Optimising the remote sensing of Mediterranean land cover
Optimising the remote sensing of Mediterranean land cover
The aim of this thesis was to develop an effective procedure (by means of maximising the percentage accuracy) for classifying Mediterranean land cover with remotely sensed data. Combinations of a maximum likelihood classifier (ML), an artificial neural network (ANN) and texture analysis, on both a per-pixel and per-field basis, were used to classify land cover using three Landsat Thematic Mapper images of Cukurova, Turkey.
This study was designed to integrate spectral and spatial information. The spatial (textural) information was in the form of the standard deviation, variance, geostatistical measures of texture and the co-occurrence matrix. The dominant spatial units (field boundary information) were utilised in two ways: (i) after classification ('per field majority rule') and (ii) prior to classification ('per-field').
In most of the cases, the accuracy of the ANN was greater than that of the ML when using spectral data alone and when using spectral and textural data together. However, when the classes were spectrally distinct ML provided more accurate results than ANN. The use of texture measures through the per-pixel and per-field majority rule approaches, were found to reduce the classification accuracy because the field boundaries were enlarged and so overwhelmed the measures of texture. In contrast, the per-field approach (where the field was specified prior to analysis) combined with texture information, markedly increased the classification accuracy. However, the accuracy decreased as the variogram lag increased and the accuracy of the resultant semivariance estimates decreased. The accuracy with which the land cover of the study area could be classified was maximised at 89% using a per-field, ANN approach in which semivariance at a lag of 1 pixel was incorporated as textural information. This was 15% greater than the accuracy achieved using a standard per-pixel ML classification. This research also demonstrated that geostatistical measures of spatial variability contribute to more accurate classifications of Mediterranean land cover than do statistics derived from the co-occurrence matrix in a per-field format. The primary limitation of the use of the per-field approach was noted to be the need for prior knowledge of field boundaries, which may be resolved by using existing data or through some form of edge-detection routine.
University of Southampton
1999
Berberoǧlu, Süha
(1999)
Optimising the remote sensing of Mediterranean land cover.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
The aim of this thesis was to develop an effective procedure (by means of maximising the percentage accuracy) for classifying Mediterranean land cover with remotely sensed data. Combinations of a maximum likelihood classifier (ML), an artificial neural network (ANN) and texture analysis, on both a per-pixel and per-field basis, were used to classify land cover using three Landsat Thematic Mapper images of Cukurova, Turkey.
This study was designed to integrate spectral and spatial information. The spatial (textural) information was in the form of the standard deviation, variance, geostatistical measures of texture and the co-occurrence matrix. The dominant spatial units (field boundary information) were utilised in two ways: (i) after classification ('per field majority rule') and (ii) prior to classification ('per-field').
In most of the cases, the accuracy of the ANN was greater than that of the ML when using spectral data alone and when using spectral and textural data together. However, when the classes were spectrally distinct ML provided more accurate results than ANN. The use of texture measures through the per-pixel and per-field majority rule approaches, were found to reduce the classification accuracy because the field boundaries were enlarged and so overwhelmed the measures of texture. In contrast, the per-field approach (where the field was specified prior to analysis) combined with texture information, markedly increased the classification accuracy. However, the accuracy decreased as the variogram lag increased and the accuracy of the resultant semivariance estimates decreased. The accuracy with which the land cover of the study area could be classified was maximised at 89% using a per-field, ANN approach in which semivariance at a lag of 1 pixel was incorporated as textural information. This was 15% greater than the accuracy achieved using a standard per-pixel ML classification. This research also demonstrated that geostatistical measures of spatial variability contribute to more accurate classifications of Mediterranean land cover than do statistics derived from the co-occurrence matrix in a per-field format. The primary limitation of the use of the per-field approach was noted to be the need for prior knowledge of field boundaries, which may be resolved by using existing data or through some form of edge-detection routine.
This record has no associated files available for download.
More information
Published date: 1999
Identifiers
Local EPrints ID: 463602
URI: http://eprints.soton.ac.uk/id/eprint/463602
PURE UUID: c5d54203-2176-40e3-9c1a-835e9d3a857e
Catalogue record
Date deposited: 04 Jul 2022 20:54
Last modified: 04 Jul 2022 20:54
Export record
Contributors
Author:
Süha Berberoǧlu
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics