Methodology for oil film identification and feature extraction based on remote sensing data acquisition experiment

Marine oil spill incidents occur with alarming frequency, raising significant environmental and safety concerns. In the later stages of such incidents, detecting thin oil films on the sea surface poses substantial challenges. Firstly, this study conducted an offshore experiment to acquire ultraviolet remote sensing waveband image data of thin oil films with varying thicknesses. Subsequently, by conducting quantitative analysis on the acquired data and employing total variation vector calculation, we propose a novel method called adjacency total variation constraint method that yields superior outcomes exhibiting prominent structural characteristics within the sample dataset. At the same time, we innovatively designed a semi-supervised mechanism network framework named Oil Film Feature Extraction Constrained Network (TV-GAN), which is based on residual attention mechanism and generative adversarial network technology. The proposed framework enhances both data capacity and quality while effectively extracting supplementary feature information from the underlying area's structure. The experimental results demonstrate that this method exhibits robust performance in detecting thin oil films of varying thicknesses.