Deep Learning-Based Image Processing for Whitecaps on the Ocean Surface

<p> Whitecaps generated by wave breaking on the ocean surface play an important role in the local interaction across the air-sea interface. Whitecap coverage is defined by the area of whitecaps per the unit ocean surface. It has been recognized as one of the most valuable physical quantities for describing the ocean surface fluxes such as the momentum, heat and carbon dioxide, so that the quantitative evaluation of whitecap coverage becomes significant from viewpoints of coastal and ocean engineering. In this study, a progressive high-precision whitecap extraction model is first built by using the algorithm of deep learning. Compared with a traditional whitecap extraction model based on threshold value, the algorithm is found to solve problems caused by illuminance condition and color change on the ocean surface, and effectively extracts fine whitecaps with complicated structures. Further, through comparisons with previous algorithms such as Automatic Whitecap Extraction (AWE), Iterative Between Class Variance (IBCV) and the whitecap extraction based on fixed threshold value, the present algorithm is demonstrated to be more accurate for identifying whitecaps, and it reduces the amount of evaluation load, and can effectively apply for changeable ocean conditions. The new whitecap extraction technology is used to determine whitecap coverage when shooting digital images under complicated sea surface conditions. Due to the progressive characteristics of this algorithm, it has not only a high precision processing effect on images taken by a fixed camera, but also has the potential to analyze accurately images from a non-fixed camera system, such as an observation ship equipped with camera system, unmanned aerial vehicle and so on.</p>