Automatic Retrieval Technology Of Optical Remote Sensing Images Of Nansha Coral Reefs

The South China Sea is the key area of China's strategic policy of "promote land and marine development in a coordinated way" and "building a maritime power".It is the key place to promote the "twenty-first Century Sea Silk Road".The Nansha Coral reef is the southernmost,most widely distributed and most coral reef group in the South China Sea.The study established a coral reef image library for Nansha Island Reef.Then the study designed the automatic retrival technology for Nansha Island reef images to realize the update of image library from unclear source images.In addition,it will eventually serve the Nansha Islands in real-time monitoring,safeguard Nansha's homeland security and swear the national sovereignty.In order to realize the normalizaition monitoring for Nansha Island Reef,this study established a multi-source,multi-target,multi-platform,multi-temporal,multi-scale Nansha Island reef image library,containing 4691 images of Nansha Island Reef from 1988 to 2018.Then,an image index set for Nansha Island Reef serving for image retrieval was established based on the reef image library,and local and global features of the image index set were extracted and stored.Finally,an image retrieval technique for reef images which considering local and global features was proposed.The technique achieved a better retrieval effect than the traditional methods,and could realize the updating and supplementation for image library with reef images from unknown sources.The specific research contents and conclusions are as follows:(1)This study established a multi-source,multi-target,multi-platform,multi-temporal,multi-scale Nansha Island reef image library,in order to grasp as much as possible the historical image information of Nansha Island Reef.This study collected 4691 images of 30 years' Nansha reefs from 1988 to 2018,including remote sensing images and Internet images,in which remote sensing images include medium and high resolution images.For the medium resolution images,GF4 satellite imges with 50 meters resolution have 300 scenes from 2016 to 2018,HJ satellite imges with 30 meters resolution have 863 scenes from 2009 to 2017,Landsat satellite images with 30 meters resolution have 1714 scenes from 1988 to 2018,and GF1 satellite images with 16 meters resolution have 627 scenes from 2014 to 2018.High reslotion images from Ikonos satellite with 1 meter resolution,Quickbird satellite with 0.6 meter resolution,Worldview satellite with 0.5 meter resolution and GeoEye satellite with 0.4 meter resolution have 167 scenes from 1999 to 2016.The Internet images containing multi views have 1020 scenes mostly from 2012 to 2018.(2)This study established an image index set for Nansha Island Reef and extracted and stored the local and global features of these images,serving for image retrieval experiment.In this study,the Nansha reef image index set,covering 18 key reefs and having 760 scenes from 2004 to 2018,was established through the Nansha reef image library(4691 scenes).At the same time,the local features were extracted by MSER(Maximally Stable External Region),described by SIFT(Scale Invariant Feature Transformation)and stored by high-dimensional feature index,while the global features—color histogram and GIST feature—were extracted and stored by matrix,for further image retrieval experiment.(3)This study proposed an image retrieval technique for reef images which considering local and global features,which achieved a better retrieval effect than the traditional methods,and could realize the updating and supplementation for image library with reef images from unknown sources.The proposed technology optimized the traditional technology which combined the local features and Ann algorithm,and then introduced the global features.With the help of GMCP(Generalized Minimum Clique Problem)graph theory model,the nearest neighbors with the most similar global features were selected,rather than the first nearest neighbor,to vote for the most matching image.The experimental results showed that this method has great results for remote sensing images with medium and high resolution,and aerial images close to orthophoto view,which can effectively identify the types,sources,and shooting time for island reefs.It was found that the global features of GIST were more suitable than the color features as the global feature input for the technology,and achieved the correct rate with 71.6%.At the same time,through comparative analysis,the island reef image retrieval technique considering local features and global features(GIST features)had achieved better results than the traditional technology combining local features and first nn,and could optimize the 23.9%of the results of the traditional method.The innovativeness of this research is to use local and global features to represent remote sensing images of islands and reefs,and establish a highly efficient storage feature index.Then,a reef image retrieval technique,which considering the local features and global features and adopts the graph theory model,is used to achieve a better retrieval effect than the traditional methods.However,there are still some deficiencies.In the future work,our research will enrich and expand the Nansha Island Reef image index set,and then make the registration and retrieval of images from different perspectives be our key research object.