Research On Submarine Oil Pipeline Spill Detection Method Based On Cell UNet++ And Feature Judgment

As an economical and fast way of transportation,submarine pipeline is widely used.However,the submarine pipeline oil spill accident brings huge economic loss and environmental damage.The existing detection methods of submarine pipeline oil spill are mainly synthetic aperture radar(SAR)image detection and underwater sensor detection.With the development of deep learning algorithm and the technique of image processing,target detection algorithm based on machine vision is emerge in endlessly,the target detection method based on image compared with existing methods have good timeliness,the advantages of low cost.Therefore,it is of great practical significance to apply the image-based target detection method to the spill detection of submarine oil pipeline.At present,the spill detection of submarine oil pipeline is mostly used in simulation experiments.In this paper,a large number of detection methods have been studied for the spill video of real submarine oil pipeline.The main work contents are as follows:Firstly,this paper systematically analyzes the problem of submarine pipeline spill detection based on machine learning,puts forward the theoretical basis of submarine pipeline spill video detection,summarizes the problems existing in submarine pipeline spill image,and discusses the existing methods to solve these problems.A pipeline spill detection dataset is made by using the spill video collected on the network.Then,based on the obvious texture features of underwater oil spill targets,a new detection method based on texture super-pixel segmentation is proposed in this paper.It is improved based on UNet++ network to achieve segmentation of texture superpixels.Experimental results show that the proposed algorithm has higher accuracy than the traditional texture operator method and deep learning algorithmFinally,considering the dynamic and static characteristics of oil spill targets,an algorithm structure of cascading image processing decider for texture super-pixels classification is proposed to solve the problem of misdetection in the proposed classification results of texture super-pixels.A decider which fused multiple image features is cascaded into the output results of the texture super-pixels detector is used to eliminate the misdetection in the classification results of super-pixels texture.Experimental results show that the proposed classifier-decider cascading structure can effectively classify texture super-pixels and eliminate errors.Compared with existing methods,the algorithm proposed in this paper uses a small number of samples to realize spill detection of submarine oil pipelines.It has better timeliness and economic efficiency,and can be applied to seabed oil spill warning and other problems.