Target Detection For Remote Sensing Imagery Coupling Linearly Constraint And Local Contrast

Target detection is one of the important content in remote sensing imagery information extraction,and has been used in many different applications,such as target classification,reconnaissance,surveillance,mineral mapping,and environmental monitoring.However,with targets generally existing as a small number of pixels and even as a subpixel in remote sensing images,signals of target are easy to be interfered by the complicated background and similar objects,leading to troubles for target detection.Thus,it is so important to build a more stabilized and effective target detection method.The existing research results show that principal questions in target detection methods based on remote sensing images contain:(1)the prior knowledge is fragmentary;(2)the existing of admixture of pixels would decrease the detection accuracy;(3)"spectral similar" objects would increase the error extraction rate in target detection.This paper has chosen typical target detection methods based on second order statistics and multivariate normal distribution to compare and analysis,proposed a target detection method with higher accuracy and lower false alarm rate,and used Landsat multispectral remote sensing imagery and AVIRIS hyperspectral remote sensing imagery to validate the method.The principal results and conclusions as follows:(1)The analysis and amelioration for methods based on second order statistics and multivariate normal distributionFor existing methods based on second order statistics,like CEM,LCMV,and TCIMF,and existing methods based on multivariate normal distribution,like RXD,DWEST,and LCM,this paper has done the extracting boats experiments on Landsat 8 OLI multispectral remote sensing imagery,and the extracting rhyolites experiments on AVIRIS hyperspectral remote sensing imagery,respectively.In these two kinds of methods,the appropriate methods LCMV and LCM were selected as subsidiary methods of the finial detecting method.This paper has developed the anomaly detecting method LCM,which only enhances the lightest pixel in the target window,and improved the accuracy of target detection.(2)The problem weakening for "spectral similar" objects on multispectral remote sensing imagesThe root cause of "spectral similar" objects lies in the finiteness of the spectral and spatial resolution ratio in remote sensing images.However,existing target detection methods usually extract signals that are similar with target spectral vectors and spatial characteristics,and this situation will increase the probability of extracting non-target objects,which have similar spectral vectors and spatial characteristics.To solve this problem,this paper not only thinks about spectral vectors of signals,but also thinks about spatial characteristics of targets.It means that combining spectral and spatial characteristics for enhancement targets to weaken the problem of "spectral similar" objects.In the evaluation of the proposed method,some "spectral similar"objects have been compared using the original images and finial images.The result shows that the proposed method could weaken the problem of "spectral similar"objects effectively on multispectral images.(3)Building and evaluating the target detection method coupling second order statistics and multivariate normal distributionThis paper builds a new target detection method LCLCM coupling the LCMV and the developed LCM.Two target detection experiments have been done on two images of different sensors,Landsat 8 OLI and AVIRIS,and the experimental results show that the proposed method would have preeminent suitability in different sensors.When the result of LCLCM compared with the Constraint Methods,like CEM and TCIMF,LCLCM has solved the interfered by similar spectral vectors to some extent,increasing the detecting accuracy and decreasing the false alarm rate.When the result of LCLCM compared with the Local Methods,like RXD,DWEST and LCM,LCLCM has weakened the interfered by non-targets that have the similar spatial characteristics with the target.The comparison experiments show that the proposed method could improve the detecting accuracy.For result of the Landsat 8 OLI image,the right detecting rate,95.68%,of LCLCM is far better than other methods,and the error detecting rate and the missing detecting rate are far lower than other methods.For result of the AVIRIS image,the right detecting rate,93.79%,of LCLCM is far better than other methods,and the error detecting rate and the missing detecting rate are far lower than other methods.