Behavior Classification of Egyptian Fruit Bat (Rousettus aegyptiacus) From Calls With Deep Learning

Introduction

The bats had separated from other mammals towards the end of the Cretaceous period (~70 million years ago). But main diversification had occurred around 52 million years ago. They are an important order of mammals that have gained the ability to fly like birds and that recognized about 1386 species all over the world with that the second largest order of the mammals (Dietz et al., 2009, Burgin et al., 2018). Bats have ecologically and economically important roles in the ecosystem as predators, prey for vertebrates, hosts for parasites, pollination, seed dispersal, soil fertility, and nutrient distribution, bioindicators, and they serve as a biological control agent for pests (Kasso & Balakrishnan, 2013). Because of ecological and economic importance, all bat species are protected according to Bern Convention in appendices II and III, and some of their conservation statuses are evaluated near threatened, vulnerable, endangered, or critically endangered in the IUCN (International Union for Conservation Nature).

There are 39 bat species living in Turkey (Özkurt & Bulut, 2020). Among 39 bat species in Turkey, there is a special one because of its foraging type. As its name suggests, the Egyptian fruit bat feeds on fruits and it differs from the other 38 insectivorous species. Egyptian fruit bats are considered as important agents for the ecological balance due to their diet. Their preferred meals are seeded fruits and flower nectars such as figs, cherry, plum, Persian lilac, apricots, peaches, apples, citrus plants, bananas, as well as mulberries and dates. In winters, they eat the leaves of carob and figs as well. Egyptian fruit bats have a minor role in pollination, however, when they travel through different locations, they play important role in seed dispersal of their food plants. They leave feces with undigested plant seeds, increasing the spread of plants with enhanced fertility, and impacting the preservation of plant species positively (Albayrak et al., 2008; Dietz et al., 2009). Egyptian fruit bats are important for the biodiversity of Turkey, because of being Afro-tropical species reaching the northernmost of its distribution (Benda et al., 2012), and regarding as a separate, endemic in terms of 10% mtDNA divergence from sub-Saharan Populations (DelVaglio et al., 2011). Egyptian fruit bats use echolocation and sights to find their ways or foods during the flights. If the light is sufficient, orientation takes place by sight. Echolocation calls are short-paired clicks and emitted between frequencies from 12 to 70 kHz, and duration 0.3 – 0.5 milliseconds long. Echolocation calls provide much information about bats' ecology and behavior. Furthermore, Egyptian fruit bats are social animals: the colonies may comprise from 50 to 500 individuals in some caves. Thus, that will help us to understand the habitat preference, social relationships, foraging, reproduction, mobility and migration, the anomalies in colonies such as diseases, food shortage, underpopulation, and overpopulation. In this chapter, the authors aim to detect the important behaviors of bats by utilizing deep learning methods on their calls, rather than only finding different species. The obtained results have the potential to enlighten the path for understanding the bats' or other animals’ vocal communication and their social interaction.

The rest of this chapter is organized as follows. The next section presents a literature review on classification based on animal calls. Then, the contributions of the chapter with respect to the previous literature are given. After the explanation of the database used in this work and its data processing, two classification approaches are introduced. First, conventional machine learning approaches based on the one-dimensional (1D) features generated from the bat calls are investigated. Then, for the same purpose a deep learning method, i.e., convolutional neural network (CNN) is also considered. Upon the performance comparison between classical machine learning and deep learning techniques, the chapter is finalized with conclusions and a short discussion of future research directions.