Authors: Phuong Dao*, University of Toronto, Dept of Geography, Kiran Mantriprigada, University of Ontario Institute of Technology, Faculty of Science, Yuhong He, University of Toronto, Department of Geography, Faisal Qureshi, University of Ontario Institute of Technology, Faculty of Science
Topics: Remote Sensing, UAS / UAV, Land Use and Land Cover Change
Keywords: Outlier detection, Optimal scale selection, Image segmentation, OBIA, Hyperspectral image classification
Session Type: Paper
Presentation File: No File Uploaded
Image segmentation is a critical component of the Object-Based Image Analysis and interpretation. Several supervised and unsupervised methods have been proposed to obtain optimal segmentation scale parameters. Among them, unsupervised parameter optimization methods that consider image object or inter-object local statistics are widely used given their capability to quantitatively assess output segments. Unsupervised parameter optimization methods typically use global measures of spatial and spectral properties calculated from all image objects in all bands as the target criteria to determine the optimal segmentation scale. However, no studies consider the effect of noise levels of bands in different spectral regions into the segmentation assessment and scale selection. This ignorance can cause inaccurate segmentation, especially when there are high-noise in NIR and SWIR bands. Furthermore, these global measures are affected by outliers or extreme values from a small number of objects, which lead to incorrect assessment and selection of optimal scales for the majority of other objects. These issues become more critical when segmenting hyperspectral data that exhibit large spectral variability in different spectral bands and across the spectrum. In this study, we propose an enhanced method that 1) incorporates the band’s inverse noise weighting in the segmentation and 2) detects and removes outliers before determining segmentation scale parameters. The proposed method is evaluated on three well-established segmentation approaches – k-means, mean-shift, and watershed. The generated segments are validated by comparing them with reference polygons. The results demonstrated that this proposed scale selection method produces more accurate and reliable segmentation results.