Research On Target Feature Extraction Method Based On GF-3 SAR Imagr

Synthetic Aperture Radar(SAR),because of its excellent imaging ability,is not affected by weather such as clouds and rains,can still provide high-quality,high-timeliness images for earth observation in harsh environments,which makes up for the defect that optical sensors can not obtain data due to weather.It has been applied in more and more fields and is an important means to obtain ground information.GF-3 is a civil radar satellite independently developed in China.Its advantages of all-weather,multi-mode and multi-polarization provide a new way of thinking for Earth Observation Technology in China.With the development of radar satellite,the resolution of SAR image is higher and higher,and data acquisition is more and more convenient.It can provide real-time and scientific data support for urban planning and monitoring of urban resources.However,due to the complexity and diversity of urban features,there are few studies on urban target extraction based on GF-3.At the same time,for SAR images with high resolution and high data volume,the processing method is still an urgent problem to be solved in the application of SAR images.Therefore,based on high-resolution GF-3 SAR image,this paper takes the water and buildings in the city as the target objects to extract the features of the target objects.GF-3 SAR image processing is the basis of image application.Because of its unique imaging mechanism,SAR image processing methods are also different from traditional optical images.Therefore,this paper firstly analyzed the processing flow of GF-3 SAR image,and analyzed the SAR image from the imaging principle,geometric features,image factors and other aspects.At the same time,it analyzed the characteristics of the target buildings and water bodies in SAR image.For speckle noise in radar imaging process,different filtering methods are selected to suppress it.The results are compared and analyzed from the perspective of quantitative evaluation.The analysis results show that the improved Lee filtering method is superior to other filtering methods in image detail preservation.Then,the features of two kinds of target objects are extracted by using the gray level co-occurrence matrix based on statistics,and the extraction parameters are optimized by using the Papanicolaou distance.The extracted feature data and the gray level feature map of the image are constructed into a high-dimensional data set of target objects.For the extracted feature data set,this paper introduces manifold learning algorithm,which reduces the dimension of the data while maintaining the original features of the object,reduces the amount of data in the application of SAR image,and extracts the new features of the target object.In this paper,PCA algorithm,LE algorithm and LTSA algorithm are selected to reduce the dimension of high-dimensional feature datasets.According to the result of data dimension reduction,the same classification method,K-means clustering algorithm is used to extract target objects.By comparing the accuracy of three kinds of algorithms,we can find that LTSA algorithm has low false alarm rate and high precision detection rate for building extraction,and LTSA algorithm has high precision detection rate for building areas with dispersed distribution.It is obviously superior to other algorithms.For water body extraction,all three algorithms can extract water body range accurately,but LE algorithm has the lowest false alarm rate and the most accurate water body extraction range.Based on the quantitative evaluation of the classification results,this paper summarizes the optimal extraction method of building and water body based on high-score GF-3 SAR image.