False Alarm Source Detection In Infrared Imagery From Earth Observation

With the development of modern space exploration technology,various satellite images can be acquired and analyzed.The infrared imagery from earth observation has been widely applied in different fields because of its imaging mechanism and functions.Moreover,target detection in infrared images is crucial to modern warfare such as space guidance,security and early warning.Infrared target detection is a challenge problem for the far distance imaging,complex background and limited features.On the other hand,infrared imagery is formed from invisible infrared radiation via receiving the radiation by infrared sensors and then converting this thermal signal into electrical signal.Nevertheless,there are amount of areas such as snow mountain,frozen river,cirrus,etc.that will also be captured with our target at the same time because they have high radiation as well.Therefore,these regions will affect target detection and produce a lot of false alarms.To increase the performance of earth observation systems and the accuracy of target detection,this dissertation will focus and analyze the scenes which may produce the false alarms,i.e.,the false alarm sources to decrease the false alarm rate through detecting these false alarm sources.Combined with different kinds of detection methods from infrared remote sensing images,this dissertation mainly researches on the feature analysis and extraction of false alarm sources and targets,the detection of typical false alarm sources and the corresponding application in infrared imaging system from earth observation.The main contents are listed as follows:(1)Based on the infrared imaging mechanism,we analyze the scenes that may generate high radiation and difficulties of their detection.Moreover,we research and analyze the universal and general features to some extent by analyzing false alarm sources in the images such as texture,grey,shape,visual salience,fractal feature and so on.The features of the targets in different false alarm backgrounds are studied as well.By researching on the features of false alarm sources and targets and analyzing their imaging mechanism,the theoretical base has been built and serves our research on detection algorithms in this dissertation.(2)An improved local binary pattern algorithm combined with morphological theory for river segmentation is proposed based on the differences of grey and texture features between the river and other regions.Traditional local binary pattern feature whose neighborhoods are 8 pixels and the radius is 2 which will ignore the information of the nearest neighbor pixels.In this dissertation,we change the radius and compute each value again via linear interpolation to involve as much local information as possible which will solve the problem of information lost.Through the feature extraction and utilizing the appropriate threshold and morphological methods,the river can be segmented rapidly and easily.(3)A river detection method using Frangi filter and shearlet feature is proposed based on the shape and directional feature of the river.Frangi filter can enhance the tubular or strip shape in the image.Therefore,rivers can be enhanced by Frangi filter because they are similar to the strip.In addition to their shape feature,different from other scenes like buildings,lake or etc.,rivers will flow in some certain directions.Thus,shearlet feature which can study features in various directions is utilized in this dissertation after Frangi filter.The rivers' feature will be more apparent in some directions than other scenes so that we can build the apparent feature maps again to enhance the rivers and suppress other regions.Then the river can be extracted successfully and efficiently by utilizing active contour model on the feature map.Moreover,to increase the precision of the detection,an approach which selects the regions based on their ratios between long axis and short axis and the area is proposed to rule out the detected non-river regions and to improve the performance of the algorithm.(4)Based on the visual salience feature of infrared imagery from earth observation,a cirrus detection method based on multi salience feature fusion is proposed.Cirrus is usually apparent in the infrared image and the gray value is always different from others.Moreover,single feature cannot express the target well but different scales or kinds of features could represent different characters.Thus,this dissertation extracts the salience features and proposes feature fusion strategy based on principle component analysis to enhance the cirrus and detect it accurately via active contour model.(5)An automatic single-image based cloud detection method based on random fractal model and sparse representation is proposed.As we all know,remote sensing images,especially the infrared satellite images,are difficult to collect as well as their corresponding labels.However,the detection approaches based on machine learning demand a lot of data and labels which are very high cost.Therefore,to solve the problem,this dissertation proposes a cloud detection algorithm that firstly builds the cloud model by random fractal method.Then the suitable cloud atoms are selected through calculating the histogram similarities between random fractal model we produced previously and the image patches.Finally the atoms are trained to generate the cloud dictionary and the cloud can be extracted via sparse representation.This method realizes the cirrus detection from single image without any prior information so that the demand of data and labels can be avoided.Thus,the proposed method can extract the cirrus accurately while reduce the cost of the algorithm.(6)Based on the previous research on false alarm sources in infrared remote sensing images from earth observation,we applied these false alarm source detection methods to target detection in infrared images from earth observation.Meanwhile,on the basis of considering the detected false alarm sources as prior information,the novel target detection theory is built and we analyze the rationality and efficiency of its future application.Furthermore,we research on other applications of false alarm source detection methods in infrared system from earth observation as well.Different experiments and results show that the proposed approaches can detect the typical false alarm sources successfully and receive the competitive performances.Moreover,the target detection theory based on the pre-extracting false alarm sources is built.The discussions on reducing the false alarm rate to some extent and improving the accuracy are also demonstrated.Additionally,the theoretical system of false alarm sources detection we built in this dissertation is a good starting point for the following research and applications in infrared system from earth observation.