Classification of animal species via deep neural networks and species distribution modeling: a systematic review

Abstract

The automated classification of animal species is a challenging task of great ecological importance, which is usually carried out through species distribution modeling (SDM), relying on animal locations and their environment, and/or through image classification from animal photos. On the one hand, there is the well-known SDM, used mainly to estimate the existence of species in certain regions and their ideal conditions. On the other hand, the use of deep neural networks is gaining popularity in ecology, with substantial use in animal identification from photos. A more recent trend is to combine both approaches to improve the final accuracy, from simple to sophisticated combination strategies. This review focuses on works that combine animal image classification models through deep learning with animal SDMs. We obtained 728 articles from the literature, from which we selected and synthesized 13 studies related to the simultaneous use of deep learning and ecological modeling of species in the context of environmental conservation. Thus, we present a summary of applications that integrate deep learning in ecology and SDMs and discuss their limitations and challenges to overcome them.