Translator Disclaimer
Paper
20 January 2005 Application of a multilayer perceptron neural network to phytoplankton concentration using marine reflectance measures
Author Affiliations +
Abstract
The multilayer perceptron (MLP) neural network have been widely used to fit non-linear transfer function and performed well. In this study, we use MLP to estimate chlorophyll-a concentrations from marine reflectance measures. The optical data were assembled from the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) Bio-optical Algorithm Mini-workshop (SeaBAM). Most bio-optical algorithms use simple ratios of reflectance in blue and green bands or combinations of ratios as parameters for regression analysis. Regression analysis has limitations for nonlinear function. Neural network, however, have been shown better performance for nonlinear problems. The result showed that accuracy of chlorophyll-a concentration using MLP is much higher than that of regression method. Nevertheless, using all of the five bands as input can derive the best performance. The results showed that each band could carry some useful messages for ocean color remote sensing. Only using band ratio (OC2) or band switch (OC4) might lose some available information. By preprocessing reflectance data with the principle component analysis (PCA), MLP could derive much better accuracy than traditional methods. The result showed that the reflectance of all bands should not be ignored for deriving the chlorophyll-a concentration because each band carries different useful ocean color information.
© (2005) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Feng-Chun Su, Chung-Ru Ho, and Nan-Jung Kuo "Application of a multilayer perceptron neural network to phytoplankton concentration using marine reflectance measures", Proc. SPIE 5656, Active and Passive Remote Sensing of the Oceans, (20 January 2005); https://doi.org/10.1117/12.578661
PROCEEDINGS
8 PAGES


SHARE
Advertisement
Advertisement
Back to Top