Research On Oil Spill Identification From Ships Based On Remote Sensing Monitoring Data

The research on identifying oil spills from ships holds significant importance for monitoring unlawful maritime activities,ensuring marine environmental safety,and aiding in marine disaster prevention and mitigation.Current methodologies for detecting ship oil spills face challenges,including delays in detection,inefficiencies in identification,and shortcomings in traceability.This thesis employs the modern remote sensing tool of unmanned aerial vehicles(UAVs)to conduct research on the identification of ship oil spills based on image data.It explores methods for identifying oil spills,measuring the extent of oil-contaminated areas,and developing traceability techniques for oil spill pollution.This work aims to provide a theoretical foundation for the prevention,control,and investigation of marine oil spill incidents.Firstly,from the perspective of ship oil spill detection,a machine learning method for oil spill identification based on the histogram of oriented gradient(HOG)feature combined with a support vector machine(SVM)is proposed.This method takes the thermal infrared image as the research object,extracts the HOG feature of the sample by preprocessing the image,and uses the radial basis function as the kernel function to train the SVM classifier,realizing the all-day fast recognition of oil spills.Compared with BP and LBP-SVM,it is found that this method has high recognition accuracy,with the recognition accuracy of oil film test samples reaching up to 91.3%,which can meet certain application requirements.Secondly,addressing the issue of incomplete recognition due to single-feature extraction in environments with large sample sizes,a machine learning oil spill recognition method based on the fusion of Histogram of Oriented Gradient(HOG)and Gray-Level Co-occurrence Matrix(GLCM)features,combined with an Extreme Learning Machine(ELM),is proposed.This approach enhances the infrared oil spill image sample data and employs the combined HOG and GLCM features for feature extraction,which are then input into the ELM classification model to perform oil spill identification.Repeated experimental results demonstrate that,in a large sample size environment,the oil spill identification method utilizing HOG-GLCM multi-feature fusion with ELM not only ensures training time and recognition efficiency but also significantly improves the recognition accuracy rate,reaching as high as 96%.Thirdly,from the perspective of oil spill cleanup decision-making and claims,an improved Canny algorithm based on the principle of the pixel area method is proposed to measure the oil spill area.The feasibility of this method is confirmed through the collection of experimental data from oil spill simulation experiments in a tank,followed by calculating the oil film area.The simulation results demonstrate that the improved Canny algorithm effectively extracts the edges of oil films from crude oil and diesel oil.Moreover,the calculation error of the oil spill area under calm water conditions is less than 6% within the range of low altitude and small aerial photography angles.This makes it suitable for the preliminary estimation of oil spill areas in low-altitude and small-scale waters near the sea.Finally,in cases where AIS is not installed or terminals are shut down,this thesis proposes an oil spill tracing method that leverages coordination between satellites and aircraft.Employing local threshold segmentation theory,this study investigates the feasibility and performance of the maximum inter-class variance method,the maximum entropy method,and a maximum entropy method enhanced by a genetic algorithm for segmenting remote sensing images in large-scale water areas using optical satellite data.This approach facilitates the initial identification of oil spill pollution sources based on contour characteristics.The results indicate that the maximum entropy method augmented with a genetic algorithm features rapid segmentation,high efficiency,and robust segmentation performance.Furthermore,for remote sensing data gathered from UAVs and small-scale waters near the coast,an improved YOLOv8 s recognition algorithm is introduced.This algorithm utilizes a spatial attention mechanism and multi-scale convolution fusion method to precisely identify the types of oil pollution sources.The results demonstrate that the enhanced YOLOv8 s algorithm increases the m AP@0.5 by 6.7% compared to the original YOLOv8 s algorithm,and the detection speed reaches 92f/s,indicating strong practical applicability.In summary,this study introduces a method for identifying oil spills in small-scale waters using UAV-based remote sensing monitoring.It also details the techniques for measuring the area of oil films after spills and outlines methods for tracing the sources of oil spill pollution.The conclusions of this study can provide references for the clean-up treatment of oil spills from ships and the claims and liabilities of oil spills.