Research On Detection Of Aircraft In High-resolution Optical Remote Sensing Images

Automatic remote sensing image target recognition is an important research topic in the field of image processing and pattern recognition,which has the very high academic and practical application value.Optical image target detection and recognition is to transform optical remote sensing image from data to information and then complete the key stage of the decision,interpretation and target monitoring.It has very important significance in enhancing the automation of image interpretation,the efficiency of the utilization of remote sensing image data,so it's a hot issue in the field of intelligent processing of remote sensing information.Among them,the aircraft,as an important military targets in the military field,its detection and recognition on remote sensing images has important military significance,which will help to analysis war situation,assist to combat decision and improve operational efficiency.This paper aims at the particularity of automatic remote sensing image target recognition,and on the basis of reference to human visual system,utilizes the method of superpixel division to detect the aircraft from the remote sensing images.The main innovation work of this article can be divided into the following two points.Firstly,Aiming at the aircraft target detection problem in high resolution optical remote sensing image recognition,this article studies the key technologies of the superpixel division,global significance testing,regional feature extraction and so on.And we have proposed a fast target detection method of optical remote sensing image.The method of target detection based on the target-superpixel of significant degree model,through using machine vision field of visual significance model,the concept of target feature analysis,so as to realize the goal of high efficiency and high precision detection.The algorithm improves detection efficiency which can be reduced by noise interference,part of the loss,the shadow shade,local deformation when applied to practical application.Secondly,This article has put forward a target detection algorithm which builds on top of the superpixel division.For each high resolution optical remote sensing image,our algorithm first calculates the visual significant figure,and selects screening significant point area as candidate.And then performs super pixel division on gouge candidate area,computing the color,space distance and shape descriptor for each super pixel.Finally utilizes k-means clustering algorithm to remove redundant information,generating visual word and building word package model.By using samples to construct classification model through off-line training and classifying word package,which is used to realize the automatic detection of the target.The experimental results show that this method,compared with the traditional traversal search algorithm,sharply reduces the searching space and calculation complexity.