extreme gradient boosting classifiers.The time-series approach uses data patterns throughout the whole growth period, while the day-wise approach analyzes the PolSAR data from each acquisition into a single data stack for training and validation. The main advantage of this approach is the possibility of generating an intermediate crop map, whenever a SAR acquisition is available for any particular day. Besides, the day-wise approach has the least climatic influence as compared to the time series approach. However, as time-series data retains the crop growth signature in the entire growth cycle, the classification accuracy is usually higher than the day-wise data.Within the Joint Experiment for Crop Assessment and Monitoring (JECAM) initiative, in situ measurements collected over the Canadian and Indian test sites and C-band full-polarimetric RADARSAT-2 data are used for the training and validation of the classifiers. Besides, the sensitivity of the Kennaugh matrix elements to crop morphology is apparent in this study. The overall classification accuracies of 87.75% and 80.41% are achieved for the time-series data over the Indian and Canadian test sites, respectively. However, for the day-wise data, a ∼6% decrease in the overall accuracy is observed for both the classifiers.     

Interference of Human Impacts in Urban Growth Modelling with Transition Rules of Cellular Automata, GIS and Multi-Temporal Satellite Imagery: A Case Study of Maraghe, Iran

Urban growth is the result of physical and human impacts. In this study Cellular Automata (CA) has been used to analyze physical suitability and human forces in urban growth modelling of Maraghe. The multi-temporal satellite imagery, physical suitability and human impacts Layers have been applied to the modelling. In order to evaluate the accuracy of the image classification methods, Fuzzy ARTMAP is compared with Maximum Likelihood Classification (MLC) and Minimum Distance Classification (MDC) methods. The image classification results showed an overall accuracy of 93 %. Therefore, it is employed for classification of multi-temporal satellite imagery. In order to weight physical suitability and human impacts layers or geographical transition rules in the modelling, regression analysis, the correlation coefficient, trial-and-error method and visual comparison used. The statistical methods are presented to validate neighbourhood scales in the urban growth modelling. The calibration of the model is in fact to the estimate value of the physical suitability and human impacts layer (combinatory layer of demand for urban land and the government facilities) in the modelling. The results obtained from the model calibration showed that human impacts have the highest influence in the urban growth among other factors. Also a small neighbourhood scale (25:5?×?5 cells) is more realistic in the modeling. The accuracy of final validation is 83 % and the final scenario is based on this validation. A fuzzy CA has been used in urban growth modeling of Maraghe. The final scenario shows that Maraghe will growth on the east side, where the land demand for built up area and government facilities plays the significant role.     

机器学习算法在森林地上生物量估算中的应用

森林地上生物量是森林生产力的重要评价指标,对其进行高效监测对维持全球碳平衡和保护生态系统具有重要意义。本文首先基于冠层高度模型数据,通过分水岭分割算法得到单木冠幅边界;然后在单木冠幅范围内提取23个LiDAR变量,结合佩诺布斯科特试验森林的87组实测数据,利用随机森林和支持向量机建立森林地上生物量估算模型;最后对样地模型估算的结果进行了比较,讨论了预测结果及其精度。结果表明:本文选用的随机森林模型和支持向量机模型在估算森林地上生物量的应用中获得了较高的精度;并且,随机森林模型在基于机载雷达数据估测森林地上生物量中的估算精度更高,模型泛化能力更强,制图精度也更好,具有更好的适用性。