The Detection Of Airport Runways Based On Random Forest In PolSAR Image

Airport is an important civil and military infrastructure that carries missions such as takeoff and landing,maintenance of aircraft.Runway is the core facility of the airport.Automatic detection of airport runway is widely used in practical issues such as military strikes,disaster relief,and aircraft navigation.Synthetic aperture radar(SAR)is an active imaging radar system which has characteristics of full-day,all-weather,strong penetrability,and long-range imaging.Through transceiving different kinds of polarization microwave,polarimetric synthetic aperture radar(PolSAR)can obtain polarization scattering characteristics of ground objects.Compared with single-polarimetric SAR,full-polarimetric SAR has more transmitting and receiving modes,which can obtain more comprehensive polarization scattering information of ground objects.Therefore,Airport runway detection in PolSAR image has a very broad application prospect.Based on real PolSAR data collected by UASAR system,several studies of airport runway detection in PolSAR images have been done.The details are as follows:Classification is an important part of the interpretation of PolSAR images,which can be used to extract the region of interest(ROI).In order to reduce the number of classification categories while ensuring detection accuracy,an airport runway detection algorithm in PolSAR image based on Random Forest is proposed.Firstly,training samples are manually selected and a variety of features are extracted from the samples.Secondly,based on Fast Correlation-Based Filter algorithm(FCBF),some features which are weakly relevant and redundant are removed by analyzing the relevance and redundancy among the target features.Then,the ROI is extracted by Random Forest classifier which is designed to divide the image contents into two parts.Finally,the parallel-line and structural features of the runway are used to identify the real runway area.Parallel line feature is widely used in airport runway detection.Detection result obtained by parallel line feature is easily interfered by roads,rivers,bridges and other ground objects which also have parallel-line features.Therefore,a suspected runway area identification method,which doesn’t rely on parallel line detection,is proposed.Firstly,the sliding detection window is set in the PolSAR power image corresponding to each independent ROI area.Secondly,based on Histogram of Oriented Gradient(HOG),feature vector of detection window is calculate.Then,by using Support Vector Machine(SVM),‘Mask binary image’ of independent ROI is generated.Next,by conducting AND operation between ‘Mask binary image’ and original binary image of independent ROI area,the ratio of changed pixels is calculate.The real runway area is finally detected by the ratio value.Several experiments are carried out with real full-polarimetric SAR image data collected by UAVSAR system.The detected runway areas have a complete structure and clear outline,which demonstrate that the proposed method is effective.And a low false alarm rate is obtained by proposed suspected runway area identification method.