Abnormal Detection And Recognition Of Ship Oil Spill Targets In Hyperspectral Images

Oil spill on the sea has seriously affected the marine environment and maritime transportation,and these oil spills mainly come from oil storage bases,offshore drilling platforms and ships.After the oil spill,it is very important to grasp the distribution information of the oil spill in real time for the subsequent emergency decision making and cleaning work.The remote sensing technology which can obtain large area information plays an important role in oil spill monitoring.Compared with radar,laser and multispectral image data,hyperspectral remote sensing image has the characteristics of wide range,continuous spectrum and high dimension feature.It plays an important role in environmental monitoring.In recent years,with the development of sensor technology and machine learning technology,hyperspectral image recognition technology is gradually developing to spectral-spatial recognition technology and the direction of deep learning.However,at present,in the study based on oil film recognition in hyperspectral data,the technology of spectral-spatial and deep learning is seldom used.In terms of system integration,the current research is only aimed at a certain model,and has not integrated multiple models to form software system.Aiming at the above problems and requirements,the paper has studied the relative thickness of oil film and the preliminary detection of ships based on hyperspectral data.First,the method based on anomaly detection is used to find the both ROI of oil film and ROI of ship.Then applying the SVM,BP-neural network and SVM-SAE models to distinguish the oil spill of ship.The SVM-SAE's overall accuracy was 71%,and Kappa was 0.635 in noise data sets,achieved the extraction of the information of the relative thickness of the oil spilled on the sea surface.Second,SVM-SAE's overall accuracy was improved by 2%in noisy data with spatial and spectral characteristics.The CNN model established is suitable for the identification of oil film in the paper.The model's AUC value is 0.72 and 0.96 on the validation dataset and training dataset,and it solves the problem of spatial homogeneity and heterogeneity.Finally,integrating the algorithms involved in the paper to form a trainable and reload model software system.In this paper,technology of spectral-spatial and deep learning is used to the target recognition of oil spill of ship.It provides a complete set of technical solutions for marine oil spill information acquisition,and integrates various algorithm models.The integrated system has strong practical application value,and provides effective technical support for emergency and disposal of ship oil spill accidents.