Enhancing Offshore Infrastructure Monitoring: Synthetic Data Generation for Deep Learning-Based Object Detection on Sentinel-1 Radar Imagery

The recent and ongoing expansion of marine infrastructure, including offshore wind farms, oil and gas platforms, artificial islands, and aquaculture facilities, highlights the need for effective monitoring systems. Precise quantification in space and time is crucial to planning the future expansion, usage, management, and impact of marine offshore infrastructure. In the past decade, numerous studies have explored the detection and monitoring of offshore infrastructure using space-borne data and remote sensing techniques. Recently, deep learning-based approaches have emerged as a powerful tool for these tasks. However, the development of robust and reliable object detection models depends on the availability of comprehensive, balanced training datasets. Manual annotation of existing objects is the standard method for dataset creation, but it falls short when samples are scarce, particularly for underrepresented object classes, shapes, and sizes. To address this limitation, we propose a deep learning-based approach for generating synthetic training data by modifying and retraining a stable diffusion model. The goal of this approach lies within the augmentation of manual image-label pairs and the enhancement of the dataset quality and diversity. We validate this approach by applying the object detector YOLOv10 to efficiently detect and classify offshore infrastructure objects (specifically offshore oil and gas platforms) on Sentinel-1 radar imagery in three diverse test regions: the Gulf of Mexico, the North Sea, and the Persian Gulf. We will present an analysis of the impact of our synthetic data generation approach on training results with a focus on how unbalanced classes can be better represented and model performance improved. This study underscores the critical importance of balanced datasets and highlights synthetic data generation as an effective strategy to address common challenges in remote sensing. Furthermore, it reaffirms the pivotal role of Earth observation in advancing offshore infrastructure monitoring by demonstrating the first test results of our model on unseen data.