Demonstration of Wildfire Detection Using Image Classification Onboard CubeSat

Abstract

In the past decade, a massive number of wildfire events have been reported worldwide. An economic loss every year shows the importance of having satellites to overcome this catastrophe. Earth observation CubeSat can be a solution to detect, monitor, and provide data for the fire departments from the aerial view. For this purpose, the utilization of the deep learning (DL) algorithm to process the images captured onboard CubeSat before they are downlinked to the ground station is demonstrated. As a proof-of-concept, a single-board computer (SBC), Raspberry Pi, has been integrated into the KITSUNE 6U CubeSat to run the model. In this paper, the results of functional and environment tests of our proof-of-concept implementation have shown the feasibility of using image classification onboard the CubeSat. The DL algorithm's accuracy has reached 95% in the ground test, and this accuracy can be improved with further verifications of the platform and analysis of flight results. Therefore, DL's advancement in CubeSat is the state-of-art that can contribute significantly to address the wildfire issues.