Segmentation through patch classification: A neural network approach to detect Posidonia oceanica in underwater images

This paper focuses on the detection of Posidonia oceanica in underwater images. The input image is split into a set of patches that are classified as depicting Posidonia or not. Two different Neural Networks are proposed to perform the classification. A region growing algorithm able to accurately detect the contours of the Posidonia oceanica from the output of the classifier is also described. The experimental results, performed using images gathered in coastal areas of Mallorca (Spain), show that our proposal reaches the 95.5% of accuracy, the 95.8% of precision and the 95.4% of recall with only a 4.6% of fallout. When compared to previous studies, these results surpass those of methods based on Machine Learning and are comparable, even superior in some cases, to recent Deep Learning approaches. The advantages in terms of computational requirements, which are crucial in underwater robotics, are also highlighted. In particular, our proposal works up to 8.6 times faster than other Deep Learning methods.