Deep learning and t-SNE projection for plankton images clusterization

In this paper we present a pipeline to cluster unlabelled image samples. Although not restricted to plankton image applications, we present the system within this context. Feature maps obtained from a deep learning architecture (DenseNet) are fed to the t-SNE projection in order to obtain 2D clusters. The method successfully creates clusters that can be used in interactive software, for quick manual classification of images batches.