Autonomous Data Acquisition and Processing Payload for Small Unmanned Aircraft Systems

In response to NOAA-OAR-OAR TPO-2019-2005899 topic 9.5.01 , Kitware proposes a compact data collection and processing payload that can be hosted from small unmanned aircraft systems to support autonomous beyond line of sight missions. This innovative payload will manage collections from onboard sensors with optimizations collection paradigms. It will produce curated, geo-registered data archives, which can be exfiltrated and ingested into Kitware’s VIAME application. VIAME, which is in use at all six NOAA National Marine Fisheries Science Centers, allows NOAA personnel to annotate data and train deep neural network models providing mission-critical image processing. These models can be uploaded back onto the payload to enable real-time, in flight processing. For large-scale surveys of ice mammals, in-flight animal detection allows the majority of imagery without sufficient evidence to be omitted from the data archive, allowing greater ground coverage before filling onboard storage. Autonomous, online mission planning could incorporate real-time detections to adjust the flight plan for higher-quality data where animals are detected by revisiting at lower altitude or with additional sensors. Another example is in-flight semantic image segmentation of boundaries between regions, such as land–water, habitat, biomass fire interfaces, which could be used to alter flight path for optimal collection.