CDD & CloudNet: A Benchmark Dataset & Model for Object Detection Performance

Aerial imagery obtained through remote sensing is extensively utilized across diverse industries, particularly for object detection applications where it has demonstrated considerable efficacy. However, clouds in these images can obstruct evaluation and detection tasks. This study therefore involved the compilation of a cloud dataset, which categorized images into two classes: those containing clouds and those without. These images were sourced from the publicly available Maxar ‘Hurricane Ian’ repository, which contains images from various natural events. We demonstrated the impact of cloud removal during pre-processing on object detection using this dataset and employed six CNN models, including a custom model, for cloud detection benchmarking. These models were used to detect objects in aerial images from two other events in the Maxar dataset. Our results show significant improvements in precision, recall, and F1-score for CNN models, along with optimized training times for object detection in the CloudNet+YOLO combination. The findings demonstrate the effectiveness of our approach in improving object detection accuracy and efficiency in remote sensing imagery, particularly in challenging cloud-covered scenarios.