Radiometric Normalization and Cloud Detection of Multi-Temporal Optical Satellite Images Using Invariant Pixels

  • 林 柏毅

Student thesis: Master's Thesis

Abstract

Clouds in optical satellite images can be regarded as information for liquid water measurement or as contaminations that obstruct landscape observation Thus cloud detection that discriminates cloud and clear-sky pixels in images is a necessary processing step in remote sensing applications With radiometric correction/normalization preprocessing most previous studies utilized temporal and spectral information to develop thresholding-based filters with the aid of a cloud-free reference image Although this strategy can efficiently and accurately identify cloud pixels detection accuracy mainly relies on the success of radiometric correction/normalization and quality of the selected cloud-free reference image Radiometric normalization generally suffers from cloud covers; cloud detection is sensitive to radiometric normalization In this study a method based on weighted invariant pixels is proposed for radiometric normalization and cloud detection Utilizing temporal correlations a set of invariant pixels obtained from the scatterplot of two adjacent images in time series through weighted principal component analysis are extracted from a time series of cloud-contaminated images with the reason that the variations of image digital counts during a period are linear The image is normalized by using the selected invariant pixels with quality control or the so-called pseudo-invariant features In addition a composed cloud-free reference image is generated for each cloud-contaminated image by using the selected invariant pixels with a proposed weighting scheme In the experiments qualitative analyses of image sequences acquired by Landsat-7 Enhanced Thematic Mapper Plus (ETM+) sensor and quantitative analyses of image sequences with various cloud contamination conditions and landscapes are conducted to evaluate the proposed method The experimental results show that the proposed radiometric normalization has the ability to deal with images that contain various clouds Moreover cloud detection accuracy is improved by 0 14% to 4 40% with the use of the generated reference images with a thresholding-based detection method
Date of Award2014 Aug 29
Original languageEnglish
SupervisorChao-Hung Lin (Supervisor)

Cite this

'