A review of acoustic systems for insect monitoring: From recording technologies to species recognition performance

The key underlying principle of bioacoustics is the recording of acoustic data and the subsequent identification of taxa from these recordings. Acoustic data can originate from different animal behaviors, such as birdsong, echolocation calls, wing flap, or even crawling on hard surfaces. Typical sound production mechanisms specific to insects include chewing, stridulation in grasshoppers, and the use of tymbals in cicadas, each of which generates distinctive acoustic patterns that can be leveraged for taxon identification. While theoretically a broad diversity of animal taxa can be detected by means of bioacoustic methods, prior research has predominantly focused on birds and bats. Only in recent years there has been much work dedicated to insect sound detection, underpinning the great importance of insects in ecosystems and demonstrating the wider application potential of bioacoustics. Using a methodological survey approach, this paper provides an overview of the current state of the art regarding the use of artificial intelligence and machine learning techniques for acoustic insect recognition. Besides examining data processing methods, we also survey the recording systems that were used to collect insect sounds as well summarizing the characteristics of insect sound datasets used in corresponding studies. Our review revealed that the majority of works focuses on a recognizing particular group of insects only, rather than seeking to identify larger parts of the entomofauna. This is in contrast to recognition approaches for other animals, e.g., birds, where generally all species native to a certain region are considered. Second, real-world insect recordings were only collected in slightly more than half of the surveyed studies, whereas most relied on data from the few datasets published online. Moreover, most studies only recorded acoustic data in controlled environments rather than in the field, thus leaving it unclear whether they will also work in practical settings. Third, in a considerable number of cases, the recording setup used a sampling rate too low to fully capture all sounds produced by insects, leading to the potential omission of important characteristics in the analyses.