Abstract
In recent years, the growing amount of space debris has posed a significant threat to space operations. On-orbit visual sensors used to detect space debris have emerged as a promising solution for maintaining spacecraft safety. However, the performance of visual sensors is highly dependent on illumination, which is often insufficient under low-light conditions. Since visible and thermal images complement each other effectively in such environments, a visible and thermal image fusion technique is utilized to enhance the application of visual sensors. A dataset containing recorded visible and thermal images of space debris was developed for model validation. This paper presents a novel approach for the fusion of visible and thermal images, followed by the classification of the fused images into space debris or satellites. The fused images are then fed into a classifier to distinguish between space debris and satellites. Experimental results are compared using different classification algorithms to identify the most effective algorithm for achieving better performance. The proposed approach shows promise for the accurate identification and categorization of objects in space. Future work will focus on improving real-time implementation, expanding the dataset to include diverse orbital scenarios and exploring advanced deep learning models to further enhance detection accuracy. Additionally, the motion of space objects may be incorporated into the classification process, using video fusion techniques.
Keywords: Deep Learning, Image Classification, Image Fusion, Space Debris, Thermal Imaging