Getting the act together : segmentation-based land cover classification using rapideye imagery and open street map ancillary data
The research deals with land cover classification by means of segmentation-based image analysis and geoprocessing in GIS software. The goal of the research is the overall improvement of classification results obtained from rule-based land cover classification, by utilising ancillary vector data. The...
Lưu vào:
| Tác giả chính: | |
|---|---|
| Định dạng: | BB |
| Ngôn ngữ: | eng |
| Thông tin xuất bản: |
2020
|
| Chủ đề: | |
| Truy cập trực tuyến: | http://tailieuso.tlu.edu.vn/handle/DHTL/5192 |
| Từ khóa: |
Thêm từ khóa bạn đọc
Không có từ khóa, Hãy là người đầu tiên gắn từ khóa cho biểu ghi này!
|
| Tóm tắt: | The research deals with land cover classification by means of segmentation-based image analysis and geoprocessing in GIS software. The goal of the research is the overall improvement of classification results obtained from rule-based land cover classification, by utilising ancillary vector data. The area of interest is the City of Zagreb in Croatia. This research is primarily based on 2015 RapidEye satellite imagery. The assumption that the mixture of urban, peri-urban, rural areas, that makes the territory of the City of Zagreb, is sufficiently interwoven with OpenStreetMap data was confirmed after examination of acquired shapefiles. Intent of the research was to find a solution for land cover classification of geographically heterogeneous region by reconciling the need for convenient and affordable spatial data. Preprocessing and postprocessing of spatial data, as well as the principle component analysis of the imagery, were performed in ESRI ArcGIS. The classification process was performed using Trimble eCognition Developer. Rule-based classification was employed on image segments that were created by the multiresolution segmentation algorithm. Threshold values for statistical features, vegetation indices, and values derived from PCA were set after careful examination of all input datasets. Overall classification accuracy of 91.5% was assessed by error matrix. Imagery of higher spatial resolution were used as reference data for the error matrix. |
|---|