Extreme‐learning‐machine‐based land cover classification

The extreme learning machine (ELM), a single hidden layer neural network based supervised classifier is used for remote sensing classifications. In comparison to the backpropagation neural network, which requires the setting of several user‐defined parameters and may produce local minima, the ELM requires setting of one parameter, and produces a unique solution for a set of randomly assigned weights. Two datasets, one multispectral and another hyperspectral, were used for classification. Accuracies of 89.0% and 91.1% are achieved with this classifier using multispectral and hyperspectral data, respectively. Results suggest that the ELM provides a classification accuracy comparable to a backpropagation neural network with both datasets. The computational cost using the ELM classifier (1.25 s with Enhanced Thematic Mapper (ETM+) and 0.675 s with Digital Airborne Imaging Spectrometer (DAIS) data) is very small in comparison to the backpropagation neural network.     

A WWW‐assisted fax system for Internet fax‐to‐fax communication

This paper describes a WWWassisted fax system (WAX) that is developed to provide reliable and enhanced Internet faxtofax communication. It integrates the easytouse WWW interface with conventional faxing procedures, resulting in an Internet fax system which not only circumvents the cost of long distance fax charges but also adds enhanced functionality not otherwise possible. The WAX system comprises two gateways, namely, the FaxIn and the FaxOut Gateways. The FaxIn Gateway accepts fax messages over Public Switched Telephone Network (PSTN) and stores them in a transit database. The system interfaces with the user over the WWW to provide access to his stored faxes, with the basic ability to send them out over the Internet to recipients. The FaxOut Gateway receives fax files from the FaxIn Gateway through the Internet and transmits them out to the intended recipients via the local PSTN. WAX users do not require any additional hardware except for a fax machine and a personal computer with Internet connectivity to gain access to WAX via any WWW browser. In addition, WAX provides a host of other enhanced features such as the ability to construct minifaxes from a single incoming fax as well as dynamically attach cover notes to outgoing faxes.     

Issues in modified problem‐based learning: A self‐study in pre‐service science‐teacher education

Abstract:

In this self‐study, the author gained in‐depth understanding of how to plan and implement problem‐based learning (PBL), a student‐centred approach to teaching and learning that is driven by messy, open‐ended problems. This paper focuses primarily on the issues and concerns that arose as she developed and implemented a modified form of traditional PBL (Barrows, 1996) in large, pre‐service science‐teacher education classes. To view the research from many perspectives, a variety of data collection methods and sources were used, including field notes, semi‐structured interviews, student‐generated documents, and student journals. The outcomes of this study describe challenges (problem development, facilitation of groups, and assessment) encountered by the author as she planned for and implemented PBL. Furthermore, changes in the author's classroom practice, the connection between these changes and constructivist learning principles, and implications for science‐teacher education are addressed.     

Cross‐calibration of MSG1‐SEVIRI infrared channels with Terra‐MODIS channels

This paper addresses the cross‐calibration of the infrared channels 4 (3.9 µm), 9 (10.8 µm) and 10 (12.0 µm) of the Spinning Enhanced Visible and Infra‐Red Imager (SEVIRI) onboard the Meteosat Second Generation 1 (MSG1) satellite with the channels of the MODerate resolution Imaging Spectroradiometer (MODIS) onboard Terra. The cross‐calibrations, including the Ray‐Matching (RM) method and the Radiative Transfer Modelling (RTM) method, were developed and implemented over a tropical area using SEVIRI and MODIS measurements of July 2005 and July 2006 with absolute view zenith angle differences (|ΔVZA|)<0.5°, absolute view azimuth angle differences (|ΔVAA|)<0.5° and absolute time differences (|ΔTime|)<10 min. The results obtained by the RM and RTM methods revealed calibration discrepancies between the two sensors. The results obtained by the RM method were consistent with previously published results. The results obtained by the RTM method were consistent with the results obtained by the RM method if the temperature differences caused by the spectral differences between the two sensors were taken into account. From the cross‐calibration results obtained by the two methods, the use of the results obtained by the RTM method to recalibrate the SEVIRI data is recommended. The recalibrations remove the overestimation of the Land Surface Temperature (LST) retrieved from the SEVIRI data by a split‐window method.     

Analysis of co‐occurrence and discrete wavelet transform textures for differentiation of forest and non‐forest vegetation in very‐high‐resolution optical‐sensor imagery

The extraction of texture features from high‐resolution remote sensing imagery provides a complementary source of data for those applications in which the spectral information is not sufficient for identification or classification of spectrally similar landscape features. This study presents the results of grey‐level co‐occurrence matrix (GLCM) and wavelet transform (WT) texture analysis for forest and non‐forest vegetation types differentiation in QuickBird imagery. Using semivariogram fitting, the optimal GLCM windows for the land cover classes within the scene were determined. These optimal window sizes were then applied to eight GLCM texture measures (mean, variance, homogeneity, dissimilarity, contrast, entropy, angular second moment, and correlation) for the scene classification. Using wavelet transformation, up to five levels of macro‐texture were computed and tested in the classification process. Comparing the classification results, (1) the spectral‐only bands classification gave an overall accuracy of 58.69%; (2) the statistically derived 21×21 optimal mean texture combined with spectral information gave the best results among the GLCM optimal windows with an accuracy of 73.70%; and (3) the combined optimal WT‐texture levels 4 and 5 gave an accuracy of 63.56%. The combined classification of these three optimal results gave an overall accuracy of 77.93%. The results indicate that even though vegetation texture was generally measured better by the GLCM‐mean texture (micro‐textures) than by WT‐derived texture (macro‐textures), the results show that the micro–macro texture combination would improve the differentiation and classification of the overall vegetation types. Overall, the results suggests that computer‐assisted classification of high‐spatial‐resolution remotely sensed imagery has a good potential to augment the present ground‐based forest inventory methods.     

Phase‐correlation‐based hyperspectral image classification using multiple class representatives obtained with k‐means clustering

In this letter, a modification to a phase–correlation‐(PC‐)based supervised classification method for hyperspectral data is proposed. An adaptive approach using different numbers of multiple class representatives (CRs) extracted using PC‐based k‐means clustering for each class is compared with the use of selecting a small, pre‐determined number of dissimilar CRs. PC is used as a distance measure in k‐means clustering to determine the spectral similarity between each pixel and cluster centre. The number of representatives for each class is chosen adaptively, depending on the number of training samples in each class. Classification is performed for each pixel according to the maximum value of PCs obtained between test samples and the CRs. Experimental results show that the adaptive method gave the highest classification accuracy (CA). Experiments on the effect of reducing the size of the feature vectors found that CA increased as the feature vector decreased.     

Corn‐yield estimation through assimilation of remotely sensed data into the CSM‐CERES‐Maize model

One of the applications of crop simulation models is to estimate crop yield during the current growing season. Several studies have tried to integrate crop simulation models with remotely sensed data through data‐assimilation methods. This approach has the advantage of allowing reinitialization of model parameters with remotely sensed observations to improve model performance. In this study, the Cropping System Model‐CERES‐Maize was integrated with the Moderate Resolution Imaging Spectroradiometer (MODIS) leaf area index (LAI) products for estimating corn yield in the state of Indiana, USA. This procedure, inversion of crop simulation model, facilitates several different user input modes and outputs a series of agronomic and biophysical parameters, including crop yield. The estimated corn yield in 2000 compared reasonably well with the US Department of Agriculture National Agricultural Statistics Service statistics for most counties. Using the seasonal LAI in the optimization procedure produced the best results compared with only the green‐up LAIs or the highest LAI values. Planting, emergence and maturation dates, and N fertilizer application rates were also estimated at a regional level. Further studies will include investigating model uncertainties and using other MODIS products, such as the enhanced vegetation index.     

RVM‐based multi‐class classification of remotely sensed data

The relevance vector machine (RVM), a Bayesian extension of the support vector machine (SVM), has considerable potential for the analysis of remotely sensed data. Here, the RVM is introduced and used to derive a multi‐class classification of land cover with an accuracy of 91.25%, a level comparable to that achieved by a suite of popular image classifiers including the SVM. Critically, however, the output of the RVM includes an estimate of the posterior probability of class membership. This output may be used to illustrate the uncertainty of the class allocations on a per‐case basis and help to identify possible routes to further enhance classification accuracy.     

Artificial immune‐based supervised classifier for land‐cover classification

This paper explores the potential of an artificial immune‐based supervised classification algorithm for land‐cover classification. This classifier is inspired by the human immune system and possesses properties similar to nonlinear classification, self/non‐self identification, and negative selection. Landsat ETM+ data of an area lying in Eastern England near the town of Littleport are used to study the performance of the artificial immune‐based classifier. A univariate decision tree and maximum likelihood classifier were used to compare its performance in terms of classification accuracy and computational cost. Results suggest that the artificial immune‐based classifier works well in comparison with the maximum likelihood and the decision‐tree classifiers in terms of classification accuracy. The computational cost using artificial immune based classifier is more than the decision tree but less than the maximum likelihood classifier. Another data set from an area in Spain is also used to compare the performance of immune based supervised classifier with maximum likelihood and decision‐tree classification algorithms. Results suggest an improved performance with the immune‐based classifier in terms of classification accuracy with this data set, too. The design of an artificial immune‐based supervised classifier requires several user‐defined parameters to be set, so this work is extended to study the effect of varying the values of six parameters on classification accuracy. Finally, a comparison with a backpropagation neural network suggests that the neural network classifier provides higher classification accuracies with both data sets, but the results are not statistically significant.     

Phase correlation pixel‐to‐pixel image co‐registration based on optical flow and median shift propagation

With singular value decomposition (SVD) and robust 2‐dimensional fitting phase correlation algorithms, it is possible to achieve pixel‐to‐pixel image co‐registration at sub‐pixel accuracy via local feature matching. However, the method often fails in featureless and low correlation areas making it not robust for co‐registration of images with considerable spectral differences and large featureless ground objects. A median shift propagation (MSP) technique is proposed to eliminate the problem, in a phase correlation and Normalized Cross‐Correlation (NCC) combined approach. The experiment results using images from different sensor platforms and spectral bands indicate that the new method is very robust to featureless and low correlation areas and can achieve very accurate pixel‐to‐pixel image co‐registration with good tolerance of spectral and spatial differences between images. The method will significantly improve change detection in various remote sensing applications.     

Small‐footprint,waveform‐resolving lidar estimation of submerged and sub‐canopy topography in coastal environments

The experimental advanced airborne research lidar (EAARL) is an airborne lidar instrument designed to map near‐shore submerged topography and adjacent land elevations simultaneously. This study evaluated data acquired by the EAARL system in February 2003 and March 2004 along the margins of Tampa Bay, Florida, USA, to map bare‐earth elevations under a variety of vegetation types and submerged topography in shallow, turbid water conditions. A spatial filtering algorithm, known as the iterative random consensus filter (IRCF), was used to extract ground elevations from a point cloud of processed last‐surface EAARL returns. Filtered data were compared with acoustic and field measurements acquired in shallow submerged (0–2.5 m water depth) and sub‐canopy environments. Root mean square elevation errors (RMSEs) ranged from 10–14 cm for submerged topography to 16–20 cm for sub‐canopy topography under a variety of vegetation communities. The effect of lidar sampling angles and global positioning system (GPS) satellite configuration on accuracy was investigated. Results show high RMSEs for data acquired during periods of poor satellite configuration and at large sampling angles along the edges of the lidar scan. The results presented in this study confirm the cross‐environment capability of a green‐wavelength, waveform‐resolving lidar system, making it an ideal tool for mapping coastal environments.     

Fire‐induced variability in satellite SPOT‐VGT NDVI vegetational data

Is it possible to quantitatively characterize the dynamic variability induced by fires on vegetation? To answer this question, we performed a fluctuation analysis of time series of satellite SPOT‐VEGETATION Normalized Difference Vegetation Index (NDVI) data from 1998 to 2003 in order to discriminate fire‐induced variability in the vegetation dynamics of forests in southern Italy. We used the Detrended Fluctuation Analysis (DFA), which permits the detection of persistent properties in non‐stationary signal fluctuations. We analysed two forest sites, one fire‐unaffected and the other fire‐affected. Our findings suggest that fires play an important role in the temporal evolution of forests, increasing the persistence of the vegetation dynamics.     

Estimating area errors for fine‐scale feature‐based ecological mapping

High spatial resolution feature‐based approaches are especially useful for ecological mapping in densely populated landscapes. This paper evaluates errors in estimating ecological map class areas from fine‐scale current (～2002) and historical (～1945) feature‐based ecological mapping by a set of trained interpreters across densely populated rural sites in China based on field‐validated interpretation of high spatial resolution (1 m) imagery. Median overall map accuracy, corrected for chance, was greater than 85% for mapping by trained interpreters, with greater accuracy for current versus historical mapping. An error model based on feature perimeter proved as reliable in predicting 90% confidence intervals for map class areas as did models derived from the conventional error matrix. A conservative error model combining these approaches was developed and tested for statistical reliability in predicting confidence intervals for ecological map class areas from fine‐scale feature‐based mapping by a set of trained interpreters across rural China, providing a practical basis for statistically reliable ecological change detection in densely populated landscapes.     

E‐Learning with visual math: An E‐book review

Yerushalmy, M., Katriel, H., & Sternberg, B. Visual Math: Functions. Available: http://www.cet.ac.il/math/function/english/ (accessed 26 July 2005).     

Using low‐spectral‐resolution images to acquire simulated hyperspectral images

We propose a method to acquire simulated hyperspectral images using low‐spectral‐resolution images. Hyperspectral images provide more spectral information than low‐spectral‐resolution images, because of the additional spectral bands used for data acquisition in hyperspectral imaging. Unfortunately, original hyperspectral images are more expensive and more difficult to acquire. However, some research questions require an abundance of spectral information for ground monitoring, which original hyperspectral images can easily provide. Hence, we need to propose a method to acquire simulated hyperspectral images, when original hyperspectral images are especially necessary. Since low‐spectral‐resolution images are readily available and cheaper, we develop a method to acquire simulated hyperspectral images using low‐spectral‐resolution images. With simulated hyperspectral images, we can acquire more ‘hidden’ information from low‐spectral‐resolution images. Our method uses the principles of pixel‐mixing to understand the compositional relationship of spectrum data to an image pixel, and to simulate radiation transmission processes. To this end, we use previously obtained data (i.e. spectrum library) and the sorting data of objects that are derived from a low‐spectral‐resolution image. Using the simulation of radiation transmission processes and these different data, we acquire simulated hyperspectral images. In addition, previous analyses of simulated remotely sensed images do not use quantitative statistical measures, but use qualitative methods, describing simulated images by sight. Here, we quantitatively assess our simulation by comparing the correlation coefficients of simulated images and real images. Finally, we use simulated hyperspectral images, real Hyperion images, and their corresponding ALI images to generate several classification images. The classification results demonstrate that simulated hyperspectral data contain additional information not available in the multispectral data. We find that our method can acquire simulated hyperspectral images quickly.     

A Web‐based multi‐agent system for interpreting medical images

A difficult problem in medical image interpretation is that for every image type such as xray and every body organ such as heart, there exist specific solutions that do not allow for generalization. Just collecting all the specific solutions will not achieve the vision of a computerized physician. To address this problem, we develop an intelligent agent approach based on the concept of active fusion and agentoriented programming. The advantage of agentoriented programming is that it combines the benefits of objectoriented programming and expert system. Following this approach, we develop a Webbased multiagent system for interpreting medical images. The system is composed of two major types of intelligent agents: radiologist agents and patient representative agents. A radiologist agent decomposes the image interpretation task into smaller subtasks, uses multiple agents to solve the subtasks, and combines the solutions to the subtasks intelligently to solve the image interpretation problem. A patient representative agent takes questions from the user (usually a patient) through a Webbased interface, asks for multiple opinions from radiologist agents in interpreting a given set of images, and then integrates the opinions for the user. In addition, a patient representative agent can answer questions based on the information in a medical information database. To maximize the satisfaction that patients receive, the patient representative agents must be as informative and timely as communicating with a human. With an efficient pseudonatural language processing, a knowledge base in XML, and user communication through Microsoft Agent, the patient representative agents can answer questions effectively.     

Assessment of stand‐wise stem volume retrieval in boreal forest from JERS‐1 L‐band SAR backscatter

JERS‐1 L‐band SAR backscatter from test sites in Sweden, Finland and Siberia has been investigated to determine the accuracy level achievable in the boreal zone for stand‐wise forest stem volume retrieval using a model‐based approach. The extensive ground‐data and SAR imagery datasets available allowed analysis of the backscatter temporal dynamics. In dense forests the backscatter primarily depended on the frozen/unfrozen state of the canopy, showing a ～4 dB difference. In sparse forests, the backscatter depended primarily on the dielectric properties of the forest floor, showing smaller differences throughout the year. Backscatter modelling as a function of stem volume was carried out by means of a simple L‐band Water Cloud related scattering model. At each test site, the model fitted the measurements used for training irrespective of the weather conditions. Of the three a priori unknown model parameters, the forest transmissivity coefficient was most affected by seasonal conditions and test site specific features (stand structure, forest management, etc.). Several factors determined the coefficient's estimate, namely weather conditions at acquisition, structural heterogeneities of the forest stands within a test site, forest management practice and ground data accuracy. Stem volume retrieval was strongly influenced by these factors. It performed best under unfrozen conditions and results were temporally consistent. Multi‐temporal combination of single‐image estimates eliminated outliers and slightly decreased the estimation error. Retrieved and measured stem volumes were in good agreement up to maximum levels in Sweden and Finland. For the intensively managed test site in Sweden a 25% relative rms error was obtained. Higher errors were achieved in the larger and more heterogeneous forest test sites in Siberia. Hence, L‐band backscatter can be considered a good candidate for stand‐wise stem volume retrieval in boreal forest, although the forest site conditions play a fundamental role for the final accuracy.     

Indicators of Northern Eurasia's land‐cover change trends from SPOT‐VEGETATION time‐series analysis 1998–2005

The Boreal and Tundra ecosystems of the mid to high latitudes are sensitive indicators of environmental impacts from both climate change and direct human activities. This study uses inter‐annual and inter‐seasonal SPOT‐VGT mosaics for recent years from 1998 to 2005 covering the entire boreal ecosystems of northern Eurasia. Linear trends could be detected in the NDVI and NDWI time series that differ by season, land‐cover type, and latitude. Significant positive NDVI trends are described for spring and related negative trends for NDWI over the boreal forest zone. They indicate an earlier onset of the vegetation green‐up. Similar vegetation dynamics can be described for autumn. The tundra ecosystems of the northern Eurasia latitudes exhibit trends of negative NDVI and positive NDWI. This may be explained by earlier snowmelt and increasing amounts of surface water from positive temperature anomalies. The non‐ambiguous coarse‐scale indicators require further detailed studies to identify driving factors and amount for positive feedbacks in boreal ecosystems.     

Feature‐selection ability of the decision‐tree algorithm and the impact of feature‐selection/extraction on decision‐tree results based on hyperspectral data

The decision‐tree (DT) algorithm is a very popular and efficient data‐mining technique. It is non‐parametric and computationally fast. Besides forming interpretable classification rules, it can select features on its own. In this article, the feature‐selection ability of DT and the impacts of feature‐selection/extraction on DT with different training sample sizes were studied by using AVIRIS hyperspcetral data. DT was compared with three other feature‐selection methods; the results indicated that DT was an unstable feature selector, and the number of features selected by DT was strongly related to the sample size. Trees derived with and without feature‐selection/extraction were compared. It was demonstrated that the impacts of feature selection on DT were shown mainly as a significant increase in the number of tree nodes (14.13–23.81%) and moderate increase in tree accuracy (3.5–4.8%). Feature extraction, like Non‐parametric Weighted Feature Extraction (NWFE) and Decision Boundary Feature Extraction (DBFE), could enhance tree accuracy more obviously (4.78–6.15%) and meanwhile a decrease in the number of tree nodes (6.89–16.81%). When the training sample size was small, feature‐selection/extraction could increase the accuracy more dramatically (6.90–15.66%) without increasing tree nodes.     

First algorithm for chlorophyll‐a retrieval from MODIS‐Terra imagery of Sun‐induced fluorescence in the Southern Ocean

The measurements of in situ samplers, the ENEA Light Detection and Ranging (Lidar) Fluorosensor (ELF) and Moderate Resolution Imaging Spectroradiometer on‐board the Terra satellite (MODIS‐Terra), carried out in the Southern Ocean during the Austral summer 2002–2003, were used to provide the first algorithm for chlorophyll‐a (Chl‐a) retrieval from MODIS‐Terra imagery of Sun‐induced fluorescence in the Southern Ocean. The results of the algorithm indicate that the standard MODIS‐Terra algorithm underestimated Chl‐a. The discrepancy (20%) is below the expected error of MODIS (35%).