Detail publikace

Metoda vyhodnocení chybovostiklasifikace obrazu

BARTONĚK, D. BUREŠ, J. OPATŘILOVÁ, I. VITULA, A.

Originální název

Method of error assessment in image classification

Český název

Metoda vyhodnocení chybovostiklasifikace obrazu

Anglický název

Method of error assessment in image classification

Typ

článek ve sborníku

Jazyk

en

Originální abstrakt

The article describes a method for the determination of error rate in automated image classification and the improvement of accuracy of results by the filtering using the selected vector datasets of ZABAGED (Fundamental Base of Geographic Data in the Czech Republic) in geographic information system (GIS). The principle of method is based on metric spaces and uses the interpretation of three types of metrics: euclidian, time and thematic. In our case, the evaluation of error rate of the classification results is based on the thematic distance. Special method of assessment was proposed for this type of distance and its essence is a classification tree encoded into chain codes. The distance between values of thematic data is determined by comparing their chain codes. The method is semi-automatic with controlled degree of objectivity of achieved results. The proposed method was tested in the project of data analysis of storage of gas facilities under certain types of terrain surface in the Czech Republic (CR). This analysis was done in order to determine reproductive values of gas facilities (pipelines) and the valuation of costs which would be necessary to spend for building new networks. The authors solved this project for the GasNet, Ltd. Company which is a part of a RWE group in the CR. Input data were raster datasets of orthophoto with the resolution of 25 cm/1 pixel and vector layers of the route of communications of the ZABAGED CR. Due to the territorial coverage of the CR with the area of 64,350 km2, these were massive tasks with data volume of 500 GB. The whole data analysis was carried out in ArcGIS 10.0 environment with using purpose-built applications in Python language with support for ESRI libraries. The used technology of data analysis demonstrated the low error rate in the range of 2% - 3% on the whole modeled area.

Český abstrakt

V článku je popsána metoda pro stanovení chybovosti při automatizované klasifikaci obrazu se zpřesněním výsledků filtrací s využitím výbraných vektorových datových sad ZABAGED (Fundamental Base of Geographical Data) v geografickém informačním systému (GIS). Podstata metody vychází z metrických prostorů a využívá interpretaci 3 typů metrik: 1. euklidovské, 2. časové a 3. tématické. V našem případě je vyhodnocení chybovosti výsledků klasifikace založeno na tématické vzdálenosti. Pro tento typ vzdálenosti byla navržena speciální metoda vyhodnocení, jejíž podstatou je klasifikační strom zakódovaný do řetězcových kódů. Vzdálenost mezi hodnotami tématických dat je dána porovnáním jejich řetězcových kódů. Metoda je poloautomatická s řízeným stupněm objektivity dosažených výsledků. Navržená metoda byla ověřena v rámci projektu datové analýzy uložení plynárenských zařízení pod určitými typy povrchů terénu na území České republiky (ČR) za účelem stanovení reprodukčních hodnot plynárenských zařízení (plynovodů) a ocenění nákladů, které by bylo nutné vynaložit na vybudování nových sítí, který řešili autoři pro společnost GasNet, s.r.o., která je součástí koncernu RWE Česká republika (ČR). Vstupními daty byly datové sady ortofoto – rastr s rozlišením 25 cm/pixel a vektorové vrstvy průběhu komunikací národní Základní báze geografických dat (ZABAGED) ČR. Vzhledem k územnímu pokrytí ČR s rozlohou 64 350 km2 šlo o řešení masivní úlohy s objemem dat ~500 GB. Celá datová analýza probíhala v prostředí ArcGIS 10.0 s využitím k tomu účelově vytvořených aplikací v jazyku Python s podporu knihoven ESRI. Použitá technologie datové analýzy se vyznačuje nízkou absolutní chybovostí v rozmezí 2% – 3% na celém modelovaném území stanovenou metodou popsanou v tomto článku.

Anglický abstrakt

The article describes a method for the determination of error rate in automated image classification and the improvement of accuracy of results by the filtering using the selected vector datasets of ZABAGED (Fundamental Base of Geographic Data in the Czech Republic) in geographic information system (GIS). The principle of method is based on metric spaces and uses the interpretation of three types of metrics: euclidian, time and thematic. In our case, the evaluation of error rate of the classification results is based on the thematic distance. Special method of assessment was proposed for this type of distance and its essence is a classification tree encoded into chain codes. The distance between values of thematic data is determined by comparing their chain codes. The method is semi-automatic with controlled degree of objectivity of achieved results. The proposed method was tested in the project of data analysis of storage of gas facilities under certain types of terrain surface in the Czech Republic (CR). This analysis was done in order to determine reproductive values of gas facilities (pipelines) and the valuation of costs which would be necessary to spend for building new networks. The authors solved this project for the GasNet, Ltd. Company which is a part of a RWE group in the CR. Input data were raster datasets of orthophoto with the resolution of 25 cm/1 pixel and vector layers of the route of communications of the ZABAGED CR. Due to the territorial coverage of the CR with the area of 64,350 km2, these were massive tasks with data volume of 500 GB. The whole data analysis was carried out in ArcGIS 10.0 environment with using purpose-built applications in Python language with support for ESRI libraries. The used technology of data analysis demonstrated the low error rate in the range of 2% - 3% on the whole modeled area.

Klíčová slova

vyhodnocení chybovosti, klasifikace obrazu, GIS

Rok RIV

2014

Vydáno

17.06.2014

Nakladatel

SGEM2014

Místo

Bulgaria

ISBN

978-619-7105-12-4

Kniha

14th GeoConference on Informatics, Geoinformatics and Remote Sensing

Edice

1

Číslo edice

1

Strany od

745

Strany do

752

Strany počet

8

BibTex


@inproceedings{BUT108187,
  author="Dalibor {Bartoněk} and Jiří {Bureš} and Irena {Opatřilová} and Alexej {Vitula}",
  title="Method of error assessment in image classification",
  annote="The article describes a method for the determination of error rate in automated image classification and the improvement of accuracy of results by the filtering using the selected vector datasets of ZABAGED (Fundamental Base of Geographic Data in the Czech Republic) in geographic information system (GIS). The principle of method is based on metric spaces and uses the interpretation of three types of metrics: euclidian, time and thematic. In our case, the evaluation of error rate of the classification results is based on the thematic distance. Special method of assessment was proposed for this type of distance and its essence is a classification tree encoded into chain codes. The distance between values of thematic data is determined by comparing their chain codes. The method is semi-automatic with controlled degree of objectivity of achieved results. The proposed method was tested in the project of data analysis of storage of gas facilities under certain types of terrain surface in the Czech Republic (CR). This analysis was done in order to determine reproductive values of gas facilities (pipelines) and the valuation of costs which would be necessary to spend for building new networks. The authors solved this project for the GasNet, Ltd. Company which is a part of a RWE group in the CR. Input data were raster datasets of orthophoto with the resolution of 25 cm/1 pixel and vector layers of the route of communications of the ZABAGED CR. Due to the territorial coverage of the CR with the area of 64,350 km2, these were massive tasks with data volume of 500 GB. The whole data analysis was carried out in ArcGIS 10.0 environment with using purpose-built applications in Python language with support for ESRI libraries. The used technology of data analysis demonstrated the low error rate in the range of 2% - 3% on the whole modeled area.",
  address="SGEM2014",
  booktitle="14th GeoConference on Informatics, Geoinformatics and Remote Sensing",
  chapter="108187",
  doi="10.5593/SGEM2014/B23/S11.095",
  edition="1",
  howpublished="print",
  institution="SGEM2014",
  number="III",
  year="2014",
  month="june",
  pages="745--752",
  publisher="SGEM2014",
  type="conference paper"
}