Supervised Classification Techniques Using Histogram Overlay Method and Its Application to an Investigation on the Turbidity around Kanmon Channel

Abstract

A new classification method has been produced to decrease the computing time of multispectral satellite data classification analysis, maintaining the accuracy as precise as the Maximum Likelihood Method (MLM). In this method, the dimensions of the original data are arbitrarily changeable, ie., M * N original data reduce to M'* N' cells which consist of L * L pixels (L=M/M'=N/N'). The dimensions of training areas are usually chosen to be the same as that of a cell.
The similarity between a cell and a training area is defined as the degree of the overlap of both histograms. This idea is analogous to the Fuzzy theory in a sense. The effect of the cell size and histogram smoothing on classification was carefully discussed in connection with the MLM. It has been found that this method (called Histogram Overlay Method, HOM in abbreviation)results in fair agreement with the MLM.
HOM was applied to the investigation of the turbidity on the Kanmon area between Shimonoseki and Kitakyushu, and reasonable results were obtained. In general, the variation of CCT counts in the ocean is small in comparison with those on land, so that "cell by cell" classification would be sufficient. In addition to this, HOM can be used on personal computers since it takes much less time in classification compared to MLM.