Marine Floating Raft Aquaculture Detection of GF-3 PolSAR Images Based on Collective Multikernel Fuzzy Clustering

Marine floating raft aquaculture (MFRA) is a primary human activity on oceanic resources utilization and development, the physical structure of which is so particular that conventional optical remote sensing images cannot discover thoroughly. GaoFen-3 (GF-3), as the first launched full polarimetric synthetic aperture radar satellite of China, can offer a unique opportunity to perform long-term precise monitoring of the nearshore MFRA. In this paper, GF-3 product description is discussed to choose applicable modes to analyze MFRA imaging characters. Moreover, the structure of MFRA is decomposed into different parts to reveal its specific polarimetric scattering mechanism for the first time via the electromagnetic wave propagation theory. Based on the above-mentioned physical analysis, appropriate target decomposition methods are selected to combine other source features of MFRA to integrate into initial distinct features. After that, a novel unsupervised collective multikernel fuzzy C-means (CMKFCM) algorithm with collective neurodynamic optimization approach is proposed to exploit these multisource features of MFRA. The effective feature combination based on proposed CMKFCM can take full advantage of various features through establishing a unified framework, which can optimize local and global parameters hierarchically and improve the entire clustering accuracy in the end. The experiments substantiate the efficacy and superiority of GF-3 MFRA detection based on our proposed algorithm compared with other state-of-the-art unsupervised approaches.