OilSAM2: Memory-Augmented SAM2 for Scalable SAR Oil Spill Detection

A research team led by Shuaiyu Chen has developed OilSAM2, a specialized AI framework that significantly improves oil spill detection in Synthetic Aperture Radar (SAR) satellite imagery. Built on Meta's Segment Anything Model 2 (SAM2) foundation, the system addresses critical limitations in current approaches that struggle with severe appearance variability, scale heterogeneity, and the absence of temporal continuity in real-world monitoring scenarios. Traditional SAM-based methods operate on single images and cannot effectively reuse information across scenes, while memory-augmented variants like SAM2 assume temporal coherence, making them prone to semantic drift when applied to unordered SAR image collections.

OilSAM2 introduces two key innovations: a hierarchical feature-aware multi-scale memory bank that explicitly models texture, structure, and semantic level representations, enabling robust cross-image information reuse; and a structure-semantic consistent memory update strategy that selectively refreshes memory based on semantic discrepancy and structural consistency to mitigate memory drift. The framework demonstrates state-of-the-art segmentation performance on two public SAR oil spill datasets, delivering stable and accurate results under noisy monitoring scenarios where traditional computer vision approaches typically fail. This represents a significant advancement for environmental monitoring applications that require processing large collections of unordered satellite imagery without temporal continuity assumptions.

The researchers have made the source code publicly available, enabling environmental agencies, oil companies, and monitoring organizations to implement more reliable detection systems. The approach's ability to handle the challenging characteristics of SAR imagery—including speckle noise, varying incidence angles, and complex ocean surface conditions—makes it particularly valuable for real-world deployment where image quality and ordering cannot be guaranteed. This work bridges the gap between foundation models and specialized domain applications, showing how memory-augmented architectures can be adapted for critical environmental monitoring tasks.