Classification of land cover observed by airborne polarimetric SAR and its validation

本研究旨在探讨机载偏光合成孔径雷达对具有偏光功能的地表覆盖物进行分类识别的可能性。l波段全极化散射矩阵由美国CRL/NASDA公司开发的Pi-SAR （polarimatric and Interferometric SAR）获得。在1999-2001年期间，Pi-SAR多次飞越新泻地区。幸运的是，美国宇航局-喷气推进实验室开发的AIR SAR也在2000年飞越了同一地区，最终由于他们的环太平洋运动而发生了。在此背景下，我们开展了雷达偏振原理、SAR图像分类、实验室测量验证等方面的基础研究，主要集中在以下几个方面：基于理论的雷达偏振测量原理及分类方案的建立采集数据的偏振校正所开发方案在实验室实验中的验证。人工目标分类有效极化指数的推导对抗自然目标。分类软件的开发基于地面真值测量的SAR数据验证在此期间成功地开展了上述主题，并在国际研讨会和研讨会上发表了一些成果。其中大多数是成功的，然而，一些仍然需要解决。主要难点在于雷达获取的信息是特定于雷达场景的，需要特定于场景的信息才能进行准确的分类和识别。与单一极化通道的60%的分类准确率相比，极化信息分类的总体准确率超过80%。这显示了偏振信息在遥感技术中的优势。

The purpose of this research is to pursuit the possibility of AIRBORNE POLARIMATRIC SYNTHETIC APERTURE RADAR for classifying and identifying land cover with polarimatric function. Fully polarimatric scattering matrix in the L-band is obtained by Pi-SAR (Polarimatric and Interferometric SAR), developed by CRL/NASDA. During the period of 1999-2001, Pi-SAR flew over Niigata Area several times. Fortunately, AIR SAR developed by NASA-JPL also flew over the same area in 2000, which eventually happened due to their Pacific Rim Campaign. Under these circumstances, we carried out our basic research on principle of radar polarimetry, SAR image classification, as well as laboratory measurement with validation, mainly focused on the following topics :1. Establishment of radar polarimetry principle and classification scheme based on the theory2. Polarimetric calibration of acquired data3. Validation of developed scheme in the laboratory experiment4. Derivation of efficient polarimatric index for classifying mart made targets ; against natural targets5. Development of classification software6. Validation of SAR data with ground truth measurementThe above topics were successfully carried out in the period and several results were presented in international symposia and workshops. Most of them were successful, however, some still need to be resolved. The main difficulty was the information obtained by radar is radar scene specific, which requires specific information on the scene for accurate classification and identification. The overall classification accuracy with polarimetric information exceeded more than 80 % compared to 60 % for single polarization channel. This shows the advantage of polarization information for remote sensing technology.