Research on sunken & submerged oil detection and its behavior process under the action of breaking waves based on YOLO v4 algorithm

Marine oil spill pollution is one of the most serious marine pollution issues. Aiming at the problems of low accuracy and slow speed in the process of detecting the behavior of sunken and submerged oil by traditional methods, a technology of sunken & submerged oil tracking based on YOLO v4 (YOLO refers to ‘you look only once’) algorithm is proposed in this paper. The image data used in this study are pictures of real oil pollution moving under breaking waves, and they are collected in the laboratory. First, the YOLO v4 model under CSPDarknet53 framework was established. Then, in order to simplify the oil detection model and ensure the efficiency of the model, this research used Mosaic data enhancement, random flipping, and Gaussian noise fuzzy data enhancement, as well as Cosine Annealing Learning Rate, and Label Smoothing to improve the effect of deep learning model. After data enhancement, the final data set was divided into a training set and a test set proportionally. The training set had 878 pictures, and the test set had 1945 pictures. The test set contained the situation where oil droplets were completely occluded by waves, so that the detection accuracy was closer to the real situation. The results show that the oil droplet is hit and then sunk, forming ‘sunken and submerged oil’ under the action of breaking waves of wave heights of 10 cm, 15 cm, 20 cm, 25 cm and 30 cm. The submergence time enhances with the increase of wave height of breaking wave, that is, the residence time of oil droplet for 10 cm, 15 cm, 20 cm, 25 cm and 30 cm breaking waves is 2.32 s, 2.52 s, 2.62 s, 3.20s, 7.12 s, respectively. The deepest position of oil droplet under the water for 10 cm, 15 cm, 20 cm, 25 cm and 30 cm breaking waves is 0.165 m, 0.179 m, 0.226 m, 0.297 m, 0.428 m, respectively. However, the drift velocity and sinking velocity of oil droplet show nonlinear variation. The speed of sinking to the deepest is 0.208 m/s, 0.222 m/s, 0.212 m/s, 0.359 m/s, 0.303 m/s, respectively. •A technology of submerged oil detection and behavior process tracking is proposed.•Interdisciplinary research of marine science and environmental science is conducted.•Convolution layer is used to improve the ability of YOLOv4 to extract object feature.•Oil's submergence time enhances with the increase of wave height of breaking wave.•The drift and sinking velocities of sunken & submerged oil show nonlinear variation.