Implementation of a fast coral detector using a supervised machine learning and Gabor Wavelet feature descriptors

Abstract

The task of reef restoration is very challenging for volunteer SCUBA divers, if it has to be carried out at deep sea, 200 meters, and low temperatures. This kind of task can be properly performed by an Autonomous Underwater Vehicle (AUV); able to detect the location of reef areas and approach them. The aim of this study is the development of a vision system for coral detections based on supervised machine learning. In order to achieve this, we use a bank of Gabor Wavelet filters to extract texture feature descriptors, we use learning classifiers, from OpenCV library, to discriminate coral from non-coral reef. We compare: running time, accuracy, specificity and sensitivity of nine different learning classifiers. We select Decision Trees algorithm because it shows the fastest and the most accurate performance. For the evaluation of this system, we use a database of 621 images (developed for this purpose), that represents the coral reef located in Belize: 110 for training the classifiers and 511 for testing the coral detector.