Extraction Algorithms Of Aquaculture Ponds In Coastal Zones Based On Moderate Resolution Remote Sensing Imagery

As an important form among aquaculture catalogues,coastal pond aquaculture plays an important role in global food,nutrition and economic benefits.However,it also brings great challenges to the sustainability of coastal ecological environment.Therefore,there is an urgent need for effective dynamic monitoring of pond aquaculture in order to achieve sustainable marine management of fisheries.Remote sensing technology can provide an important tool for the monitoring of aquaculture ponds.Especially,moderate resolution remote sensing imagery has the ability to provide large coverage.However,the edges of the aquaculture ponds are weak due to the imagery resolution,and some interferences nearby such as other water bodies,which bring great difficulties to the extraction of aquaculture ponds from moderate resolution images.In optical remote sensing,the classification method usually adopted in aquaculture ponds extraction relies on a large number of training samples and time-consuming training process,and has poor portability.Therefore,it is necessary to study the extraction algorithm with less manual participation and good portability.Currently,the existing Synthetic Aperture Radar(SAR)aquaculture pond extraction methods depend on many auxiliary data,so it is necessary to study the extraction algorithm without requiring auxiliary data,and less mannual operations.Meanwhile,it is also necessary to analyze the performance of these two extraction algorithms in comparison with existing algorithms.Therefore,this paper starts from the moderate resolution multi-spectral image Landsat-8 and SAR image Sentienl-1 respectively to study the extraction algorithm of coastal zone aquaculture ponds.The main research contents of this paper are as follows:(1)An extraction algorithm of aquaculture ponds from Landsat-8 multi-spectral image by combining index method and simple linear iterative clustering(SLIC)is proposed.In order to avoid the problems of pepper and salt noise of pixels and different land covers with the same spectrum,the algorithm uses superpixels as the basic unit,and provides a strategy of edge breakpoints connection to form closed contours for different objects,and exploits it to guide the generation of superpixels.A model based on modified normalized difference water index(MNDWI)linear mixing and water fraction is constructed to automatically select sea and land seed points in order to solve the problem of artificial selection on water index threshold.Finally,a region merging criterion integrating edge features and spectral features is proposed to remove interference targets in the sea.The experiments utilize the images from four regions of Bohai,Liaoning,Shandong and Guangdong with differences in time and space,and two performance indexes including Kappa index and intersection over union(IOU)have been used to quantatively assess the accuracy,respectively.The Kappa index of our algorithm is close to that of the comparison algorithm,and IOU is about 5%greater than that of the comparison algorithm.For landsat-8 images,the experiment shows that our algorithm does not need manual participation.Meanwhile,the portability of our algorithm is verified through the extraction experiment of aquaculture ponds in different regions,and the extraction performance of aquaculture ponds is better than the comparison algorithm.(2)A aquaculture ponds extraction algorithm based on radon transform and multi-feature fusion is proposed for Sentinel-1 SAR image.In order to improve the extraction efficiency of aquaculture ponds in coastal zones,a coastline detection algorithm based on Sentinel-1 image is proposed to automatically restrict the scope of coastal zones through the coastline.Due to the difficulties in extracting intensive aquaculture ponds in SAR images,Radon transform has been introduced into the extraction of aquaculture ponds for the first time,and the characteristics of the aquaculture ponds have been constructed from four aspects including the standard deviation of the polar angle,the arithmetic sequence matching degree of the polar diameter,the entropy of the polar angle histogram and the central aggregation degree of the polar angle histogram.Finally,a construction method based on the straight line number rule of Radon transform is proposed to extract the aquaculture ponds from sentinel-1 image.The experiments utilize the images from two regions of Bohai and Guangdong with differences in space,and two performance indexes including Kappa index and intersection over union(IOU)have been exploited,respectively.In comparion to the existing algorithm,the proposed algorithm could achieve more than 3%in Kappa index and IOU,which verifies the performance advantage of our algorithm,meanwhile our algorithm does not need auxiliary data,and has less mannual works.