Session

Weekend Session 8: Next on the Pad

Location

Utah State University, Logan, UT

Abstract

Bronco Ember is a nascent wildfire detection system that leverages edge computing capabilities, multi-spectral imaging, and artificial intelligence to greatly increase the performance of small satellite remote sensing payloads. The core hardware onboard is a SWIR InGaAs camera imaging in the 900nm to 1700nm wavelength and a GPU enabled single board computer. Artificial intelligence is used for fire detection and analysis using computer vision and neural networks being able to detect fires only filling a few pixels in each image. The system is based on traditional CNN networks and includes time series analysis that gives the system an 85% success rate in being able to detect wildfires with about a 50m diameter from a high-altitude balloon technology demonstration flight. The neural net is trained to monitor the movement and spread of the fire compared to prediction maps. This greatly reduces the number of false positive detected. The development of this payload has been supported through the NASA TechLeap Autonomous Observation Challenge No. 1 that has pushed the technology from concept to test flight in less than one calendar year. The system acts a rapid response remote sensing technology.

Share

COinS
 
Aug 7th, 5:45 PM

Bronco Ember An Edge Computing Acceleration Platform with Computer Vision

Utah State University, Logan, UT

Bronco Ember is a nascent wildfire detection system that leverages edge computing capabilities, multi-spectral imaging, and artificial intelligence to greatly increase the performance of small satellite remote sensing payloads. The core hardware onboard is a SWIR InGaAs camera imaging in the 900nm to 1700nm wavelength and a GPU enabled single board computer. Artificial intelligence is used for fire detection and analysis using computer vision and neural networks being able to detect fires only filling a few pixels in each image. The system is based on traditional CNN networks and includes time series analysis that gives the system an 85% success rate in being able to detect wildfires with about a 50m diameter from a high-altitude balloon technology demonstration flight. The neural net is trained to monitor the movement and spread of the fire compared to prediction maps. This greatly reduces the number of false positive detected. The development of this payload has been supported through the NASA TechLeap Autonomous Observation Challenge No. 1 that has pushed the technology from concept to test flight in less than one calendar year. The system acts a rapid response remote sensing technology.