Urban Change Detection Using Coherence and Intensity Characteristics of Multi-temporal ERS-1/2 Imagery
Mingsheng Liao(1) , Liming Jiang(1)
, and Lin Hui(2)
P. O. Box C310, 129 Luoyu Road,
(2) Chinese University of Hong Kong, Shatian, Hong Kong, Hong Kong
The advantages and possibilities of radar remote sensing as an active detection technique have been known and demonstrated in many research works related to land-cover/use change detection. But the potential of single-band and single-polarization intensity imagery is limited for this purpose due to speckle noise and errors in SAR calibration. Recently the development of SAR interferometry has proved that not only the amplitude of the radar echo but also its phase reveals important information for remote sensing applications. Particularly the interferometric correlation, as a measure for the variance of the interferometric phase, provides complementary information about the properties of a 'scene' to the intensity of backscattering. It has been demonstrated by some investigators, that the coherence is a valuable information source for land-use classification if interferometric data set is introduced. In this presentation, the method for detection of land-use change based on coherence and intensity is investigated. Herewith the multi-temporal ERS-1/2 InSAR data is used for experiment. The test site is selected within Shanghai city, China.
In this presentation, a novel supervised approach is proposed for change detection with multi-temporal InSAR data. This proposed approach is characterized by joint analysis of backscattering intensity temporal variability and long-term coherence variability for detecting urban land-use changes in the study area, which mainly includes the following three steps:
(1) Data Preprocessing. It includes radiometric calibration, co-registration, geo-coding and adaptive de-specking filtering within multitemporal InSAR data, which aims at reducing the effect of speckle and terrain slope on backscattering intensity and coherence value.
(2) Features extracting. Based on the analysis of multitemporal SAR signals physical properties in the presence of urban and non-urban areas, three attributes are extracted in terms of backscattering intensity, temporal variability and long-term coherence variability. They reveal the presence of temporal changes (flooded areas, vegetated fields, etc.) and new stable spatial features (build-up, road, etc.). In addition, the difference by subtracting between two ERS-1/2 long-term coherence images is utilized to reveal temporal information about the build-up development within the urban area.
(3) Supervised classifying. A supervised classifier based on radial basis functions (RBF) neural network is constructed to properly exploit the temporal variation characteristics and produce land-use changes maps.
Experiment is carried out on a time-series of InSAR images, including 6 scenes of ERS-1/2 images in Shanghai area, China. Preliminary results shows the effectiveness of the proposed approach in accuracy and reliability, and demonstrated that both interferometric coherence and SAR intensity information extracted from ERS-1/2 InSAR data could be exploited for detecting the land-use changes from rural area to urban area.