Onboard-Targeted Segmentation of Broken Pixels, Lines, and Dust on Optics Faults in Space Camera Sensors

Riccardo Gallon, Alessandra Menicucci, Edoardo Caroselli

Keywords: Vision-Based Navigation, Artificial Intelligence, Anomaly Detection, Onboard Processing

TL;DR: A CNN-based approach to segmentation of camera faults, with focus on onboard compatibility and integration

Abstract: This study proposes a deep learning framework for the semantic segmentation of faults in space camera sensors, specifically addressing dust grains deposition on the optics and defect pixels and lines caused by radiation impinging the detector. The artifacts caused by these faults in the final image pose significant challenges in the field of thin structures segmentation. To address this, the DeepLabV3 architecture is employed, with various configurations evaluated to improve performance. A MobileNetV3 backbone is utilized to target compatibility with onboard deployment on resource-constrained hardware. Furthermore, an integration strategy to operate the model within the onboard navigation pipeline is proposed, highlighting its utility in improving autonomy and availability of the mission. These improvements are evaluated by means of integration-aware metrics in order to quantify the functional benefits of the proposed model.

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: 8