Experimental scheme design of airborne sar combined thermal infrared and hyperspectral for oil spill detection
[Objective]Oil spills severely harm marine ecosystems,necessitating their rapid and accurate detection.Remote sensing,with its unique advantages of high-vantage,long-range,comprehensive observation,has become an important tool for offshore oil spill detection.Imaging characteristics of microwave remote sensing,thermal infrared,and hyperspectral methods suffer limitations in the extraction of oil film information.Oil spill detection using combined satellite data from multiple sensors is very challenging because of rapid changes in oil spills at sea,difficulty in simultaneous multisensor observation,and considerable observational angle deviations.To address these issues,this study designs an experimental plan for oil spill detection that combines airborne synthetic aperture radar(SAR)with airborne thermal infrared sensors and airborne hyperspectral sensors.[Methods]The plan includes the acquisition of data,construction of multisource datasets,detection of oil spills,and analysis of multisource data characteristics.A field simulation oil spill experiment was conducted from October 22 to 23,2022.Unmanned aerial vehicles were used to simultaneously acquire airborne SAR,airborne thermal infrared,and airborne hyperspectral data.The remote sensing data were preprocessed,and image registration was performed.The support vector machine(SVM)and random forest(RF)methods were used for oil spill detection,and the advantages associated with the incorporation of the hyperspectral and thermal infrared data into SAR-based oil spill detection were analyzed.Differences in the appearance of crude oil,clean seawater,and ice between SAR,thermal infrared,and hyperspectral images were examined.[Results]1)Upon comparing various detection schemes on single SAR data in Dataset#1,the optimal SAR data was found to be the combination of the SAR backscattering data and SAR texture data,with F1-Score being 24.21%for the SVM model and 40.56%for the RF model.2)Combining the airborne SAR data with the thermal infrared data,and combining the airborne SAR with the hyperspectral data,showed some improvement in oil spill detection accuracy compared with using airborne SAR data alone.3)Incorporation of airborne thermal infrared data and airborne hyperspectral data on the basis of SAR data can result in the best oil spill detection effect.For the airborne SAR combined airborne thermal infrared and airborne hyperspectral detection scheme,the F1-Score of the RF model detection on Dataset#1 reached 72.74%with notably reduced mismarking.F1-Score of the airborne SAR combined airborne thermal infrared and airborne hyperspectral detection scheme was 6.39%higher than that of the airborne SAR combined with airborne thermal infrared detection scheme and 4.65%higher than that of the airborne SAR combined with airborne hyperspectral detection scheme.On Dataset#2,F1-Scores of SVM and RF reached 99.52%and 99.64%,respectively.4)By analyzing the backscattering characteristics of crude oil,seawater and ice in Ku-band airborne SAR data,it is verified that the backscattering coefficient of crude oil is lower than that of seawater,while that of ice is similar to that of crude oil and difficult to distinguish.By analyzing the spectral curve extracted from the airborne hyperspectral data,it is found that the spectral reflectance of ice is the highest and that of seawater is greater than that of crude oil.From the airborne thermal infrared data,it is confirmed that the bright temperature value of crude oil is higher than that of seawater.[Conclusions]The use of multidimensional remote sensing data provides technical support for the application of multisource remote sensing methods in marine oil spill detection.
万方数据
维普
