Enhancing Ship Detection in Remote Sensing: A Data Augmentation Approach Using State-of-the-Art Text-to-Image Diffusion

Ship detection in remote sensing imagery is vital for maritime surveillance and security. However, detecting small vessels in complex environments remains difficult due to limited data diversity. We propose a novel augmentation pipeline combining large-language and diffusion models to enrich training data. Scene descriptions generated by ChatGPT are used to guide FLUX, a text-to-image diffusion model, in synthesizing diverse ship imagery. A hybrid labeling strategy using open-vocabulary object detector YOLO-World and human annotators improves synthetic data accuracy. The synthetic dataset enhances diversity of training data and boosts detection accuracy. Experiments on ShipRSImageNet demonstrate better results across detectors, notably with YOLOv11 increasing mAP from 0.518 to 0.527 after incorporating synthetic data. These outcomes emphasize the proposed method’s effectiveness in enhancing performance of ship detection models.