We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
基于耀光反射差异的海面溢油遥感识别提取.
- Authors
朱小波; 沈亚峰; 刘建强; 丁静; 焦俊男; 居为民; 陆应诚
- Abstract
After an oil spill accident, it is essential to quickly detect the location, spatial coverage, pollution type, and amount of oil spills, so as to measure the impacts of different oil spill types, clean up the oil, and help the ocean recover. Although the mechanism and characteristics of optical remote sensing of oil spills have been basically clarified, the research on automatic oil spill extraction algorithms is still insufficient, which still needs to be addressed. The main challenge is that the significant variation of sunglint is helpful to the oil spill identification, but also brings many uncertainties to the extraction. Hence, the marine remote-sensing community is always committed to developing remote-sensing methods for improving the performance of aspects, such as preprocessing, segmentation, and classification. The scale of extraction is deemed the key to eliminating sunglint interference. As the conventional extraction method cannot be applied on the optical images directly for the reasons outlined in the context, a new man-machine interactive oil spill extraction method (more specific an oil–water mixture detector), which is able to eliminate sunglint interference was introduced. It accomplishes this by splitting the oil spill global region into adjacent sub-windows, as the sunglint can be considered constant in a small region. The proposed method and densitybased clustering is used cooperatively in the method. The detector first discretizes different types of oil spills in images based on the specific optical detection principle of oil spills. Then the clustering uses the auxiliary information of multispectral images to achieve high confidence clustering output. The Coastal Zone Imager (CZI) onboard China’s HaiYang-1C/D (HY-1C/D) satellites can provide multispectral images with high spatial resolution and wide coverage for operational monitoring of oil spills. The proposed method was applied to HY/CZI oil spill dataset and other optical images of several oil spill events. The spatial differentiation of sunglint and remote sensing response characteristics of different oil spills in CZI were analyzed, and the accurate identification and extraction of oil spills in CZI images were realized. The results show that the optical remote sensing extraction method considering the variation of sunglint can effectively identify and extract the oil spills, and has good anti-interference ability. Based on testing of CZI data covering different areas, the method was proved to be effective in removing interference caused by sunglint, image interference (e.g. cloud, rough surface texture, ship wakes, illumination, shadowing), and so on. In addition, the method can further distinguish oil slicks and oil emulsions under the condition of weak sunglint, showing the ability to identify different oil spills, which can provide a reference for the operational application in oil spill monitoring. The results show stable variable-scale extraction accuracies of approximately 90.24% and 80.55% for oil slicks and oil emulsions, respectively. It is also applicable for the optical images with lower resolution, but the effect is inferior to that in CZI because of the influence of mixed pixels. As aforementioned, the accurate results are attributed to appropriate parameter adjustment under different spatial resolutions, oil spil types, and sunglint reflections. The satisfactory transfer applicability is mainly attributed to the variable-scale detector for sunglint reflection differences. In summary, the precondition for accurate oil spill extraction is to eliminate the sunglint interference, which depends on not only the sunglint model but also appropriate local scale, not global. Without a doubt, to what extent the coordination and utilization of sunglint and spectral information is the key to breakthrough for accurate oil spill extraction and quantification. After an oil spill accident, it is essential to quickly detect the location, spatial coverage, pollution type, and amount of oil spills, so as to measure the impacts of different oil spill types, clean up the oil, and help the ocean recover. Although the mechanism and characteristics of optical remote sensing of oil spills have been basically clarified, the research on automatic oil spill extraction algorithms is still insufficient, which still needs to be addressed. The main challenge is that the significant variation of sunglint is helpful to the oil spill identification, but also brings many uncertainties to the extraction. Hence, the marine remote-sensing community is always committed to developing remote-sensing methods for improving the performance of aspects, such as preprocessing, segmentation, and classification. The scale of extraction is deemed the key to eliminating sunglint interference. As the conventional extraction method cannot be applied on the optical images directly for the reasons outlined in the context, a new man-machine interactive oil spill extraction method (more specific an oil–water mixture detector), which is able to eliminate sunglint interference was introduced. It accomplishes this by splitting the oil spill global region into adjacent sub-windows, as the sunglint can be considered constant in a small region. The proposed method and densitybased clustering is used cooperatively in the method. The detector first discretizes different types of oil spills in images based on the specific optical detection principle of oil spills. Then the clustering uses the auxiliary information of multispectral images to achieve high confidence clustering output. The Coastal Zone Imager (CZI) onboard China’s HaiYang-1C/D (HY-1C/D) satellites can provide multispectral images with high spatial resolution and wide coverage for operational monitoring of oil spills. The proposed method was applied to HY/CZI oil spill dataset and other optical images of several oil spill events. The spatial differentiation of sunglint and remote sensing response characteristics of different oil spills in CZI were analyzed, and the accurate identification and extraction of oil spills in CZI images were realized. The results show that the optical remote sensing extraction method considering the variation of sunglint can effectively identify and extract the oil spills, and has good anti-interference ability. Based on testing of CZI data covering different areas, the method was proved to be effective in removing interference caused by sunglint, image interference (e.g. cloud, rough surface texture, ship wakes, illumination, shadowing), and so on. In addition, the method can further distinguish oil slicks and oil emulsions under the condition of weak sunglint, showing the ability to identify different oil spills, which can provide a reference for the operational application in oil spill monitoring. The results show stable variable-scale extraction accuracies of approximately 90.24% and 80.55% for oil slicks and oil emulsions, respectively. It is also applicable for the optical images with lower resolution, but the effect is inferior to that in CZI because of the influence of mixed pixels. As aforementioned, the accurate results are attributed to appropriate parameter adjustment under different spatial resolutions, oil spil types, and sunglint reflections. The satisfactory transfer applicability is mainly attributed to the variable-scale detector for sunglint reflection differences. In summary, the precondition for accurate oil spill extraction is to eliminate the sunglint interference, which depends on not only the sunglint model but also appropriate local scale, not global. Without a doubt, to what extent the coordination and utilization of sunglint and spectral information is the key to breakthrough for accurate oil spill extraction and quantification.
- Subjects
CHINA; OPTICAL remote sensing; WAKES (Fluid dynamics); OIL spills; HIGH resolution imaging; MULTISPECTRAL imaging; SURFACE texture; OIL spill cleanup
- Publication
Journal of Remote Sensing, 2023, Vol 27, Issue 1, p197
- ISSN
1007-4619
- Publication type
Article
- DOI
10.11834/jrs.20221688