Journal of The Remote Sensing Society of Japan
Online ISSN : 1883-1184
Print ISSN : 0289-7911
Search
OR
Browse
Search
Volume 36 , Issue 1
Showing 1-8 articles out of 8 articles from the selected issue
    • |<
    • <
    • 1
    • >
    • >|
Foreword
Engineering Report
  • Shinya ODAGAWA, Tomomi TAKEDA, Hiroya YAMANO, Tsuneo MATSUNAGA
    Volume 36 (2016) Issue 1 Pages 1-10
    Released: July 30, 2016
    JOURNALS FREE ACCESS
    This technical report proposes an estimation method for cover degree in shallow coral reef areas using hyperspectral bottom index (BI) imagery. Monitoring of coral reefs is important because coral reefs offer important ecosystem services for fisheries, tourism and other socio-economic sectors. However, it is difficult to monitor coral reefs using multispectral sensors due to spectral confusion between coral, algae and seagrass. Previous studies that attempted bottom-type classification using hyperspectral data reduced the influence of water attenuation and diffusion using BI. This proposed method employs a support vector machine (SVM) regression model to estimate the cover degree for each bottom-type. The cover degree is generally used for estimation of coral reef quality. As a result of this analysis, the proposed method achieved a highly accurate estimation model for each bottom-type. The determined coefficient of dead framework, sand, and coral is 0.86, 0.74 and 0.62, respectively. Furthermore, this report attempted to make a bottom-type classification map using estimated cover degrees. Overall accuracy and kappa coefficient of the classification map are 0.77 and 0.70, respectively.
    The proposed method needs the robustness of the water attenuation coefficient ratio (K) used to calculate BI. K is calculated from the reflectance of homogeneous bottom-type sand. The robustness of K was confirmed by reducing the sample size of sand. In the result, sample size did not affect the estimation accuracy of cover degree. BI was calculated by the double logarithm ratio of two bands. The number of combinations is huge in hyperspectral data. This report suggests adopting a base band for calculation of BI. It is possible to avoid the enormous combinatorial calculations using combinations of a base band and other bands. Differences in base bands may affect the estimation accuracy. In the results of the confirmation, bottom-type has an optimal base band for building the regression model.
    View full abstract
    Download PDF (1248K)
Explanation
From Universities and colleges
Laboratory Introduction
Doctorial Thesis Abstracts
Book Review
    • |<
    • <
    • 1
    • >
    • >|
feedback
Top