Determination of Harmful Algal Blooms in Satellite Images using Decision Tree Algorithm

Abstract

Using a decision tree algorithm, as a kind of machine learning technique, we distinguished the area where the red tide occurred or not occur. The satellite image was divided into smaller size lattices, and each lattice is classified by the inclusion of red tide pixels. 5 filters with different optical properties were created for quantify the features of each lattices. The generated filters were used to computing similarity with each sample lattice. As a result of the classification model, we confirmed 4 features related to the presence or absence of red tide pixels in the grid. As a result of comparison with the verification data, it was confirmed that the grid with the size of 15 km had excellent classification performance.inclusion of red tide pixels. 5 filters with different optical properties were created for quantify the features of each lattices. The generated filters were used to computing similarity with each sample lattice. As a result of the classification model, we confirmed 4 features related to the presence or absence of red tide pixels in the grid. As a result of comparison with the verification data, it was confirmed that the grid with the size of 15 km had excellent classification performance.