D. Rokos

A texture-based classification method for classifying built areas according to their density

Quantitative information about building density can serve as a useful tool for urban applications, such as the study of urban land changes, illegal building development, urban expansion, etc. This paper presents a method that classifies built areas according to their density into three categories using SPOT panchromatic remote sensing images. It is based on the concept of statistical measurement of texture. Specifically, three algorithms, which employ occurrence frequency or co-ocurrence matrices concepts on binary data, were developed, tested over the broader Athens area in Attica and evaluated. The developed algorithms were equally effective when windows larger than 31 x 31 pixels were used. For such windows, the overall accuracy of the method ranges from 83.40 to 89.61%. These results are better than the 79.70% that was obtained using the maximum likelihood classifier. The kappahat coefficient was also improved by about 0.16 units, while the accuracy of the classification for each class of built areas was improved by about 50-60%.

An object‐oriented methodology to detect oil spills

A new automated methodology for oil spill detection is presented, by which full synthetic aperture radar (SAR) high‐resolution image scenes can be processed. The methodology relies on the object‐oriented approach and profits from image segmentation techniques to detected dark formations. The detection of dark formations is based on a threshold definition that is fully adaptive to local contrast and brightness of large image segments. For the detection process, two empirical formulas are developed that also permit the classification of oil spills according to their brightness. A fuzzy classification method is used to classify dark formations as oil spills or look‐alikes. Dark formations are not isolated and features of both dark areas and sea environment are considered. Various sea environments that affect oil spill shape and boundaries are grouped in two knowledge bases, used for the classification of dark formations. The accuracy of the method for the 12 SAR images used is 99.5% for the class of oil spills, and 98.8% for that of look‐alikes. Fresh oil spills, fresh spills affected by natural phenomena, oil spills without clear stripping, small linear oil spills, oil spills with broken parts and amorphous oil spills can be successfully detected.

Dark formation detection using neural networks

Synthetic Aperture Radar (SAR) images are extensively used for dark formation detection in marine environment, as they are not affected by local weather conditions and cloudiness. Dark formations can be caused by man‐made actions (e.g. oil spills) or natural ocean phenomena (e.g. natural slicks and wind front areas). Radar backscatter values for oil spills are very similar to backscatter values for very calm sea areas and other ocean phenomena because they dampen the capillary and short gravity sea waves. Thus, traditionally, dark formation detection is the first stage of the oil‐spill detection procedure and in most studies is performed manually or using a fixed size window in which a threshold value is adopted. In high‐resolution imagery, dark formation detection may fail due to the nonlinear behaviour of the pixel values contained in the dark formation and in the area around it. In this paper, we examine the ability of two feed‐forward neural network families, i.e. Multilayer Perceptron (MLP) and the Radial Basis Function (RBF) networks, to detect dark formations in high‐resolution SAR images. The general objective of this paper is to test the potential of artificial neural networks for dark formation detection using SAR high‐resolution satellite images. Both the type and the architecture of the network are subjects of research. The inputs into the networks are the original SAR images. Each network is called to classify an area of the image as dark area or sea. The group of MLP networks can be recognized as the most suitable group for dark formation detection, as it presents reliable stable results for all the examined accuracies. Nevertheless, in terms of single topology, there is no an MLP topology that performs significantly better than the others.

Potentiality of feed-forward neural networks for classifying dark formations to oil spills and look-alikes

Radar backscatter values from oil spills are very similar to backscatter values from very calm sea areas and other ocean phenomena. Several studies aiming at oil spill detection have been conducted. Most of these studies rely on the detection of dark areas, which have high Bayesian probability of being oil spills. The drawback of these methods is a complex process, mainly because non-linearly separable datasets are introduced in statistically based decisions. The use of neural networks (NNs) in remote sensing has increased significantly, as NNs can simultaneously handle non-linear data of a multidimensional input space. In this article, we investigate the ability of two commonly used feed-forward NN models: multilayer perceptron (MLP) and radial basis function (RBF) networks, to classify dark formations in oil spills and look-alike phenomena. The appropriate training algorithm, type and architecture of the optimum network are subjects of research. Inputs to the networks are the original synthetic aperture radar image and other images derived from it. MLP networks are recognized as more suitable for oil spill detection.

The integration of spatial context information in an experimental knowledge-based system and the supervised relaxation algorithm—two successful approaches to improving SPOT-XS classification

Abstract This paper describes two different methods which integrate contextual information in a classification process. This process aims to refine the map products given by the application of a common parametric classification algorithm. The first method is the well known Supervised Relaxation Algorithm, and makes use of the first classification, with additional contextual information. The contextual information is derived either from texture features or from other map products introducing additional information on the existing land use classes. The second method is a knowledge-based system, which makes use of image and geographical context rules. The probability figures, derived from the image classifier and the rule base are combined by the use of the Dempster-Shafer reasoning scheme. Experiments using satellite data from the Loir et Cher region (Central France), together with the appropriate ground truth data, have shown that both methods return improved classification products in terms of thematic an...

A thinning-based method for recognizing and extracting peri-urban road networks from SPOT panchromatic images

In this paper we describe a method for recognizing and extracting the road network in peri-urban areas using SPOT panchromatic images. A particular combination of image representation description algorithms is proposed, which recognizes road features-not clearly defined in remotely sensed images and often confused with other features- and extracts them. The method consists of five algorithms-thresholding, morphological, thinning, linking, and gap filling-that arc used sequentially. The only human intervention required is the definition of a threshold. The proposed approach produces a raster road network representation that is highly complete and locationally accurate. Some experimental results are given in this paper.

Dark formation detection using recurrent neural networks and SAR data

In this paper a classification scheme based on recurrent neural networks is presented. Neural networks may be viewed as a mathematical model composed of many non-linear computational elements, called neurons, operating in parallel and massively connected by links characterized by different weights. It is well known that conventional feedforward neural networks can be used to approximate any spatially finite function given a set of hidden nodes. Recurrent neural networks are fundamentally different from feedforward architectures in the sense that they not only operate on an input space but also on an internal state space - a trace of what already has been processed by the network. This capability is referred as internal memory of the recurrent networks. The general objectives of this paper are to describe, demonstrate and test the potential of simple recurrent artificial neural networks for dark formation detection using SAR satellite images over the sea surface. The type and the architecture of the network are subjects of research. Input to the networks is the original SAR image. The network is called to classify the image into dark formations and clean sea. Elmans and Jordans recurrent networks have been examined. Jordans networks have been recognized as more suitable for dark formation detection. The Jordans specific architecture with five inputs, three hidden neurons and one output is proposed for dark formation detection as it classifies correctly more than 95.5% of the data set.

The Use of Expert System and Supervised Relaxation Techniques to Improve Spot Image Classification Using Spatial Context

Two methods have been tested in order to improve land use mapping in a post-classification refinement process: supervised relaxation and an expert system. Both methods use multiple sources of information and return satisfactory results. Statistical measurements of texture have been used to provide ancillary information for land use mapping at a super-class level (general classes) in a land cover classification tree. The reasoning model of the supervised relaxation technique is based upon the Bayesian theory. In contrast, the expert system uses the Dempster-Shafer reasoning scheme and allows evidence to be propagated at various levels in the land cover taxonomic hierarchy. The result of this approach may be a mixed-level map product, if the available amount of evidence is insufficient to decide among singleton competing labels. Thus, limitations in the entry-data set for accurate and f i e classifications can be defined and resolved. The knowledge base of the expert system contains a set of 40 spatial context rules.

Integrating thermal and hyperspectral remote sensing for the detection of coastal springs and submarine groundwater discharges

This research focuses on the investigation of remote-sensing techniques for the detection of coastal sub-aerial springs and submarine groundwater discharges using airborne thermal and hyperspectral imagery. Very high spatial resolution thermal and hyperspectral images were acquired using Thermal Airborne Broadband Imager 320 (TABI-320) and Compact Airborne Spectrographic Imager 550 (CASI-550) sensors. Extensive in situ spectroradiometer and oceanographic measurements were carried out in parallel with thermal and hyperspectral image acquisitions. Experiments and analysis of the data show that the combined use of very high spatial resolution airborne thermal and hyperspectral sensors for the detection of relatively small sub-aerial coastal springs and submarine groundwater discharges proves to be a very efficient and operational method. Very high spatial resolution thermal data were able to detect even very small coastal sub-aerial springs. On the other hand, the hyperspectral data were the most appropriate for detecting relatively small submarine groundwater discharges, which were not detected on thermal imagery, due to the increase in turbidity that these discharges cause. This is confirmed by the strong correlations between the hyperspectral data and the in situ measured turbidity-related water inherent optical properties.

Hyperspectral data and methods for coastal water mapping

Motivated by the increasing importance of hyperspectral remote sensing, this study investigates the potential of the current-generation satellite hyperspectral data for coastal water mapping. Two narrow-band Hyperion images, acquired in summer 2004 within a nine day period, were used. The study area is situated at the northern sector of south Evvoikos Gulf, in Central Greece. Underwater springs, inwater streams, urban waste and industrial waste are present in the gulf. Thus, further research regarding the most appropriate methods for coastal water mapping is advisable. In situ measurements with a GPS have located the positions of all sources of water and waste. At these positions groundspectro-radiometer measurements were also implemented. Two different approaches were used for the reduction of the Hyperion bands. First, on the basis of histogram statistics the uncalibrated bands were selected and removed. Then the Minimum Noise Fraction was used to classify the bands according to their signal to noise ratio. The noisiest bands were removed and thirty-eight bands were selected for further processing. Second, mathematical and statistical criteria were applied to the in situ radiometer measurements of reflectance and radiance in order to identify the most appropriate parts of the spectrum for the detection of underwater springs and urban waste. This approach has determined nine hyperspectral bands. Τhe Pixel Purity Index and the n-D Visualiser methods were used for the identification of the spectra endmembers. Both whole (Spectral Angle Mapper or Spectral Feature Fitting) and sub pixel methods (Linear Unmixing or Mixture-Tuned Matched Filtering) were used for further analysis and classification of the data. Bands resulting from processing the groundspectro-radiometer measurements produced the highest classification results. The spatial resolution of the Hyperion hyperspectral data hardly allows the detection and classification of underwater springs. Contrary, inwater streams and chlorophyll are satisfactorily classified. The SAM classification method seems to work better as the number of endmembers increases. The Linear Unmixing classification method gives better results as the number of endmembers decreases.