Internet of Birds (IoB): Song Based Bird Sensing via Machine Learning in the Cloud : How to sense, identify, classify birds based on their songs?

Nagy, Kristof; Cinkler, Tibor [Cinkler, Tibor (Távközlés), szerző] Távközlési és Médiainformatikai Tanszék (BME / VIK); Simon, Csaba [Simon, Csaba (Informatikai tudo...), szerző] Távközlési és Médiainformatikai Tanszék (BME / VIK); Vida, Rolland [Vida, Rolland (Számítógép hálózatok), szerző] Távközlési és Médiainformatikai Tanszék (BME / VIK)

Angol nyelvű Konferenciaközlemény (Könyvrészlet) Tudományos
    How to sense a bird? How to recognize, identify, classify birds? When to involve a "virtual" bird of prey to protect agricultural goods from real birds (game damage)? A machine learning / deep learning (ML/DL), two-level cloud-based sensor network has been implemented to protect agricultural goods from certain bird attacks or attacks by other animals with specific sounds/songs/noises. The background noise will be neglected via the teaching process. The accuracy (lower ratio of both false negative and false positive cases) will be drastically decreased via our two phase network agile - solution. In the first phase, the sensor module placed in the field decides whether there is any likelihood of a critical bird being there, or there is certainty about no bird being around the sensor. This phase relies on a rather simple pre-filtering that is compatible with the limited resources of the sensor module. Then, if there is a certain likelihood of the sought bird, the song - or at least some sound snippets of it - is sent via a radio link, a gateway, and the Internet to a cloud based software unit that performs ML based reliable recognition. This is the second, much more accurate, much more reliable, however, much more computationally intense phase. Once the bird is detected, some actions can be triggered (e.g., playing a bid of prey song, or firing an alarm) remotely from the cloud.
    Hivatkozás stílusok: IEEEACMAPAChicagoHarvardCSLMásolásNyomtatás
    2026-02-08 11:23