Eavesdropping on insect herbivory: Deep learning-enabled audio monitoring of insect feeding

Herbivorous insects are major crop pests, globally resulting in an estimated $430 billion in economic damage annually. Detecting and monitoring insect herbivory is crucial for effective pest management. Conventional methods can be time-consuming, inefficient, and occasionally destructive. This presentation explores the use of $8 contact microphones to eavesdrop on insect herbivores.

The tool was evaluated first in the lab, then in the field. Controlled laboratory experiments compared the signal from a pair of microphones observing plants with and without insects. Specifically, microphones were clipped stems of plants with and without tobacco hornworm (Manduca sexta Linnaeus), Colorado potato beetle (Leptinotarsa decemlineata Say), European corn borer (Ostrinia nubilalis Hübner), and northern corn rootworm (Diabrotica longicornis Say) on tobacco, potato and corn plants, respectively. The microphone was field tested to determine corn rootworm feeding in high-density fields. Recordings were analyzed using spectrograms and deep-learning convolutional neural networks to determine insect presence, quantify herbivore differences, and identify insect species. Microphones detected all herbivorous insects feeding on the tobacco, potato, and corn plants. The recordings also revealed differences in the insect species' feeding patterns across frequency, amplitude, and circadian activity.

Microphones offer a promising non-invasive herbivory monitoring method, especially for boring or root-feeding insects. This approach has great potential for early detection and precision practices. Results suggest sound can better inform on the behavior and identity of insect herbivores. By providing a simple and effective tool for detecting and monitoring insect activity, microphones have widespread potential applications in agriculture and ecology.