Multi-Source Image Classification II: An Empirical Comparison of Evidential Reasoning and Neural Network Approaches

Peddle, Derek R.; Foody, Giles M.; Airong, Zhang; Franklin, Steven E.; LeDrew, E.

doi:10.1080/07038992.1994.10874582

Cited by 61 publications

(37 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The parameters of elevation, slope and aspect are strongly linked to vegetation cover type, abundance and diversity. Peddle (1993) and Peddle et al (1994) provide good comparisons of land cover classification using various combinations of spectral, textural and DEM derivatives. Other types of information that may improve forest mapping and modelling include soils, drainage, and climate data.…”

Section: Integration Of Other Types Of Geographic Informationmentioning

confidence: 99%

Airborne remote sensing in forestry: Sensors, analysis and applications

King¹

2000

The Forestry Chronicle

View full text Add to dashboard Cite

This paper discusses the aspects of airborne remote sensing that are critical to forestry applications, the imaging characteristics of the most common sensors currently available, and analytical techniques that make use of the great amount of information content in airborne imagery. As the first paper in the CIF technical meeting to which this issue of the Forestry Chronicle is devoted, the paper is intended to provide an overview and context for subsequent papers and not a presentation of specific research methods or results.Key words: airborne remote sensing, forestry, photography, digital cameras, hyperspectral sensors, radar, laser remote sensing, image analysis Cet article discute des diffkrents aspects de la tklkdktection akroportCe qui touchent particulikrement son utilisation dans les sciences forestikres, des caractCristiques des images produites par les senseurs couramment disponibles, et des techniques d'analyse qui permettent de tirer parti de la grande quantitk d'information contenue dans ces images. Cet article, le tout premier de la confkrence technique de 1'IFC rapportCe dans ce numkro, a pour but de prksenter un cadre gCnCral et une vue d'ensemble du sujet, plutdt que des mCthodes de recherche ou des conclusions.Mots-clCs: tklCdktection akroportCe, foresterie, photographie, camkras numeriques, capteurs hyperspectrals, radar, tClCdCtection laser, analyse d'image IntroductionIn forestry, traditional information needs to manage the resource for economic gain now represent just one component in a complex set of applications that increasingly have biological and ecological significance at a variety of scales. Reporting to international agencies is required on the quantity and status of our forests. Provincial agencies require diverse information in order to balance resource extraction and employment goals with environmental and multiple use goals. Forest companies and other stakeholders require local, site-specific information for cost-effective management of their investments. With increasing demands on forest resources, the types of information required to improve our understanding and management of these resources have multiplied in quantity and diversity. Forest management must therefore be interdisciplinary and multi-scale. Information derived from field studies, airborne remote sensing and satellite remote sensing must be integrated for a wide range of applications and users.The purpose of this paper is to present the current status of airborne remote sensing capabilities in forestry. Emphasis is placed on remote sensing issues and techniques but links will be made with forest management applications. The paper is divided into three primary sections: 1) an overview of critical aspects of remote sensing that are commonly integrated in applications development in order to optimally match imaging -and analysis technologies to information needs, 2) airborne sensors that are pertinent to forestry, and 3) a summary of new techniques for forest information extraction from airborne images. Each of thes...

show abstract

Section: Integration Of Other Types Of Geographic Informationmentioning

confidence: 99%

Airborne remote sensing in forestry: Sensors, analysis and applications

King¹

2000

The Forestry Chronicle

View full text Add to dashboard Cite

show abstract

“…Although the accuracy with which land cover may be classified by these techniques has often been found to be higher than that derived from the conventional statistical classifiers [e.g. [11][12][13][14] there is still considerable scope for further increases in accuracy to be obtained and a strong desire to maximise the degree of land cover information extraction from remotely sensed data. Thus, research into new methods of classification has continued and support vector machines (SVM) have recently attracted the attention of the remote sensing community [15][16][17].…”

Section: Introductionmentioning

confidence: 99%

A relative evaluation of multiclass image classification by support vector machines

Foody

Mathur

2004

IEEE Trans. Geosci. Remote Sensing

863

394

View full text Add to dashboard Cite

Support vector machines (SVM) have considerable potential as classifiers of remotely sensed data. A constraint on their application in remote sensing has been their binary nature, requiring multi-class classifications to be based upon a large number of binary analyses. Here, an approach for multi-class classification of airborne sensor data by a single SVM analysis is evaluated against a series of classifiers that are widely used in remote sensing, with particular regard to the effect of training set size on classification accuracy. In addition to the SVM, the same data sets were classified using a discriminant analysis, decision tree and multilayer perceptron neural network. The accuracy statements of the classifications derived from the different classifiers were compared in a statistically rigorous fashion that accommodated for the related nature of the samples used in the analyses. For each classification technique, accuracy was positively related with the size of the training set. In general, the most accurate classifications were derived from the SVM approach, and with the largest training set the SVM classification was significantly (p<0.05) more accurate (93.75%) than that derived from the discriminant analysis (90.00%) and decision tree algorithms (90.31%). Although each classifier could yield a very accurate classification, >90% correct, the classifiers differed in the ability to correctly label individual cases and so may be suitable candidates for an ensemble based approach to classification.3

show abstract

“…This has included the use of a range of no n-parametric classifiers, with approaches based on evidential reasoning or neural networks proving popular (Srinivasan and Richards, 1990;Benediktsson et al, 1990;Fischer et al, 1997). Comparative studies using a suite of classification methods have shown repeatedly that for the majority of imagery and landscapes, neural networks can provide the most accurate classification of land cover (Benediktsson et al, 1990;Peddle et al, 1994;Paola and Schowengerdt, 1995). From the range of network types, feedforward networks such as the multi-layer perceptron are now the most used in mapping land cover type from remotely sensed data ( Figure 5).…”

Section: Supervised Classificationmentioning

confidence: 99%

“…For example, ancillary data on altitude or soil type may help discriminate vegetation communities that have a similar appearance in the imagery but are known to be located in different environments. The integration of ancillary information into the analysis has often been found to increase the accuracy of a variety of classifications (Strahler, 1980;Hutchinson, 1982;Peddle et al, 1994). Despite these refinements to the techniques available no classification is ideal, this is because there remain several fundamental problems with classification as a tool for land cover mapping (Foody, 1999;Mather, 1999b).…”

mentioning

confidence: 99%

Linking remote sensing, land cover and disease

Curran

Atkinson

Foody

et al. 2000

Remote Sensing and Geographical Information Systems in Epidemiology

View full text Add to dashboard Cite

References List of Tables List of Figures 3 AbstractLand cover is a critical variable in epidemiology and can be characterised remotely. A framework is used to describe both the links between land cover and radiation recorded by a remotely sensed image and the links between land cover and the disease carried by vectors. The framework is then used to explore the issues involved when moving from remotely sensed imagery to land cover and then to vector density/disease risk.This exploration highlights the rôle of land cover; the need to develop a sound knowledge of each link in the predictive sequence; the problematic mismatch between the spatial units of the remotely sensed and epidemiological data and the challenges and opportunities posed by adding a temporal mismatch between the remotely sensed and epidemiological data. The paper concludes with a call for both greater understanding of the physical components of the proposed framework and the utilisation of optimised statistical tools as prerequisites to progress in this field.4 IntroductionThe application of remote sensing to epidemiology is based on the development of a logical sequence that links measures of radiation made by a sensor on board an aircraft, or more usually a satellite, to measures of a disease and its vector (e.g ., Crombie et al., 1999). At its most general, a sequence could be: (i) remotely sensed data can be used to provide information on land cover (e.g., different vegetation types or classes of vegetation amount) and thereby habitat (Innes and Koch, 1998), (ii) the spatial distribution of vector-borne disease z is related to the habitat of that vector (Pavlovsky, 1966) and (iii) therefore, remotely sensed data can be used to provide information on the spatial distribution of vector-borne disease z (Hay et al., 1997).In certain circumstances the first two propositions can be well-founded. For example, where remotely sensed data and disease data are both related to climate (Hugh-Jones, 1991a;Thomson et al., 1996;Hay et al., 1996;1998a). When this is the case researchers have sought to use remote sensing as a direct and instrumental tool (Curran, 1987) to predict and map the location of some of the major diseases affecting human health (Bailey and Linthicum, 1989;Hay et al., 1997;1998b;Malone et al., 1997;Connor, 1999)."Satellite (sensor) images can pinpoint the breeding grounds of the mosquitoes that cause malaria, pick out the tsetse fly's favourite haunts and perhaps even identify places where there is a risk of cholera" (K. Kleiner, 1995, p.9).In support of such predictions research has been undertaken in different environments, for different diseases and vectors and with various combinations of imagery and ancillary data (Hugh-Jones and O'Neil, 1986;Hay et al., 1997;Estrada-Peña, 1998). Some thirty years of experience (Cline, 1970) have shown that while the remote sensing of disease is certainly viable (Linthicum et al., 1987;Hugh-Jones, 1991a;Rogers and Williams, 1993) it will not be a robust and reliable epidemiological technique unt...

show abstract

Multi-Source Image Classification II: An Empirical Comparison of Evidential Reasoning and Neural Network Approaches

Cited by 61 publications

References 28 publications

Airborne remote sensing in forestry: Sensors, analysis and applications

Airborne remote sensing in forestry: Sensors, analysis and applications

A relative evaluation of multiclass image classification by support vector machines

Linking remote sensing, land cover and disease

Contact Info

Product

Resources

About