One-Class Classification for Mapping a Specific Land-Cover Class: SVDD Classification of Fenland

Carolina Sanchez-Hernandez,Giles M Foody,Doreen S Boyd

doi:10.1109/tgrs.2006.890414

Carolina Sanchez-Hernandez, Giles M Foody + Show 1 more

Open Access

https://doi.org/10.1109/tgrs.2006.890414

Copy DOI

Abstract

Remote sensing is a major source of land-cover information. Commonly, interest focuses on a single land-cover class. Although a conventional multiclass classifier may be used to provide a map depicting the class of interest, the analysis is not focused on that class and may be suboptimal in terms of the accuracy of its classification. With a conventional classifier, considerable effort is directed on the classes that are not of interest. Here, it is suggested that a one-class-classification approach could be appropriate when interest focuses on a specific class. This is illustrated with the classification of fenland, a habitat of considerable conservation value, from Landsat Enhanced Thematic Mapper Plus imagery. A range of one-class classifiers is evaluated, but attention focuses on the support-vector data description (SVDD). The SVDD was used to classify fenland with an accuracy of 97.5% and 93.6% from the user's and producer's perspectives, respectively. This classification was trained upon only the fenland class and was substantially more accurate in fen classification than a conventional multiclass maximum-likelihood classification provided with the same amount of training data, which classified fen with an accuracy of 90.0% and 72.0% from the user's and producer's perspectives, respectively. The results highlight the ability to classify a single class using only training data for that class. With a one-class classification, the analysis focuses tightly on the class of interest, with resources and effort not directed on other classes, and there are opportunities to derive highly accurate classifications from small training sets

Highlights

Given that the remotely sensed response is predominantly a function of Earth surface properties, remote sensing has great potential as a source of information on land cover
This paper proposes the adoption of such a one-class classification approach based on the principles of the support vector machine (SVM) for accurate classification of a single class of interest from remotely sensed data
The results confirmed the potential of the support vector data description (SVDD) for one class classification of fenland

Summary

Introduction

Given that the remotely sensed response is predominantly a function of Earth surface properties, remote sensing has great potential as a source of information on land cover. Attention has focused on deriving information on specific crop types since pioneering major programmes such as LACIE which exploited imagery from early satellite remote sensing systems [8] It is the case in many ecological studies where attention is focused perhaps on an invasive species [9,10] or a rare habitat for conservation monitoring [11]. One concern is that the correct execution of a conventional supervised classifier for the derivation of the desired information typically requires the analyst to train the classifier on every class that occurs in the study area, even if most are of no interest to the analysis, in order to satisfy the underlying assumption of an exhaustively defined set of classes. Attention is focused on a straightforward approach to classifying a specific class of interest that requires training in a manner similar to conventional supervised classifications but only training data for the class of interest.

One class classification

One-class classification by SVDD

Study Site

Data and Methods

Results and Discussion

Summary and Conclusions