A Multi-Step Detection Method for Extraction of Stable Pointwise Target in Long Temporal SAR Image Series
一种在长时间序列SAR影像上提取稳定目标点的多级探测法

Spaceborne differential interferometric SyntheticApertureRadar technology (D-InSAR) in long temporal series is an mi portant technology in the land subsidencemeasurement, extraction ofstable pointwise target is a criticalprocedure in long temporalD-InSAR processing chain. Aswe know, now the main two kinds ofmethod have been proposed. One is amplitudemethod that can be performed to model the smi ilar behaviors with phase dispersion with no resolution loss by means of a tmi e series analysis of amplitude data. The other is based on coherence stability that can be computed in the stack of interferograms. The twomethods have their own defects, the former requires aminmi um of30 mi ages to detect the stable pointwise targetprecisely, the latterwill directly lead to lower resolution than the original mi ages. In this paperwe introduce an optmi ized extractionmethod of stable pointwise target combining amplitude and coherence analysis. In this paper, amplitude detection method is first mi plemented and the number of stable pointwise targets is extracted. In the step, the data need to be radiometrically calibrated in order to ensure the mi ages to be comparable. And the experiential threshold ofamplitude dispersion is selected to extractstable pointwise targets. Ami ing atsolving problems of low density and ununiformity distribution of pointwise targets extracted by amplitude detection, amulti-step detection method is proposed. Themulti-step detection method first uses amplitude detection, the proper threshold is selected to extract candidates of stable pointwise targets, which can ensuremost of stable pointwise targets are included and greatly mi prove computation efficiency by the threshold. Further temporal coherence is utilized to extract the precise pointwise targets. In the step, themodelof temporal coherence is established, and the partialatmosphere disturb and orbiterrorare removed by lowpass filters, sequentially the phase caused by the DEM error is estmi ated by a rough search of parameter space, the average effective phase are separated in interferogram stack. As the temporal coherence coefficient is proportional to the number of interferograms, the proper threshold is chosen in terms of the proportional relationship. Lastly, pointwise targets of theirpixelvalue larger than the referred threshold are extracted.W ith themulti-step detection method, more reliable stable pointwise targets are extracted. The evaluation of the method is finally carried outwith 26 ERS-1/2 C-band SAR mi ages acquired over Shanghai urban covering 94 km2from 1992 to 2000. Amplitude detectionmethod andmulti-step detectionmethod are mi plemented at the same region respectively. The effectiveness of two methods is evaluated by the number and density of extracted pointwise targets. By comparison, pointwise targets ofextracted by amplitude detection have the lowerdensity thanmulti step detectionmethod. And reliability is evaluated by testing the target corresponding to the