Research On Object Feature Analysis And Ship Identification Application In Optical Remote Sensing Image

In recent years,the analysis of remote sensing information for imaging reconnaissance or ground observation has become an important technical measure.Optical remote sensing image is widely used in object detection and recognition in the fields of military intelligence reconnaissance and intelligent traffic monitoring because it has the characteristics of obtaining detailed information of object features.For massive remote sensing images,it is still an important research topic how to obtain object regions of interest from large scene images,extract object features,and support object detection and recognition.In this thesis,taking ship as typical object,carry out research on feature analysis and recognition application in optical remote sensing images.The main research contents and innovations are as follows:(1)In this thesis,based on the theory of superpixel segmentation and region merging,study a segmentation method for land and sea regions of remote sensing images based on spatial constraints.Use superpixel algorithm to divide the molecular regions,and then cluster the similar regions by using merge model with spatial constraints,so that we can quickly divide the land and sea regions.Experimental result shows that the segmentation accuracy of proposed method is greater than 91% for multiple scenes,the integrity of boundary is preserved,which can improve the detection performance of ships.(2)In this thesis,study a feature extraction method of ship based on Grab Cut image segmentation,propose a progressive strategy that first extract the target and then analyze the features.Combined with the energy model of Grab Cut graph and the statistical moment method,extract the minimum enclost rectangular frame of ship,and use the geometric feature for object detection.The multi-feature cascade discrimination method is used to reduce the false alarm rate of the object effectively.(3)In this thesis,study a saliency feature analysis method which combining foreground features and background prior knowledge.Calculate boundary probability to measure distance,then generate background saliency map.Introduce the eigenweight function to constrain the mean clustering results,generate foreground saliency map.Finally,obtain the ship saliency map through weighted guided fusion.The experimental demonstration shows that the precision is ≥94% and the recall rate is ≥91%,which can effectively overcome the difficulty of obtaining the ship area when the cloud and fog interfere.