Efficient and autonomous processing and classification of images on small spacecraft

Small satellites and CubeSats are becoming an indispensable platform in space-industry development, however, these systems are severely resource-limited. Depending upon mission requirements and available communication bandwidth, it can take hours to days to downlink an image from a spacecraft to the ground. Improvements in sensors tend to generate increasingly larger data products. Since small spacecraft have limited storage space, it is crucial to filter and delete images that do not meet minimum science criteria. Depending on the mission, criteria may vary. Images can be prioritized based on having features such as high land percentage or a specific land color. Certain images are rarely useful, such as pitch-black images and cloud-filled images, and can be readily deleted. This research describes an autonomous image-classification framework to efficiently use downlink bandwidth by prioritizing image products with high science value for download while deleting others, as well as a training framework for classifier calibration.