Research On The Segmentation And Change Of Coastal Zone Covers Based On Remote Sensing Images Of Donghai Island

The Donghai Island,the gateway to Zhanjiang Port,its coastal zone is the core area of social and economic development.Remote sensing target segmentation is a key technology for acquiring natural attributes in the daily work of the island.Research how to use high-resolution remote sensing images to extract and change analysis on the coastal feature information of the Donghai Island.The study shows that it has practical application value for the reasonable planning of land resources in the coastal zone of the Donghai Island and the implementation of coastal marine environmental protection measures.This research addresses the problems of high construction cost and long time span depending on manual field survey and installation of high-definition cameras,and the traditional remote sensing image feature segmentation method cannot meet the needs of high-level remote sensing image content interpretation.The Worldview-3 satellite image of Donghai Island is mainly used as the research object,and the deep learning model is combined with the high-resolution remote sensing image data for feature segmentation.Based on the segmentation results,Arc GIS software is used to analyze the feature change information of the Donghai Island Sea Coast in the past 5 years to mine the island's resource utilization situation,to further help the government to realize the transformation from informatization to intelligence,to improve the efficiency of supervision,and to accelerate the construction of smart and green islands.In the coastal zone feature segmentation task,first of all,this article uses the VGG-FCN segmentation deep learning model.This model uses FCN as the basic framework and replaces the convolutional layer in the FCN with the VGG16 network.The segmentation effect of remote sensing features in the coastal zone is relatively rough,especially the edge part.Secondly,in view of the shortcomings of FCN segmentation details insensitive and lack of spatial consistency,Unet is used as the basic framework,and the VGG-Unet model obtained by using the VGG16 network to improve the Unet coding part is used to segment the features.This model can achieve a more detailed segmentation of the edge area of the coastal features,but there is a phenomenon that local features are not segmented.Finally,a deep learning model based on Res-Unet segmentation is used,The model integrates the advantages of the Resne101 network's better feature extraction into the Unet network,and integrates techniques such as multi-scale cross training,multiple loss calculations,and SGD optimization.The Mio U segmentation accuracy rates of 92%,76% and 76% for coastal ports,aquaculture and coastlines have been achieved,and a good segmentation effect has been achieved.The experimental results show that the use of the Res-Unet model can accurately and efficiently segment the three types of features in the coastal zone,which has high practical value for the application of remote sensing image feature segmentation.In the coastal zone feature change analysis task,the Res-Unet model is used to re-integrate the results of the image segmentation of Dongjian Town,Donghai Island,to obtain the feature type map.Through a series of Arc GIS processing analysis of different time series remote sensing image types,such as the area statistics,the calculation of the degree of change,the acquisition of the transfer matrix and the statistics of the coastline length,it can be seen that in the past five years,the port and aquaculture area and transfer of Dongjian Town have changed drastically,the length of the coastline and the average rate of change vary greatly.The analysis of this information can provide planning guidance for the rational use of resources in the east island coastal zone.