Abstract: Oil spill and its emulsifications are a serious threat to the Marine environment. As a powerful earth observation technology, hyperspectral remote sensing plays a role in the early detection and qualitative analysis of Marine oil spill. Given the problem of large hyperspectral data volume and limited airborne computing resources, this paper presents a model for identifying lightweight emulsified oil, it combines incremental sparse principal component analysis methods (Incremental Sparse Principal Component Analysis, ISPCA), with the lightweight deep learning network SqueezeNet. The method is suitable for real-time data flow and large-scale data sets, from which the features can be extracted continuously, improved the response speed, and applied in the field simulation and real oil spill scene acquisition to carry out emulsified oil identification experiment. The experimental results show that the model can effectively realize the identification of oil spill emulsion and background seawater, with the overall accuracy of 83.4% and Kappa coefficient of 0.81. The lightweight model can process the data in batches, avoid the memory pressure caused by loading all the data at once, and shows stronger information retention ability and lower mean square error. The parameter optimization can ensure high identification accuracy and stability. the identification time is only 110 seconds, which is 47% shorter compared with conventional principal component analysis. In conclusion, this work not only contributes a lightweight oil spill emulsion identification model suitable for large-scale datasets and online learning scenarios, but also provides a new idea for model optimization, which integrating deep learning and real-time environmental factors, demonstrating the great potential of hyperspectral remote sensing technology in solving disaster emergency problems.