Remote sensing develops different methods and technologies for contactless and cost-effective mapping of large area land cover/use maps and other thematic maps. The key factor for availability and reliability of these maps for the use in Earth sciences is development of effective procedures for analysis and classification of satellite data. With increasing spatial resolution, pixel-based classification methods become less effective, since the relationship between the size of pixel and the dimension of observed objects on the Earth's surface is significantly changed. Therefore over the past decade object-oriented classification is used increasingly. It combines segmentation, which is a fundamental phase of the approach, and contextual classification itself. Segmentation divides image into homogeneous pixel groups (segments), which are then in the process of semantical classification arranged into classes based on their spectral, geometric, texture and other features. The purpose of this paper is to present theoretical argumentation and methodology of object-oriented analysis of remote sensing data, to provide an overview on the field and to point out certain restrictions on current operational solutions.
COBISS.SI-ID: 33536301
Algebra-based satellite imagery comparison recommends substantial data preparation (geometric, radiometric, topographic corrections) and offers numerous techniques to analyse the time sequences. Regardless of the carefully performed data preparation certain differences remain present in the series of images and these are able to drastically influence the imagery comparisons or aiming change detection. Disturbances originate from the natural and technological conditions during data acquisition as well as the pre-processing algorithms. Thus they cannot be completely removed and handled with data correction techniques. This paper examines how particular steps in the pre-processing phase may affect image characteristics and analyse their impact on the magnitude of image spectral properties and comparison feasibility. Supported by the processing of correction routines of almost 30 Landsat images, we will show that the main problem for accessing appropriate homogeneity lies in the poor preservation of histogram properties of spectral bands (distribution width range).
COBISS.SI-ID: 32694573
In September 2010 the International Charter Space and Major Disasters was activated to record extensive floods in Slovenia. A time series of medium resolution radar images (ENVISAT, RADARSAT-2) was obtained and three techniques of flood area detection were employed: pixel-based water delineation, object-based classification and machine learning procedure. The paper examines and compares advantages and restrictions of these methods. It also validates water under-detection and over-detection on radar images, and examines specifics for their occurrence. Discussion is focusing on some proposals to overcome specific drawbacks, and addressing the usefulness of mapping products for local disaster relief. The study confirms that rapid estimation of flooded areas could be achieved from medium resolution radar images. However, the accuracy is not adequate neither for timely, in-situ rescue operations nor for damage assessment, where correct location and exact delineation of the event are of utmost importance. Improving the accuracy of detection presents a future challenge for remote sensing to be truly helpful in disaster relief.
COBISS.SI-ID: 32702509
Within the activation of International Charter on Space and Major Disasters following intensive rains in Slovenia between September 17th and 19th 2010, we obtained time series of radar satellite images, which was used to study flood dynamics in the surroundings of capital Ljubljana and on karstic fields, as well as local water retention and/or retreat. Radar images enabled satisfactory near-real-time comprehension of flood dynamics within wider natural environment, however not in urban areas. Motivated with recent development of space technology in Slovenia, with this article we are also opening a public discussion about applicability and potential of remote sensing data and its rapid mapping products for various users.
COBISS.SI-ID: 32486957
The paper expose problems that can be meet while performing classification over areas of heterogenous characteristics. Two areas in different geographical regions and with different type of land cover were chosen intentionally – the intensive agricultural area of Gornja Radgona in NE of Slovenia and the sub-alpine area of Kobarid in Western part. With the use of object based image analysis approach in this research we aim to present how two main issues were minimized: reducion of negative effects of shadows on the image and improvement of wrong delineation of spectrally similar classes.
COBISS.SI-ID: 32311341