Papadopoulos Spiros, Koukiou Georgia, Anastassopoulos Vassilis
Electronics Laboratory, Physics Department, University of Patras, 26504 Patras, Greece.
J Imaging. 2024 Apr 12;10(4):91. doi: 10.3390/jimaging10040091.
Decision fusion plays a crucial role in achieving a cohesive and unified outcome by merging diverse perspectives. Within the realm of remote sensing classification, these methodologies become indispensable when synthesizing data from multiple sensors to arrive at conclusive decisions. In our study, we leverage fully Polarimetric Synthetic Aperture Radar (PolSAR) and thermal infrared data to establish distinct decisions for each pixel pertaining to its land cover classification. To enhance the classification process, we employ Pauli's decomposition components and land surface temperature as features. This approach facilitates the extraction of local decisions for each pixel, which are subsequently integrated through majority voting to form a comprehensive global decision for each land cover type. Furthermore, we investigate the correlation between corresponding pixels in the data from each sensor, aiming to achieve pixel-level correlated decision fusion at the fusion center. Our methodology entails a thorough exploration of the employed classifiers, coupled with the mathematical foundations necessary for the fusion of correlated decisions. Quality information is integrated into the decision fusion process, ensuring a comprehensive and robust classification outcome. The novelty of the method is its simplicity in the number of features used as well as the simple way of fusing decisions.
决策融合通过融合不同观点在实现凝聚性和统一性结果方面发挥着关键作用。在遥感分类领域，当综合来自多个传感器的数据以得出确定性决策时，这些方法变得不可或缺。在我们的研究中，我们利用全极化合成孔径雷达（PolSAR）和热红外数据为每个像素建立与其土地覆盖分类相关的不同决策。为了加强分类过程，我们采用保利分解分量和地表温度作为特征。这种方法便于提取每个像素的局部决策，随后通过多数投票将这些决策整合起来，为每种土地覆盖类型形成一个全面的全局决策。此外，我们研究每个传感器数据中对应像素之间的相关性，旨在在融合中心实现像素级相关决策融合。我们的方法需要对所采用的分类器进行深入探索，以及相关决策融合所需的数学基础。质量信息被整合到决策融合过程中，确保获得全面且稳健的分类结果。该方法的新颖之处在于其使用的特征数量简单，以及决策融合方式简单。
