Zhai Min, Shen Huayu, Cao Qihang, Ding Xuanhao, Xin Mingzhen
College of Geodesy and Geomatics, Shandong University of Science and Technology, Qingdao 266590, China.
Key Laboratory of Beidou Navigation and Intelligent Spatial Information Technology Application, Shandong University of Science and Technology, Qingdao 266590, China.
Sensors (Basel). 2025 Aug 7;25(15):4868. doi: 10.3390/s25154868.
Synthetic Aperture Radar (SAR) technology has the characteristics of all-day and all-weather functionality; accordingly, it is not affected by rainy weather, overcoming the limitations of optical remote sensing, and it provides irreplaceable technical support for efficient water body extraction. To address the issues of low accuracy and unstable results in water body extraction from Sentinel-1 SAR images using a single method, a water body extraction method fusing the Sentinel-1 dual-polarized water index and random forest is proposed. This novel method enhances water extraction accuracy by integrating the results of two different algorithms, reducing the biases associated with single-method water body extraction. Taking Dalu Lake, Yinfu Reservoir, and Huashan Reservoir as the study areas, water body information was extracted from SAR images using the dual-polarized water body index, the random forest method, and the fusion method. Taking the normalized difference water body index extraction results obtained via Sentinel-2 optical images as a reference, the accuracy of different water body extraction methods when used with SAR images was quantitatively evaluated. The experimental results show that, compared with the dual-polarized water body index and the random forest method, the fusion method, on average, increased overall water body extraction accuracy and Kappa coefficients by 3.9% and 8.2%, respectively, in the Dalu Lake experimental area; by 1.8% and 3.5%, respectively, in the Yinfu Reservoir experimental area; and by 4.1% and 8.1%, respectively, in the Huashan Reservoir experimental area. Therefore, the fusion method of the dual-polarized water index and random forest effectively improves the accuracy and reliability of water body extraction from SAR images.
合成孔径雷达（SAR）技术具有全天时、全天候工作的特点；因此，它不受阴雨天气影响，克服了光学遥感的局限性，为高效水体提取提供了不可替代的技术支持。针对单一方法从哨兵-1 SAR图像中提取水体时精度低、结果不稳定的问题，提出了一种融合哨兵-1双极化水体指数和随机森林的水体提取方法。这种新方法通过整合两种不同算法的结果提高了水体提取精度，减少了与单一方法水体提取相关的偏差。以大芦湖、银湖水库和华山水库为研究区域，利用双极化水体指数、随机森林方法和融合方法从SAR图像中提取水体信息。以通过哨兵-2光学图像获得的归一化差异水体指数提取结果为参考，定量评估了不同水体提取方法与SAR图像结合使用时的精度。实验结果表明，与双极化水体指数和随机森林方法相比，融合方法在大芦湖实验区总体水体提取精度和Kappa系数平均分别提高了3.9%和8.2%；在银湖水库实验区分别提高了1.8%和3.5%；在华山水库实验区分别提高了4.1%和8.1%。因此，双极化水体指数与随机森林的融合方法有效地提高了从SAR图像中提取水体的精度和可靠性。
