False Alarm Source Detection Based On Few-shot Deep Learning

Space infrared satellite plays an important role in remote sensing observation and missile early warning,and is an important part of national defense construction.Space infrared satellite uses its on-board infrared detector to detect the high-intensity infrared radiation produced by missile or aircraft in the process of flight,so as to realize the real-time early warning and monitoring of the whole world.However,there are a large number of targets with high radiation characteristics in the imaging band of the infrared detector of the space infrared satellite,which cause interference or even false alarm to the normal operation of the missile early warning system.These targets are called false alarm sources.Typical false alarm sources include snow mountains,frozen lakes and high-altitude cirrus clouds.In order to avoid the interference caused by the false alarm source and improve the detection efficiency of the real target,it is necessary to detect and eliminate the false alarm source in advance.As the most effective method in the field of image processing in recent years,deep learning can extract more comprehensive features than the traditional artificial design,which is conducive to improving the overall accuracy of the false alarm source detection algorithm.Because of the particularity of the data source,the data sample of the ground infrared detection has the weak sample condition,which shows that the image quality is poor and the number of samples is too small compared with the traditional deep learning application scene.Focusing on the algorithm of false alarm source detection based on deep learning,the core issues include image preprocessing,false alarm source candidate area detection and false alarm source classification and recognition.Aiming at these problems,this paper has carried out theoretical analysis,method research and simulation verification.The main contents of this paper are as follows:(1)Firstly,this paper studies the basic theoretical knowledge of deep convolution network,including the machine learning algorithm foundation of deep learning,the basic structure of convolution neural network including convolution layer,pooling layer and activation function,and the training optimization methods of deep network;(2)In this paper,the image preprocessing method is studied and proposed for the imaging of infrared earth detector of space infrared satellite,including the image denoising method based on bilateral filtering and adaptive median filtering and the image contrast enhancement method based on homomorphic filtering;(3)In view of the high radiation imaging characteristics of typical false alarm sources,several commonly used adaptive threshold segmentation algorithms are studied,and on this basis,the detection method of candidate areas of false alarm sources based on threshold segmentation and morphological reconstruction is proposed;(4)Aiming at the particularity of the earth observation data of the space infrared satellite,this paper studies the meta learning method based on the measurement to realize the effective feature extraction of the false alarm source when the number of samples is very small,and proposes the feature extraction network based on the separable convolution;The method proposed in this paper has been verified by simulation and measured data,and the results show that it can effectively detect the false alarm source.