研究目的
To develop an object-based method for SAR image change detection that reduces spurious changes and is insensitive to speckle noise by using multi-temporal segmentation, a new statistical distance based on Nakagami–Rayleigh distribution, and cluster ensemble.
研究成果
The proposed object-based method effectively reduces spurious changes and is robust to speckle noise in SAR images, achieving higher accuracy and lower error rates compared to existing methods, as validated with Radarsat-2 data.
研究不足
The method may be sensitive to mis-registration and different incident angles in SAR images, which can affect accuracy. The segmentation and distance calculation rely on specific parameters and models that might not generalize to all SAR data types.
1：Experimental Design and Method Selection:
The method involves multi-temporal segmentation of SAR images using eCognition software, application of Nakagami–Rayleigh distribution for noise modeling, Bayesian formulation for deriving a statistical distance, and cluster ensemble for combining results across scales.
2：Sample Selection and Data Sources:
Multi-temporal Radarsat-2 SAR images of Vancouver airport acquired in January 2008 and November 2009 with C-band, HH polarization, and 3 m resolution are used.
3：List of Experimental Equipment and Materials:
Radarsat-2 sensor for image acquisition, eCognition software for segmentation.
4：Experimental Procedures and Operational Workflow:
Co-register images, perform multi-temporal segmentation at multiple scales (scale parameters: 10, 20, 30, 35, 40), calculate distances between parcel pairs using the derived statistical distance, apply cluster ensemble with weighted voting to combine results, and compare with other methods (MR, MCP, CN, WT).
5：Data Analysis Methods:
Quantitative analysis using false alarm (FA), missed alarm (MA), false alarm rate (Pfa), missed alarm rate (Pma), Accuracy, and Kappa index.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
