A Deep Learning Approach for Satellite and Debris Detection: YOLO in Action

Satellite detection refers to the identification and recognition of satellites in satellite images. This task is crucial for various applications, including space surveillance, satellite tracking, and space debris monitoring. Several studies have been conducted to develop methods and techniques for satellite detection. The intricacy and scale of satellite images make satellite detection incredibly challenging. The process of manual identification and classification necessitates a significant amount of time. Recently, Deep Learning (DL) has emerged as a highly promising approach in Computer Vision (CV) tasks. This study evaluates the effectiveness of three widely employed DL-based object detection algorithms, specifically YOLOv4, YOLOv5, and YOLOv8, within the field of satellite detection. The algorithms have undergone training utilizing a dataset of satellite photos comprising both satellites and debris. The evaluation of the algorithm’s performance involved the utilization of broadly accepted metrics, including Precision, Recall, Mean Average Precision (mAP), Box Loss, and Class Loss. The findings indicated that all three algorithms exhibited a high level of accuracy in detecting and precisely determining the location of objects inside satellite data. The performance of YOLOv5 and YOLOv8 surpassed that of YOLOv4 across various measures, with a notable advantage observed in detecting tiny objects. Overall, this study demonstrates the capabilities of DL algorithms in the context of satellite detection and classification techniques.