Oil film identification via windspeed interference estimation using hyperspectral data

Sea surface windspeed is one of the important factors that affects the reflection and radiation intensity of the sea surface oil film, posing challenges for the oil spill classification via marine optical remote sensing. A new method is proposed to extract significant windspeed features from the original hyperspectral remote sensing data and apply them to the qualitative analysis and quantitative estimation of local windspeed. First, a portable push-broom hyperspectral sensor was utilized to collect the hyperspectral data under different experimental windspeed conditions. After that the data was divided into spatial-spectral hyperspectral data along the spectrum slit direction and finally converted into spatial-spectral joint heat maps (SSHMs). Incorporating a transfer learning methodology, we utilized SSHMs as input to optimize the training process of windspeed estimation. This approach yielded dependable evaluation results. Moreover, we developed an effective deep learning framework that integrates spatial and spectral analyses for the categorization of sea surface oil films. This framework utilizes hyperspectral image datasets which have already removed the influence of windspeed variables. The proposed method demonstrated outstanding performance in estimating windspeed and classifying oil films using both experimental and AVIRIS datasets. This paper introduces a feasibility study for ocean surface windspeed estimation and precise identification of oil spills using optical remote sensing technology.