Session
Advanced Technologies 1
Location
Salt Palace Convention Center, Salt Lake City, UT
Abstract
Earth observing satellites contribute to the monitoring of maritime activity by enabling large-scale coverage of open water. For example, spacecraft data can be used to monitor and combat Illegal, Unreported, and Unregulated (IUU) fishing that occurs in coastal and international waters across the globe. However, information latencies limit the utility of space-based sensors for operational users who need real-time situational awareness. These limitations are especially pronounced for small spacecraft with limited power generation and downlink capabilities. In recent years, the devices and tools that enable edge computing and near real-time downlink from small spacecraft have improved substantially. Equipped with these technologies, spacecraft can reduce large amounts of raw data into actionable notifications for rapid response. For IUU fishing interdiction, small spacecraft could provide maritime law enforcement with near real-time imagery of the current position of fishing fleets. Current edge computing technology enables hardware acceleration of Artificial Intelligence (AI)-based computer vision models that can fit within the Size, Weight, and Power (SWaP) constraints of small spacecraft platforms. However, limited work has characterized image classification models for maritime applications on edge processing devices that can fit within these SWaP constraints. This paper examines the performance of Convolutional Neural Network (CNN)-based classification models implemented on two flight-like embedded devices. It was found that the latest technology can result in more than an order of magnitude throughput as well as more power efficient inferencing compared to previous work characterizing an earlier generation of edge devices. This performance improvement suggests that small spacecraft equipped with current edge computing and near real time downlink capabilities can now enable missions which significantly improve the rate of decision making for operational users.
Document Type
Event
Evaluating Performance of Hardware-Accelerated Vessel Detection Algorithms
Salt Palace Convention Center, Salt Lake City, UT
Earth observing satellites contribute to the monitoring of maritime activity by enabling large-scale coverage of open water. For example, spacecraft data can be used to monitor and combat Illegal, Unreported, and Unregulated (IUU) fishing that occurs in coastal and international waters across the globe. However, information latencies limit the utility of space-based sensors for operational users who need real-time situational awareness. These limitations are especially pronounced for small spacecraft with limited power generation and downlink capabilities. In recent years, the devices and tools that enable edge computing and near real-time downlink from small spacecraft have improved substantially. Equipped with these technologies, spacecraft can reduce large amounts of raw data into actionable notifications for rapid response. For IUU fishing interdiction, small spacecraft could provide maritime law enforcement with near real-time imagery of the current position of fishing fleets. Current edge computing technology enables hardware acceleration of Artificial Intelligence (AI)-based computer vision models that can fit within the Size, Weight, and Power (SWaP) constraints of small spacecraft platforms. However, limited work has characterized image classification models for maritime applications on edge processing devices that can fit within these SWaP constraints. This paper examines the performance of Convolutional Neural Network (CNN)-based classification models implemented on two flight-like embedded devices. It was found that the latest technology can result in more than an order of magnitude throughput as well as more power efficient inferencing compared to previous work characterizing an earlier generation of edge devices. This performance improvement suggests that small spacecraft equipped with current edge computing and near real time downlink capabilities can now enable missions which significantly improve the rate of decision making for operational users.
