Onboard Image Classification Unit Implementation for AlAinSat-1 CubeSat

This study presents the implementation of an onboard Image Classification Unit (ICU) for the AlAinSat-1 CubeSat, aiming to enhance its autonomy and data processing capabilities. The focus is on integrating a trained CNN model onto AlAinSat-1’s STM32 microcontroller.The designed system employs TensorFlow models trained for image classification tasks relevant to CubeSat missions, such as target accuracy detection, to determine if the image captures the required target, and image quality assessment to estimate cloud cover percentage.The integration process onto the STM32 microcontroller involves addressing the resource constraints inherent in CubeSat platforms. The paper details the optimization techniques applied to adapt the model to the STM32 architecture, ensuring efficient execution within the available hardware resources.Key aspects covered in the study include hardware-software co-design considerations, addressing memory and computational limitations, and optimizing mission duration for power consumption efficiency. Additionally, the development of a reliable communication interface between the onboard Image Classification Unit (ICU) and CubeSat’s main control system is discussed to facilitate seamless integration into the overall satellite architecture.The presented implementation enables CubeSats to perform onboard image classification tasks, reducing the need for constant communication with ground stations and enabling quicker response times for mission decisions. This research contributes to the growing field of embedded machine learning applications in spaceborne remote sensing systems, showcasing the feasibility and benefits of incorporating image processing capabilities on resource-constrained platforms.