Land use/land cover mapping for conservation of UNESCO Global Geopark using object and pixel-based approaches

9th IGRSM International Conference and Exhibition on Geospatial & Remote Sensing

地球科学;计算机科学

Halim, M.K.A.^1 ; Ahmad, A.^1 ; Rahman, M.Z.A.^1 ; Amin, Z.M.^1 ; Khanan, M.F.A.^1 ; Musliman, I.A.^1 ; Kadir, W.H.W.^1 ; Jamal, M.H.^2 ; Maimunah, D.S.^2 ; Wahab, A.K.A.^2 ; Zabidi, M.M.A.^3 ; Suaib, N.M.^4 ; Zain, R.M.^4

Faculty of Geoinformation and Real Estate, Universiti Teknologi Malaysia, Malaysia^1

Faculty of Civil Engineering, Universiti Teknologi Malaysia, Malaysia^2

Faculty of Electrical Engineering, Universiti Teknologi Malaysia, Malaysia^3

Faculty of Computer Science, Universiti Teknologi Malaysia, Malaysia^4

关键词: Classification accuracy;    Classification process;    Digital classification;    Management and conservations;    Maximum likelihood classifications;    Pixel based classifications;    QuickBird satellite imagery;    Supervised classification;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/169/1/012075/pdf
DOI  :  10.1088/1755-1315/169/1/012075

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

Land Use/Land Cover (LULC) is essential in planning and management activities especially for conserving eco-environment, soil and vegetation research as well as urban planning. Higher resolution imagery and accuracy of LULC for monitoring ecosystem survival are preferred especially when it takes into account environmental issues. Langkawi had faced problems related to environmental issues after it has been designated as a geopark. Therefore, this study aims to map and evaluate digital classification methods of mapping of LULC using Very High Resolution (VHR) Quickbird satellite imagery in one of the Langkawi UNESCO Global Geopark, that is Kilim Karst Geoforest Park (KKGP) which is located at northeast of Langkawi, Kedah, Malaysia. Object-based and pixel-based classification methods were explored and compared. Object-based method involved multi-resolution segmentation part where scale parameter, shape and compactness should be assigned as accurate as possible, so that the image is segmented to homogenous area. Both segmentation and classification processes were conducted in e-Cognition software. While, a supervised classification, Maximum Likelihood Classification (MLC) involved selection of training areas was used for pixel-based method using ERDAS Imagine software. Then, classification accuracies were assessed by comparing both techniques using error matric and Kappa coefficient. The results from the classified image shows that the object-based approach provides more accurate results with an overall accuracy of approximately 87.91% and Kappa coefficient of 0.85 compared to the results achieved by MLC pixel-based classification with 72.21 % accuracy and Kappa coefficient of 0.66. As a conclusion, the results indicated that object-based technique has more advantages to be applied with VHR imagery for better environmental management and conservation actions.