Validation of the collection 5 MODIS FPAR product in a heterogeneous agricultural landscape in arid Uzbekistan using multitemporal RapidEye imagery

The fraction of photosynthetically active radiation (FPAR) absorbed by a vegetation canopy is an important variable for global vegetation modelling and is operationally available from data of the Terra Moderate Resolution Imaging Spectroradiometer (MODIS) satellite sensor starting from the year 2000. Product validation is ongoing and important for constant product improvement, but few studies have investigated the specific accuracy of MODIS FPAR using in situ measurements and none have focused on agricultural areas. This study therefore presents a validation of the collection 5 MODIS FPAR product in a heterogeneous agricultural landscape in western Uzbekistan. High-resolution FPAR maps were compiled via linear regression between in situ FPAR measurements and the RapidEye normalized difference vegetation index (NDVI) for the 2009 season. The data were aggregated to the MODIS scale for comparison. Data on the percentage cover of agricultural crops per MODIS pixel allowed investigation of the impact of spatial heterogeneity on MODIS FPAR accuracy. Overall, the collection 5 MODIS FPAR overestimated RapidEye FPAR between approximately 6% and 15%. MODIS quality flags, the underlying biome classification and spatial heterogeneity were investigated as potential sources of error. MODIS data quality was very good in all cases. A comparison of the MODIS land-cover product with high-resolution land-use classification revealed a significant misclassification by MODIS. Yet, we found that the overestimation of MODIS FPAR is independent of classification accuracy. The results indicate that the amount of background information, present even in the most homogeneous pixels (～70% crop cover), is most likely the reason for the overestimation. The behaviour of pure pixels could not be investigated due to a lack of appropriate pixels.     

Use of a dark object concept and support vector machines to automate forest cover change analysis

An automated method was developed for mapping forest cover change using satellite remote sensing data sets. This multi-temporal classification method consists of a training data automation (TDA) procedure and uses the advanced support vector machines (SVM) algorithm. The TDA procedure automatically generates training data using input satellite images and existing land cover products. The derived high quality training data allow the SVM to produce reliable forest cover change products. This approach was tested in 19 study areas selected from major forest biomes across the globe. In each area a forest cover change map was produced using a pair of Landsat images acquired around 1990 and 2000. High resolution IKONOS images and independently developed reference data sets were available for evaluating the derived change products in 7 of those areas. The overall accuracy values were over 90% for 5 areas, and were 89.4% and 89.6% for the remaining two areas. The user's and producer's accuracies of the forest loss class were over 80% for all 7 study areas, demonstrating that this method is especially effective for mapping major disturbances with low commission errors. IKONOS images were also available in the remaining 12 study areas but they were either located in non-forest areas or in forest areas that did not experience forest cover change between 1990 and 2000. For those areas the IKONOS images were used to assist visual interpretation of the Landsat images in assessing the derived change products. This visual assessment revealed that for most of those areas the derived change products likely were as reliable as those in the 7 areas where accuracy assessment was conducted. The results also suggest that images acquired during leaf-off seasons should not be used in forest cover change analysis in areas where deciduous forests exist. Being highly automatic and with demonstrated capability to produce reliable change products, the TDA-SVM method should be especially useful for quantifying forest cover change over large areas.     

Comparison of artificial neural networks and support vector machine classifiers for land cover classification in Northern China using a SPOT-5 HRG image

This article presents a sufficient comparison of two types of advanced non-parametric classifiers implemented in remote sensing for land cover classification. A SPOT-5 HRG image of Yanqing County, Beijing, China, was used, in which agriculture and forest dominate land use. Artificial neural networks (ANNs), including the adaptive backpropagation (ABP) algorithm, Levenberg–Marquardt (LM) algorithm, Quasi-Newton (QN) algorithm and radial basis function (RBF) were carefully tested. The LM–ANN and RBF–ANN, which outperform the other two, were selected to make a detailed comparison with support vector machines (SVMs). The experiments show that those well-trained ANNs and SVMs have no significant difference in classification accuracy, but the SVM usually performs slightly better. Analysis of the effect of the training set size highlights that the SVM classifier has great tolerance on a small training set and avoids the problem of insufficient training of ANN classifiers. The testing also illustrates that the ANNs and SVMs can vary greatly with regard to training time. The LM–ANN can converge very quickly but not in a stable manner. By contrast, the training of RBF–ANN and SVM classifiers is fast and can be repeatable.     

Integrating LiDAR and high-resolution imagery for object-based mapping of forest habitats in a heterogeneous temperate forest landscape

ABSTRACT

Land-cover mapping (LCM) at a fine scale would be useful for forest management across heterogeneous natural landscapes. However, the heterogeneity of land covers at such scales results in complex spectral and textural properties that hinder the applicability of LCM. Besides, the method suffers from, e.g. inconsistent representation of different land-cover types, lack of sufficient and balanced training samples, and instability of classifiers trained by a high number of predictor variables. Even well-known object-based classification approaches are challenged with an objective evaluation of segmentation outputs. Here we classified partially ambiguous land-cover types across heterogeneous forest landscapes in the Bavarian Forest National Park (Germany) by combining metrics from airborne light detection and ranging (LiDAR) and colour infrared (CIR) imagery data and a random forest classifier implemented in an object-based paradigm. We evaluated the segmentation results by creating a global quality score based on inter- and intra-measurements of variance and the number of segments. Selected segmentation outputs were combined with balanced training samples to run the classification algorithm based on representative blocks within the national park. The entire processing chain was implemented in an open-source domain. The final segmentation consisted of LiDAR-based height, image-based Normalized Difference Vegetation Index (NDVI) and red band, with 20 cluster seeds and a minimum segment size of 40 pixels. In the classification, the most important variables included the height of the top layer, NDVI, Enhanced Vegetation Index (EVI) and Green–Red Vegetation Index (GRVI). The average values of 500 random forest runs indicated an overall accuracy of 86.6% and an estimated Cohen’s kappa coefficient of 85.2%, with different probabilities of correct classification for land-cover classes. Mature deciduous, standing deadwood, fallen deadwood, meadow, and bare soil classes were classified most accurately, whereas classification of young coniferous, intermediate-age coniferous, mature coniferous, young deciduous, and intermediate-age deciduous were associated with the highest uncertainties. Our methodology is sufficiently robust to be applied to other similarly structured sites across temperate forested landscapes. The versatility of the method is partially guaranteed by the proposed segmentation quality score, which satisfactorily corrects under- and over-segmentation.     

Parameter tuning in the support vector machine and random forest and their performances in cross- and same-year crop classification using TerraSAR-X

This article describes the comparison of three different classification algorithms for mapping crops in Hokkaido, Japan, using TerraSAR-X data. In the study area, beans, beets, grasslands, maize, potatoes, and winter wheat were cultivated. Although classification maps are required for both management and estimation of agricultural disaster compensation, those techniques have yet to be established. Some supervised learning models may allow accurate classification. Therefore, comparisons among the classification and regression tree (CART), the support vector machine (SVM), and random forests (RF) were performed. SVM was the optimum algorithm in this study, achieving an overall accuracy of 89.1% for the same-year classification, which is the classification using the training data in 2009 to classify the test data in 2009, and 78.0% for the cross-year classification, which is the classification using the training data in 2009 to classify the data in 2012.     

Gene selection and prediction for cancer classification using support vector machines with a reject option

In cancer classification based on gene expression data, it would be desirable to defer a decision for observations that are difficult to classify. For instance, an observation for which the conditional probability of being cancer is around 1/2 would preferably require more advanced tests rather than an immediate decision. This motivates the use of a classifier with a reject option that reports a warning in cases of observations that are difficult to classify. In this paper, we consider a problem of gene selection with a reject option. Typically, gene expression data comprise of expression levels of several thousands of candidate genes. In such cases, an effective gene selection procedure is necessary to provide a better understanding of the underlying biological system that generates data and to improve prediction performance. We propose a machine learning approach in which we apply the l₁ penalty to the SVM with a reject option. This method is referred to as the l₁ SVM with a reject option. We develop a novel optimization algorithm for this SVM, which is sufficiently fast and stable to analyze gene expression data. The proposed algorithm realizes an entire solution path with respect to the regularization parameter. Results of numerical studies show that, in comparison with the standard l₁ SVM, the proposed method efficiently reduces prediction errors without hampering gene selectivity.     

基于复小波和支持向量机的纹理分类法*

针对图像纹理分类问题,提出了一种将二元树复小波变换与支持向量机相结合的分类方法,通过二元树复小波变换对纹理图像进行四层分解,提取各子频带小波系数模的均值和标准方差组成特征向量,利用支持向量机作为分类器实现纹理图像分类。对20类Brodatz纹理图像的分类实验表明,提出的方法具有较高的分类精度,在有限训练样本的情况下比传统的分类算法平均正确率有10%左右的提高,体现了该方法的有效性和良好的泛化能力。     

Two-phase optimization for support vectors and parameter selection of support vector machines: Two-class classification

Support vector machines (SVMs) are one of the most popular classification tools and show the most potential to address under-sampled noisy data (a large number of features and a relatively small number of samples). However, the computational cost is too expensive, even for modern-scale samples, and the performance largely depends on the proper setting of parameters. As the data scale increases, the improvement in speed becomes increasingly challenging. As the dimension (feature number) largely increases while the sample size remains small, the avoidance of overfitting becomes a significant challenge. In this study, we propose a two-phase sequential minimal optimization (TSMO) to largely reduce the training cost for large-scale data (tested with 3186–70,000-sample datasets) and a two-phased-in differential-learning particle swarm optimization (tDPSO) to ensure the accuracy for under-sampled data (tested with 2000–24481-feature datasets). Because the purpose of training SVMs is to identify support vectors that denote a hyperplane, TSMO is developed to quickly select support vector candidates from the entire dataset and then identify support vectors from those candidates. In this manner, the computational burden is largely reduced (a 29.4%–65.3% reduction rate). The proposed tDPSO uses topology variation and differential learning to solve PSO’s premature convergence issue. Population diversity is ensured through dynamic topology until a ring connection is achieved (topology-variation phases). Further, particles initiate chemo-type simulated-annealing operations, and the global-best particle takes a two-turn diversion in response to stagnation (event-induced phases). The proposed tDPSO-embedded SVMs were tested with several under-sampled noisy cancer datasets and showed superior performance over various methods, even those methods with feature selection for the preprocessing of data.     

Generalization performance of support vector machines and neural networks in runoff modeling

Effective one-day lead runoff prediction is one of the significant aspects of successful water resources management in arid region. For instance, reservoir and hydropower systems call for real-time or on-line site-specific forecasting of the runoff. In this research, we present a new data-driven model called support vector machines (SVMs) based on structural risk minimization principle, which minimizes a bound on a generalized risk (error), as opposed to the empirical risk minimization principle exploited by conventional regression techniques (e.g. ANNs). Thus, this stat-of-the-art methodology for prediction combines excellent generalization property and sparse representation that lead SVMs to be a very promising forecasting method. Further, SVM makes use of a convex quadratic optimization problem; hence, the solution is always unique and globally optimal. To demonstrate the aforementioned forecasting capability of SVM, one-day lead stream flow of Bakhtiyari River in Iran was predicted using the local climate and rainfall data. Moreover, the results were compared with those of ANN and ANN integrated with genetic algorithms (ANN-GA) models. The improvements in root mean squared error (RMSE) and squared correlation coefficient (R²) by SVM over both ANN models indicate that the prediction accuracy of SVM is at least as good as that of those models, yet in some cases actually better, as well as forecasting of high-value discharges.     

Detection of mountain pine beetle-killed ponderosa pine in a heterogeneous landscape using high-resolution aerial imagery

Previous studies have used remote-sensing images to map tree mortality caused by mountain pine beetle (Dendroctonus ponderosae; MPB) in relatively homogeneous lodgepole pine (Pinus contorta) forests; however, classification methods have not been tested for the patchy landscape of ponderosa pine-dominated (Pinus ponderosae) montane forests characterized by highly variable tree density. This study explores two supervised classification methods to identify MPB-caused mortality (red attack) in heterogeneous montane forests of the Colorado Front Range using 1 m-resolution 2011 imagery of the National Agriculture Imagery Program (NAIP): maximum likelihood using the red, green, and blue bands, and the red-green index (RGI), and a thresholding technique using the RGI. Two variations of the RGI threshold method were also explored: the addition of a green-band threshold and the incorporation of a focal analysis. Evaluation pixels were used to assess the accuracy of the classification methods. The maximum likelihood (97 Percentage Correctly Classified (PCC); 11% error of commission for red attack) and RGI threshold (85 PCC; 46% error of commission for red attack) classification methods overestimated the red attack. The RGI and green band threshold classification reduced the error of commission (5%) and had high overall accuracy (97 PCC). In a comparison of classification methods across tree-density sites, we found the maximum likelihood classification had a very high accuracy in the high-density site (95 PCC), but substantially lower accuracy in the low-density site (85 PCC) due to the presence of more visible cover types. The RGI threshold classification with the green band constraint produced more consistent PCCs across tree densities: high (93.7 PCC), moderate (95.2 PCC), and low (92.0 PCC). Our results indicate forest structure may affect the classification accuracy and should be considered when selecting a classification method for a landscape.     

Cross-border comparison of land cover and landscape pattern in Eastern Europe using a hybrid classification technique

Eastern Europe has experienced drastic changes in political and economic conditions following the breakdown of the Soviet Union. Furthermore, these changes often differ among neighboring countries. This offers unique possibilities to assess the relative importance of broad-scale political and socioeconomic factors on land cover and landscape pattern. Our question was how much land cover differed in the Polish, the Slovak, and the Ukrainian Carpathian Mountains and to what extent these differences can be related to dissimilarities in societal, economic, and political conditions. We used a hybrid classification technique, combining advantages from supervised and unsupervised methods, to derive a land cover map from three Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) images from 2000. Results showed marked differences in land cover between the three countries. Forest cover and composition was different for the three countries, for example Slovakia and Poland had about 20% more forest cover at higher elevations than Ukraine. Broadleaved forest dominated in Slovakia while high percentages of conifers were found in Poland and Ukraine. Agriculture was most abundant in Slovakia where the lowest level of agricultural fragmentation was found (22% core area compared to less than 5% in Poland and Ukraine). Post-socialist land change was greatest in Ukraine, were we found high agricultural fragmentation and widespread early-successional shrublands indicating extensive land abandonment. Concerning forests, differences can largely be explained by socialist forest management. The abundance and pattern of arable land and grassland can be explained by two factors: land tenure in socialist times and economic transition since 1990. These results suggest that broad-scale socioeconomic and political factors are of major significance for land cover patterns in Eastern Europe, and possibly elsewhere.     

Detection of cardiac arrhythmia in electrocardiograms using adaptive feature extraction and modified support vector machines

The electrocardiogram (ECG) analysis is one of the most important approaches to cardiac arrhythmia detection. Many algorithms have been proposed, however, the recognition rate is still unsatisfactory due to unreliable feature extraction in signal characteristic analysis or poor generalization capability of the classifier. In this paper, we propose a system for cardiac arrhythmia detection in ECGs with adaptive feature selection and modified support vector machines (SVMs). Wavelet transform-based coefficients and signal amplitude/interval parameters are first enumerated as candidates, but only a few specific ones are adaptively selected for the classification of each class pair. A new classifier, which integrates k-means clustering, one-against-one SVMs, and a modified majority voting mechanism, is proposed to further improve the recognition rate for extremely similar classes. The experimental results show that the proposed ECG analysis approach can obtain a higher recognition rate than the published approaches. By testing the system with more than 100,000 samples in MIT-BIH arrhythmia database, the average recognition rate is 98.92%, and the recognition rate for each class is kept above 92%.     

Comparing the performance of fuzzy and crisp classifiers on remotely sensed images: a case of snow classification

This study deals with the evaluation of accuracy benefits offered by a fuzzy classifier as compared to hard classifiers using satellite imagery for thematic mapping applications. When a crisp classifier approach is adopted to classify moderate resolution data, the presence of mixed coverage pixels implies that the final product will have errors, either of omission or commission, which are not avoidable and are solely due to the spatial resolution of the data. Theoretically, a soft classifier is not affected by such errors, and in principle can produce a classification that is more accurate than any hard classifier. In this study we use the Pareto boundary of optimal solutions as a quantitative method to compare the performance of a fuzzy statistical classifier to the one of two hard classifiers, and to determine the highest accuracy which could be achieved by hard classifiers. As an application, the method is applied to a case of snow mapping from Moderate-Resolution Imaging Spectroradiometer (MODIS) data on two alpine sites, validated with contemporaneous fine-resolution Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data. The results for this case study showed that the soft classifier not only outperformed the two crisp classifiers, but also yielded higher accuracy than the maximum theoretical accuracy of any crisp classifier on the study areas. While providing a general assessment framework for the performance of soft classifiers, the results obtained by this inter-comparison exercise showed that soft classifiers can be an effective solution to overcome errors which are intrinsic in the classification of coarse and moderate resolution data.     

Computer‐aided classification of breast masses in mammogram images based on spherical wavelet transform and support vector machines

Breast cancer can be effectively detected and diagnosed using the technology of digital mammography. However, although this technology has been rapidly developing recently, suspicious regions cannot be detected in some cases by radiologists, because of the noise or inappropriate mammogram contrast. This study presents a classification of segmented region of interests (ROIs) as either benign or malignant to serve as a second eye of the radiologists. Our study consists of three steps. In the first step, spherical wavelet transform (SWT) is applied to the original ROIs. In the second step, shape, boundary and grey level based features of wavelet (detail) and scaling (approximation) coefficients are extracted. Finally, in the third step, malignant/benign classification of the masses is implemented by giving the feature matrices to a support vector machine system. The proposed system achieves 91.4% and 90.1% classification accuracy using the dataset acquired from the hospital of Istanbul University in Turkey and the free Mammographic Image Analysis Society, respectively. Furthermore, discrete wavelet transform, which produces 83.3% classification accuracy, is applied to the coefficients to make a comparison with the SWT method.     

Agricultural crop-type classification of multi-polarization SAR images using a hybrid entropy decomposition and support vector machine technique

This article presents the application of a hybrid classification technique of entropy decomposition and support vector machine (EDSVM) for crop-type categorization. It takes the advantage of the desired parameters from the entropy decomposition (ED) method and the statistical learning method based on the support vector machine (SVM) method that determines the optimal separation between classes in a higher dimensional feature space to improve on the existing classification results. ED is capable of extracting valuable decomposed parameters of entropy H and alpha α for image interpretation with analysis of the underlying scattering mechanisms. H demonstrates the randomness of the underlying scattering mechanisms and α is used to define the type of scattering mechanisms. However, in the application of agricultural crops where the scattering mechanisms of the crops are quite similar to each other, the distribution of the H and α in the H–α feature space overlaps from one class to another. Moreover, the drawback of ED is the arbitrariness of the boundaries for each class. To overcome this issue, SVM classifier is deployed to determine the decision boundaries by projecting the training sets of the classes into higher dimensional feature space. Hence, the hybrid EDSVM is developed to provide an alternative solution to improve the classification accuracy. In this article, EDSVM classifier is applied on a multi-crop field Airborne Synthetic Aperture Radar (AIRSAR) image of Flevoland in the Netherlands and the robustness of the classifier is evaluated. The classification is done with the purpose of separating the different types of crops with the characteristics of the scattering mechanism. At the same time, a hybrid entropy decomposition and neural network (EDNN) classifier method is developed to validate the effectiveness of the EDSVM classifier. As a result, EDSVM is proved to be robust and to yield a superior result compared with neural network (NN), SVM and EDNN classifiers.     

Recognition of communication signal types using genetic algorithm and support vector machines based on the higher order statistics

Automatic recognition of communication signal type plays an important role in various applications. Most of the existing recognizers can only identify a few types of communication signal. This paper presents a novel intelligent technique that identifies a variety of digital signal types. Here, a hierarchical support vector machine based structure is proposed as the multiclass classifier. A proper set of the higher order moments (up to eighth) and higher order cumulants (up to eighth) are proposed as the effective features for recognizing of the digital communication signal. A genetic algorithm is used for selecting the suitable parameters of support vector machines. This idea improves the performance of the recognizer, efficiently. Simulation results show that the proposed recognizer has a high success rate for recognition of the different modulations even at very low SNRs.     

Detailed mapping of a salt farm from Landsat TM imagery using neural network and maximum likelihood classifiers: a comparison

Information on the size and distribution of various zones in a salt farm is critical to salt farm management and estimation of salt yield. The ability of neural network and maximum likelihood classifiers to classify spectrally uniform water bodies with a distinct boundary in a salt farm is comparatively studied in this paper for the Taibei Salt Field, Jiangsu Province, East China using Landsat Thematic Mapper (TM) data. In a pre‐run classification of general land covers, the salt farm was mapped 84% correctly using the neural network method, slightly higher than the 76% achieved with the maximum likelihood classifier. In another separate neural network classification the salt farm was mapped further into three zones of evaporation, condensation, and crystallization at a producer's accuracy of 76%, 84%, and 86%, respectively, with the optimum classification settings. Such a detailed classification was not possible with the maximum likelihood method. It is concluded that the neural network is superior to the maximum likelihood method for detailed mapping of the Taibei Salt Field where salty water bodies are spectrally uniform and spatially extensive on the image with clear‐cut boundaries among them.     

A decision support system for the automatic management of keep-clear signs based on support vector machines and geographic information systems

This paper presents a decision support system for automatic keep-clear signs management. The system consists of several modules. First of all, an acquisition module obtains images using a vehicle equipped with two recording cameras. A recognition module, which is based on Support Vector Machines (SVMs), analyzes each image and decides if there is a keep-clear sign in it. The images with keep-clear signs are included into a Geographical Information System (GIS) database. Finally in the management module, the data in the GIS are compared with the council database in order to decide actions such as repairing or reposition of signs, detection of possible frauds etc. We present the first tests of the system in a Spanish city (Meco, Madrid), where the systems is being tested for its application in the near future.