Buzz trap: Identificação de abelhas usando características acústicas e inteligência artificial

Abstract

There are few studies that have addressed the automatic identification of bees through their sounds. In addition, no study known by us has used the sound that bees produce while vibrating flowers for the automatic recognition of bee species. In this study, we conducted ecoacoustic analyzes and used a Random Forest algorithm to automatically recognize 5 species of bees (Bombus pauloensis, Exomalopsis sp. 1, Oxaea flavescens, Ptiloglossa sp. 1 and Xylocopa suspecta) through their flight and buzz sounds. And we tested if the buzzing sounds are better than the flight sounds to the automatic recognition of bees. It was not possible to identify bee species with high acuracy through ecoacoustic analysis. The fundamental frequency is shown to be the most important parameter for the classification of the species and the machine learning models obtained an accuracy of 90.94% and 82.22% in the classification through the flight and buzz sounds, respectively. Among the bee species studied, Bombus pauloensis was the worst classified by the machine learning models. We argue that this may have occurred because there is great variation in body size among the individuals of this species. Therefore, they generate sounds with varied characteristics that overlap the sounds of other species, increasing the confusion in the classifying algorithms. Nevertheless, bee classification by flight and buzz sounds will potentially help studies involving species richness and abundance in high diverse bee communities.