Coral reef image classifications with hybrid methods

Abstract

There are several organisms on oceans. Among the organisms coral reefs are the one with 800 species. Classifying coral is a difficult task. Scientist classify the coral organism and put in to groups based on their characteristics. There are several machine learning algorithms are implemented to analyzer and classify the coral species. The main aim of this work is to effectively use handcrafted features with deep features for classifying the coral classes. Here the state of art feature descriptors such as Local Binary Pattern, Local Arc Pattern and Improved Webbers Binary Code are proposed to extract the features of coral. The results which obtained can be further improved by combining these local descriptors with convolution neural network. The feature extracted by above methods are classified using KNN and Random Forest. Experiments with these methods are conducted using EILAT dataset. The Experimental results obtained by these methods demonstrate the effectiveness and robustness of our proposed method.