Study On The Recognition Of Bubble Curtains Based On Pneumatic Oil Barriers

With the development of Chinese marine shipping industry,the number of ships entering and leaving ports is growing rapidly,so the number of oil spill accidents is increasing.This phenomenon not only affects the ecological health of the nearby waters,but also causes serious adverse reactions to the coastal social economy.Preventing the spread of oil spills can effectively reduce the harm degree of oil spills.The solid float type oil containment boom,which is widely used now,has poor operability and can not adapt to the changes of meteorological conditions of walrus in port,resulting in poor contain oil spills.It is of great significance to develop a dynamic and intelligent pneumatic oil barriers to adapt to the changeable meteorological conditions of the maritime meteorological.Maritime meteorological accelerates the diffusion of oil spill.It is very important to strengthen the effect of contain oil spills by the intelligent monitoring and perception of the pneumatic oil barriers according to the condition of oil spill.Through dynamic monitoring of the bubble curtain state of the water surface,use image recognition technology to accurately identify the position information of bubble curtain,and combined with the distribution of spilled oil to evaluate the effect of oil containment.Set up the experimental platform of the pneumatic oil barriers,carry out the oil containment effect experiment of the pneumatic oil barriers,and combine the UAV real-time monitoring and sampling,According to the characteristics of the bubble curtain images under the natural environment,analyze the shortcomings of the existing feature extraction and recognition classification methods,a classification network of bubble curtain recognition based on Mask RCNN is proposed.The main contents of this paper are as follows:(1)Using UAV to take the working image of the pneumatic oil barriers,the collected image of the bubble curtain is marked with VIA annotation tool and made into the bubble curtain data set;(2)The existing convolution neural network and feature recognition algorithm are analyzed,combined with the characteristics of bubble curtain,Mask RCNN network is selected as the basic framework.This paper introduces residual network,regional recommendednetwork,ROI align layer and Fast RCNN + Mask segmentation are given more description.Resnet101 is selected as the residual network,and the coco data set is used to pre train the network.Fine tuning and testing are carried out on the labeled bubble curtain data set to complete the detection model.Identify the edge of bubble curtain,and output all edge point sets of bubble curtain;(3)The GUI interface of bubble screen recognition is designed.After the bubble curtain recognition based on Mask RCNN model,according to the usability,simplicity and consistency of GUI system design,the GUI interface of bubble screen recognition is designed.