Research And Application Of Remote Sensing Image Waterline Extraction Technology Based On Deep Learning

The monitoring of coastline change is significant for scientific management and protection of coastal zones.With the rapid development of high-resolution remote sensing images,the automatic extraction of water edges(especially sea-land lines)from high-resolution remote sensing images is an important measure to achieve coastline change monitoring.However,on the one hand,the features of ground thins in high-resolution remote sensing images are rather complicated,and the boundary of the closed inner water area on the land is easily misidentified as the sea-land line;on the other hand,due to the variety of water edge lines(including artificial,sandy,silt,etc.)which are easily blocked by nearby ground features,and the extracted waterline is prone to discontinuous breakpoints.Therefore,it is of great significance to study highly performed waterline extraction methods based on high-resolution remote sensing images.In this paper,the following three various works are done to solve the problem of extracting water borders from remote sensing images,which can be summarized as follows:(1)To solve the problem of misidentification in high-resolution remote sensing images,a residual sea-land semantic segmentation network(Co-ResNet)based on collaborative feature extraction is constructed.The network consists of an encoder,a cooperative feature extraction module(CFEM)and a decoder.The encoder uses ResNet-50 as the backbone network to extract the shallow and deep features of the input remote sensing image and generate feature maps with semantic information.For the extracted feature maps,the CFEM)based on grouped convolution is then used to extract local and global features,which can improve the accuracy of feature extraction and reducing the amount of parameters.Finally,4 artificial coastline images of Tianjin coastal area and a bedrock coastline image are adopted to verify the effectiveness of the proposed method.And different types of coastlines in Huangdao District and Laoshan District in Qingdao City are applied to conduct generalization performance experiments.The results show that the CFEM module can effectively solve the problem of mis-segmentation of traditional network segmentation results,which can lead to higher segmentation accuracy and better generalization performance.(2)To solve the problem of discontinuous breakpoints in the extraction of high-resolution remote sensing images,based on the Co-ResNet model,a coastline semantic segmentation network based on the spatial attention module(Am-ResNet)is designed.The AM module can eliminate noise and avoid excessive segmentation by enhancing useful low-level feature information.The same remote sensing image as Co-ResNet are applied to verify the effectiveness of the proposed method.The results show that the AM module can effectively capture the important spatial relationship between pixels,which can make edges of the sea and land segmentation results more precise and delicate.And therefore it effectively improves the overall water edge segmentation of the method.(3)By using historical remote sensing images of Tianjin coastal area as the data source,Am-ResNet is used to extract the coastline of the region in the past ten years and analyze its change trend.The experimental results show that the total length of the coastline in the region is Increased by 172.52 km in total in the past 14 years and the average growth rate is about 12.32km/a.In addition,the average end-point rate and linear regression rate of the coastline change in the region are 25.96m/yr and 27.49m/yr,respectively,which maintains a high rate of change.And,the coastline is dominated by seaward expansion and the overall trend is toward the sea.