Research

The development of future spacecraft is intricately linked to the advancement of onboard autonomy, with image processing playing a pivotal role in realizing this vision. However, the computational limitations inherent to spacecraft pose significant challenges when implementing sophisticated image-processing algorithms. Satellite image classification plays a pivotal role in augmenting spacecraft autonomy, facilitating capabilities such as identifying priority regions for transmission and adjusting mission schedules. Despite the presence of literature on onboard image processing, there remains a significant need for more specific references that comprehensively address the state-of-the-art advancements and future trends in the context of small satellite missions.

Support Vector Machines (SVM) have gained widespread recognition across various fields since their introduction in 1995, primarily due to their remarkable classification accuracy surpassing alternative techniques. However, the adoption of SVM in space applications has been limited, primarily attributed to its relative novelty compared to well-established competing methods.

The primary objective of this project is to delve into the unexplored realm of SVM applications in space. Two specific applications are proposed to unlock the potential of SVM in this domain. Firstly, SVM in classification mode will enhance onboard intelligence, empowering spacecraft to identify high-priority image data, such as natural disasters, without relying on previous image history. This approach assesses the feasibility of leveraging SVM by utilizing input features for adequate classification.

The second proposed application revolves around utilizing Support Vector Regression (SVR) in the critical area of Fault Detection, Isolation, and Recovery (FDIR) onboard spacecraft. FDIR plays a vital role in ensuring spacecraft autonomy, and the performance of SVR will be compared to existing approaches like Built-In Testing (BIT) to enhance fault detection and recovery processes.

To overcome the computational demands of machine learning in space, this project will employ a hardware-based implementation of SVM utilizing Field-Programmable Gate Arrays (FPGAs). The computational requirements of the software implementation will be analyzed, and its performance will be juxtaposed with the hardware implementation, paving the way for optimized SVM utilization in space applications.