Research On Anomaly Detection Method In Marine Red Tide Environment Based On Long-term Multi-source Remote Sensing Data

Satellite remote sensing is one of the main means for mankind to understand the ocean,which is an important part of ocean disaster prevention and mitigation.Long-term satellite remote sensing data from multi-sources provide important mean for acquiring anomalous events in the marine environment and helping us to understand and grasp the environmental quality of sea areas and their changing trends in a timely manner.Based on long-term multi-source remote sensing data,this thesis has carried out research on the detection methods of marine red tide environment anomalies.In view of the large total number of marine remote sensing data but different sources,high interference and strong uncertainty,and less effective data under limited spatio-temporal scale,the research contents of this thesis are as follows:(1)Marine anomaly detection method considering spatio-temporal correlation.The proposed method is divided into two parts: point time series anomaly detection and anomaly range extraction after determining point anomalies.Firstly,deseasonality is used to eliminate the disturbance of seasonal factor fluctuations.Then,the spatio-temporal reconstruction of multisource remote sensing data is carried out.Also,the artificial neural network is used to make time series prediction,and the difference between the prediction result and the actual value is selected as the basis for time series anomaly detection.Finally,considering the spatial correlation of remote sensing data,the extreme value theory is used to dynamically extract the abnormal range.(2)Spatio-temporal modeling for long-term multi-source remote sensing data based anomaly detection.The long-term analysis involves a large amount of historical monitoring data that requires targeted organization modeling.Based on the practical application of marine remote sensing data,we propose a spatio-temporal tile model for the unified and efficient organization.The remote sensing data are unified and reorganized in the form of tiles in the spatial dimension,and the massive reorganized tile data are managed in the time dimension.Finally,the concept of spatio-temporal feature cube is proposed to improve the efficiency of spatio-temporal analysis.And the above model is combined with the marine environment anomaly detection method to propose a marine environment anomaly detection framework based on long-term multi-source remote sensing data.(3)The application of marine environment anomaly detection framework,using red tide detection as an example.The coastal area of Zhejiang Province was selected as the research area,using 20 years of long-term remote sensing data,based on chlorophyll a concentration,and selecting sea surface temperature and photosynthetically active radiation as the dependent variables to carry out the red tide.Finally verify the feasibility and effectiveness of the proposed model and method in this thesis.(4)The extension of key technologies in the Marine Satellite Data Online Analysis Platform(Sat CO2).Apply the model,method and conclusions proposed in this thesis to Sat CO2 for a complete marine environment anomaly detection process by providing convenient access and efficient utilization of marine satellite remote sensing data.At the same time,the spatial selfawareness mechanism of the tile is designed according to the characteristics of the tile block to assist in the realization of local enhancement visualization effect.Finally,the application ability and value of the model and method proposed in this thesis are expanded.