YOOLO: You Only Orbit and Look Once

Vincenzo Mariano Scarrica, Antonio Vanzanella, Antonino Staiano

Keywords: spacecraft detection, spacecraft classification, spacecraft segmentation

Abstract:

In the context of the SPARK 2026 Challenge, a deep learning pipeline for the simultaneous classification, detection, and segmentation of spacecraft orbiting Earth is developed. The challenge provides a large-scale synthetic dataset of orbiting spacecrafts, including image-level class labels, bounding boxes, and pixel-wise segmentation masks for ten real satellite platforms. Participants are required to design compact models capable of jointly solving classification, detection, and segmentation tasks under strict efficiency constraints.

In this work, YOOLO (You Only Orbit & Look Once) is introduced as a compact and efficient deep learning pipeline capable of addressing multiple computer vision tasks within a unified framework. A YOLO11s-seg model is trained on the full training set for 20 epochs, achieving an accuracy score of 0.938 on the test set. The results demonstrate that YOOLO effectively balances model compactness and predictive performance, delivering encouraging classification, detection, and segmentation outcomes while ensuring fast inference. This makes the proposed approach promising for resource-constrained space-domain applications.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 26