Acoustic Insect Detection : exploring the potential of deep learning methods applied to airborne insect sound signals as a means of connecting digital systems to the world of insects
Acoustic insect recognition could become a key component of future digital insect monitoring systems. Such tools are urgently needed to advance the digitalization of crop protection in horticulture and agriculture, as well as to support ecological monitoring. However, a major obstacle remains the lack of suitable digital insect sensors. Only if digital systems can directly assess the state of insect populations will they unlock their full potential for improving sustainability in food production and research efficiency.
The main goal of this cumulative dissertation was to investigate the potential of microphone-based (airborne sound) acoustic insect recognition as a viable solution for future digital insect detection. A scalable and reproducible method for generating insect sound data was developed through the design and testing of an anechoic box, the selection of appropriate recording hardware, and the systematic planning of experimental setups and recording sessions. This approach led to the creation of the InsectSound1000 dataset.
Experiments on a subset of this dataset suggest that modern deep learning models can identify insects even from low-level sounds caused by non-flight movements. Simulations with environmental noise further revealed that spatial noise filtering through beamforming can enhance model robustness. A comparison of low- and high-end microphones highlighted a technological gap in available recording hardware.
Overall, this work demonstrates both the potential and the limitations of airborne acoustic signal processing for insect recognition. While louder insects can be detected with high accuracy under quiet conditions, small and silent species, as well as noisy conditions, remain challenging. Future research should focus on the development of specialized insect microphones and on sensor fusion approaches combining microphone arrays with other modalities, such as cameras, to overcome current limitations.
Preview
Rights
Use and reproduction:
Please note that individual components of the publication may be subject to other licensing or copyright conditions.