Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Chine → 2008 → Study On Automatic Classification For Land Use/Cover In Arid And Semi-arid Area Based Upon Remotely Sensed Image Cognition

Lanzhou University (2008)

Study On Automatic Classification For Land Use/Cover In Arid And Semi-arid Area Based Upon Remotely Sensed Image Cognition

李爱华;Li Ai Hua

Titre : Study on Automatic Classification for Land Use/Cover in Arid and Semi-arid Area Based upon Remotely Sensed Image Cognition

Auteur : 李爱华;Li Ai Hua

Grade : Master’s Theses 2008

Université : Lanzhou University

Résumé partiel
The classification of remotely sensed image, which can quickly obtain the spatial distribution and conduct the real-time change monitoring of land use/land cover, plays an important role in environmental change research. However, the conventional methods and means of remote sensing information extraction and classification can not meet modern scientific research depending on the high-precision data requirements. Particularly the important area for environmental studies-arid or semi arid area is different from the humid area, featuring by landscape patterns, complex causes of its own forming and the diversity of land cover. So its special characteristics made the classification more difficult than that in humid area. Many mature classification methods of remotely sensed image and even some new classification methods are subject to certain restrictions and the advantages of them are not fully used in the arid or semi arid area. So the study on high-precision classification of remotely sensed image that is suitable for arid or semi arid area, is not only a natural development direction of remote sensing technology, but also meets the needs of remote sensing image analysis and environmental monitoring, particularly LUCC and the ecological landscape research.This article aimed at the automatic classification of high accuracy and efficiency. The study focused on the classification process based on remotely sensed image cognition and understanding. The study area lies in the intersection of Gansu province, Ningxia Hui nationality autonomous region and Inner Mongolia autonomous region. The data from Landsat5 Thematic Mapper sensor was used in this project. One of the two scenes was captured on 21 Aug. 2006 while the other one was captured on 15 Sep. 2006. The relative DEM was the 90m-SRTM3 DEM data. At first, it was to co-register the 30-meter Landsat TM data to 1:50,000 topographic maps and project the 90-meter DEM to match the TM data’s projection. Then do topographic analysis on DEM to get slope as well as aspect images. Second, the reflectance image after calibration and atmospheric correction of TM data, the DEM, the aspect and slope images were taken as the layers for the segmentation. Additionally the river, railway, road and penstock collected in the study area were used as thematic layers for some line objects classification.The classification based on remotely sensed image cognition and understanding was introduced to make a simulation for the human brain’s remotely sensed image cognition and understanding, which comes of the visual interpretation and orients to be approximate to human’s thinking process and point of cognition. To begin with, make an analysis for the image to draft the classification scheme of land use/land cover in the study area. Secondly, some aspects about remotely sensed image multi-segmentation, including the effect of scale, optimizing the scale and the transformation of scale, were preliminarily studied in order to convert the basic process unit "a single pixel" to "the real meaning object", realizing the first step of human brain’s cognition and understanding. Through comparing the 15-scale segmentations of different magnitude data, the data that multiplied the reflectance image data by 10,000 and can reflect the subtle differences in ground covers, had the best subtle result. At the same time, a larger scale segmentation with 200 scale parameter labeling as the super objects level was done for the large and relatively well-proportioned area to make up the fragment caused by the segmentation with 15 scale parameter. Thirdly, the spectral features, texture features, shape features, spatial distribution features and semantic characteristics that are similar to human language expression were delineated. Following that, the class hierarchy was discussed and created. The last step is to automatically do the fuzzy classification and make the accuracy assessment.Through the study above, the conclusions of this paper are as followings

Mots clés : image cognition; object-oriented; fuzzy classification; multi-segmentation; arid and semi-arid area;

Présentation (CNKI)

Page publiée le 22 avril 2013, mise à jour le 9 novembre 2017