关键词: 海面溢油探测, 多模态遥感数据, 多模态优势互补, 光谱空间特征, 语义分割

Abstract: As a sudden Marine pollution event, marine oil spills pose a serious threat to marine and coastal ecosystems as well as human safety. With the rapid development of aerospace technology, remote sensing platforms equipped with various sensors have become the key to obtain high-resolution and high-quality imagery of marine oil spills. Satellite-borne remote sensing platforms enable large-scale and real-time detection and monitoring of oil spills, while airborne platforms based on fixed-wing aircraft, unmanned aerial vehicles (UAVs) and other equipment offer flexible and accurate on-site data acquisition, providing effective data support for the implementation of subsequent relevant emergency response measures. Considering that multi-modal remote sensing imagery can capture multi-dimensional features of targets and significantly improve detection accuracy and robustness, this paper proposes a multi-modal semantic segmentation model for marine oil spill detection based on airborne Hyperspectral Image (HSI) data and satellite-borne Synthetic Aperture Radar (SAR) data. The model utilizes a 3D convolution module to extract spectral-spatial features from HSI data, serving as a supplement to SAR features. a Multi-modal Complement Atrous Spatial Pyramid Pooling module is designed to extract SAR features at multi-scale and fuse the strengths of both modalities, mitigating the limitations of single-modality. By introducing directional gradient perception and coordinate attention, a Differential Localization Enhancement Module is constructed to enhance the perception ability of the model for the edge and small area of the oil spill.Experimental results on the Gulf of Mexico Database (GMD) demonstrate that the proposed model achieves 99.43, 99.41% and 99.47% in Kappa coefficient (×100), mIoU and F1 score, respectively, outperforming the selected comparison models. These results show the model’s strong oil spill feature extraction and detection capabilities under the support of aerospace remote sensing information. The research results can provide technical support for the intelligent monitoring of the marine environment with the integration of the integrated space-air observation systems.

Key words: oil spill detection, multi-modal remote sensing data, multi-modal complementarity, spectral spatial feature, semantic segmentation