Data associated with Implicit Labeling: Unsupervised learning bridges validation data gaps in agroecology

Published: 19 November 2026| Version 1 | DOI: 10.17632/4rvmjwgmbc.1
Contributors:
,
,
,
,
,
,

Description

This dataset contains processed audio event data extracted from contact microphone recordings of ambrosia beetle (Xylosandrus germanus and Anisandrus maiche) tunneling activity in apple (Malus domestica var. Gale Gala) wood under laboratory and field conditions. In the laboratory experiment, 24 apple bolts (approximately 2 cm diameter, 5 cm length) were monitored continuously for 96 hours using Musedo contact microphones connected to a Raspberry Pi 4b via a Sabrent sound card. Nine bolts were exposed to individual X. germanus beetles, nine to individual A. maiche beetles, and six served as uninoculated controls. In the field experiment, three potted three-year-old apple trees were each infested with a single A. maiche beetle and recorded for 24 hours. Raw one-hour WAV files were segmented into discrete audio events using mel spectrogram-based region of interest extraction. Contiguous regions exceeding five times the median background intensity were identified, bounded, merged where overlapping, and extracted as individual snippets. Each audio event was converted to a log-mel spectrogram and compressed into a 384-dimensional latent embedding using a DINO-trained Vision Transformer (ViT-S) model. The dataset includes extracted audio events, associated mel spectrograms, latent embeddings, and experimental metadata (timestamps, treatment group, bolt/tree identity, and recording device information). These data support the implicit labeling methodology described in Still et al. (2025), "Implicit Labeling: Unsupervised learning bridges validation data gaps in agroecology," submitted to Ecological Informatics. Code for data processing and analysis is available at https://github.com/lasti15/implicitlabelling.

Files

Institutions

Categories

Artificial Intelligence, Computer Vision, Insect Pest Management, Machine Learning

Funders

Licence