Building an on-board autonomous data acquisition and processing technology payload for small unmanned aircraft system

Abstract:

Aerial platform data collection is an essential component for timely and widespread monitoring of the environment and stewardship for its habitats. Transitioning these collection activities from fairly expensive manned missions to small autonomously operated unmanned aircraft systems (sUAS) provides significant benefits, especially as the community progresses towards beyond visual line of sight missions. sUAS operations impose strict payload strict size, weight, and power (SWaP) constraints, requiring sensor and processing components to be a tightly integrated lightweight, low-power payload. We present on the development of a low-SWaP data collection and processing payload that can be hosted on a variety of sUAS to support autonomous missions, Autonomous Data Acquisition and Processing Technologies (ADAPT). ADAPT will produce curated, geo-registered data archives, including vectorized maps, which will serve as key inputs to existing deep learning systems and applications. ADAPT’s in-flight capabilities greatly benefits sUAS missions to survey the boundaries between regions, such as land–water interfaces, habitat boundaries, or biomass fire perimeters. ADAPT produces pixel-wise labeled image maps, which will be distilled down to vectorized, geolocated boundaries between regions, providing significant storage savings and low-bandwidth transfer over a wireless downlink. This optimizes the data that reaches the operations center to assist them in more informed decision making. We use river-ice sUAS data from Circle, Alaska as the development test dataset and an offline browser-based computer vision annotation tool to build the database for ADAPT. We show how ADAPT will manage the data collection from onboard sensors and provide actionable intelligence through enabling in-flight, real-time application of deep neural network image processing. This work has been supported by the NOAA grant: NA20OAR0210083.