Hierarchy-guided Neural Networks for Species Classification

Fish species classification is an important task that is the foundation of many industrial, commercial, ecological, and scientific applications involving the study of fish distributions, dynamics, and evolution.While conventional approaches for this task use off-the-shelf machine learning (ML) methods such as existing Convolutional Neural Network (ConvNet) architectures, there is an opportunity to inform the ConvNet architecture using our knowledge of biological hierarchies among taxonomic classes.In this work, we propose infusing phylogenetic information into the model’s training to guide its structure and relationships among the extracted features. In our extensive experimental analyses, the proposed model, named Hierarchy-Guided Neural Network (HGNN), outperforms conventional ConvNet models in terms of classification accuracy under scarce training data conditions.We also observe that HGNN shows better resilience to adversarial occlusions, when some of the most informative patch regions of the image are intentionally blocked and their effect on classification accuracy is studied.

Assessing the spatiotemporal persistence of fish distributions: a case study on two red mullet species (Mullus surmuletus and M. barbatus) in the western Mediterranean

Understanding the spatiotemporal persistence of fish distributions is key to defining fish hotspots and effective fisheries-restricted areas (FRAs). Hierarchical Bayesian spatiotemporal models provide an excellent framework to understand these distributions, as they can accommodate different spatiotemporal behaviour in the data, primarily due to their flexibility. The aim of this research was to characterize the fundamental behavioural patterns of fish as persistent, opportunistic or progressive by comparing different spatiotemporal model structures in order to provide better information for marine spatial planning. To illustrate this method, the spatiotemporal distributions of 2 sympatric Mullidae species, the striped red mullet Mullus surmuletus and the red mullet M. barbatus, were analysed. The occurrence of each species, its conditional-to-presence abundance and median length were analysed using Mediterranean trawl survey data from the western Mediterranean between 2000 and 2016. Results demonstrate that there are various common hotspots of both species distributed along the Iberian coast. The convenient persistent spatiotemporal distribution of these hotspots facilitates the configuration of a network of connected FRAs for red mullets in the study area.