Segmentation of coring images using fully convolutional neural networks

As a first step in building a toolkit for the computer analysis of images of sea floor sediment cores, we introduce a technique to automate a time consuming manual phase of said analysis. The retrieved cores contain artifacts, e.g., induced by the extraction itself, the removal of which improves the efficiency of environmental reconstruction. From a computer vision perspective, the task of identifying those artifacts is an image segmentation problem. The method we describe as a solution uses the recently introduced fully convolutional neural networks (FCN), which have been shown to be very efficient in segmenting images.