Sara Hooker & Sean McPherson, Delta Analytics at MLconf Seattle 2017
summary
Delta Analytics and Rainforest Connection deploy recycled cellphones called Guardians streaming 24/7 audio from sites in Ecuador Brazil and Peru and train a convolutional neural network on spectrograms (AlexNet-style) to flag chainsaws, using pseudo-labeling to expand a tiny hand-labeled set and Independent Component Analysis to separate chainsaw from background. Alerts reach rangers within minutes; localization uses time‑delay‑of‑arrival triangulation across guardians to estimate direction and distance. Code and dataset release planned for July.
AI-written summary · about
chapters (6)
- 0:00Start
- 7:19Rainforest Team
- 12:32approach: turn audio detection into an image classification problem, apply a deel learning architecture.
- 18:07Binary classification problem: model outputs probability that a spectrogram is chainsaw
- 23:11Approach: Source localization based on ti delay of arrival
- 23:31Unique challenges posed by rainforest deployment environment
related talks
15mThe Internet of Wild Things: Acoustics, AI and Conservation
17mTechnology: A Changemaker in Environmental Protection
46mSaving elephants with Raspberry Pi’s & Machine Learning
25mTechnology gone wild
27mLive and Machine Learn
29mBioacoustics and Machine Learning as Key Tools in Conservation
46mData Discussion - AI for Good
58mImproving disaster response with machine learning