Optical classification of inland waters based on an improved Fuzzy C-Means method

• DOI: 10.1364/oe.27.034838
• Publication Date: 2019-11-13T20:52:34Z
• Authors: Shun Bi; Yunmei Li; Jie Xu; Ge Liu; Kaishan Song; Meng Mu; Heng Lyu; Song Miao; Jiafeng Xu
• Keywords: Ocean color; Data set

Water optical clustering based on water color information is important for many ecological and environmental application studies, both regionally globally. The fuzzy method avoids the sharp boundaries in type-memberships produced by hard methods, thus presents its advantages. However, to make good use of methods spectra data sets, determination fuzzifier parameter (m) FCM (fuzzy c-means) key factor. Usually, m set 2 default. Unfortunately, this assigned some membership degrees non-belonging type, failing obtain unitarity cluster structure cases, especially inland eutrophic water. To overcome shortcoming, we proposed an improved (namely FCM-m) classification optimizing parameter. We collected containing 1280 situ spectral co-measured quality parameters with a wide range biogeochemical variability China. Using FCM-m, seven spectrally distinct clusters Sentinel-3 OLCI (Ocean Land Colour Imager) bands were obtained optimized (m=1.36), well-performed result assessed validated index (Fuzzy Silhouette Index=0.513). Also, FCM-m-based soft framework was successfully applied atmospherically corrected images, which evaluated previous case studies. Besides, testing FCM-m three coastal oceanic verified that should be adjusted itself, general, value gradually approaches 1 increase band number (or dimension). Finally, effect tested Chlorophyll-a concentration estimation. results show algorithm------- blending performs better than original FCM, mainly because reduces estimation error from stricter assignation. sum up, believe adaptive algorithm, whose R codes are available at https://github.com/bishun945, needs more public sets.