研究目的
To present a novel method for mapping flooded areas using Sentinel-1 ground range detected products, introducing innovations at both the product and methodological levels.
研究成果
The proposed unsupervised method outperforms popular classification techniques in terms of overall accuracy and false alarms, providing rapid flood mapping with high resolution outputs (30m for Chain 1 and 10m for Chain 2) without the need for supervision or thresholding, making it suitable for emergency response.
研究不足
The processing time depends on the despeckling algorithm selected; for example, using the refined Lee filter, computational time is about one hour and a half. The method requires masks for permanent hydrography in single image processing to distinguish from flooded areas.
1：Experimental Design and Method Selection:
The methodology includes two processing chains: Chain 1 for single image analysis using Haralick textural features and a fuzzy decision system, and Chain 2 for change detection using pre- and post-event images with despeckling and cross-calibration.
2：Sample Selection and Data Sources:
Test cases were taken from the Copernicus Emergency Management Service (EMS) database, which provides ground truths and masks for permanent hydrography.
3：List of Experimental Equipment and Materials:
A 4-core, 32GB RAM machine was used for processing. Sentinel-1 GRD products with 10 meters spatial resolution were utilized.
4：Experimental Procedures and Operational Workflow:
For Chain 1, input is a single GRD product processed with multilooking, histogram clip, texture calculation, and fuzzy classification. For Chain 2, input is a couple of calibrated and coregistered GRD products processed with despeckling (refined Lee filter), cross-calibration, change index computation, and fuzzy classification.
5：Data Analysis Methods:
Performance was assessed using overall accuracy (OA) and false alarms (FA) percentages, compared with literature methods like k-means, SVM, NN, ML, and thresholding.
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
