本文選題：SAR對抗 + 卷積神經(jīng)網(wǎng)絡��； 參考：《電子科技大學》2017年碩士論文

【摘要】：SAR全天候、全天時等特點使其在軍事領域具有光學成像系統(tǒng)無法替代的優(yōu)勢。而隨著SAR軍事作用的提升,針對SAR的干擾和抗干擾技術也成為了SAR研究的重要課題。SAR欺騙式干擾通過干擾機模擬虛假的回波,在SAR圖像上形成欺騙目標,嚴重影響SAR圖像解譯的可靠性。由于SAR一般為側(cè)視成像,SAR圖像中目標陰影特征明顯。因此,本文針對SAR欺騙式干擾,開展了基于陰影特征的SAR對抗方法研究,具體內(nèi)容及創(chuàng)新如下:1.闡述了SAR基本原理,SAR欺騙式干擾的實現(xiàn)方式以及SAR圖像的陰影特征。說明了SAR欺騙式干擾本質(zhì)是干擾回波和真實回波的疊加,難以實現(xiàn)對虛假目標陰影特征的模擬,因此可以在圖像域從陰影識別的角度進行SAR抗欺騙式干擾。簡要概述了識別所用到的神經(jīng)網(wǎng)絡的基本原理,對神經(jīng)網(wǎng)絡參數(shù)訓練常用的反向傳播算法進行了說明。2.對LeNet-5結(jié)構(gòu)的卷積神經(jīng)網(wǎng)絡模型進行改進以用于SAR目標分類。將LeNet-5結(jié)構(gòu)的卷積神經(jīng)網(wǎng)絡的卷積層和全連接分類層的激活函數(shù)分別改為ReLU和softmax函數(shù),池化層采用最大采樣,對網(wǎng)絡參數(shù)調(diào)節(jié)改善SAR目標分類效果。由于SAR公開數(shù)據(jù)庫中沒有欺騙式干擾下的數(shù)據(jù),本文為此研究了基于電磁仿真軟件的SAR成像仿真。通過計算雷達照射下的目標表面電磁流分布,回波仿真和成像處理得到復雜目標及其在欺騙式干擾下的成像結(jié)果,為基于識別的抗欺騙式干擾研究提供了樣本庫。3.提出了基于圖像域SAR目標分類以及抗欺騙式干擾方法。傳統(tǒng)SAR抗欺騙式干擾通過對發(fā)射信號進行復雜調(diào)制抑制干擾回波積累成像,無法對已經(jīng)受欺騙式干擾的圖像進行抗干擾。本文首先將SAR圖像通過卷積神經(jīng)網(wǎng)絡時完成對圖像的分類,此時圖像邊緣信息明顯,能夠較好的分辨出目標類型。但是真實目標陰影特征相對較弱,難以被學習和區(qū)分。接下來結(jié)合大津法和形態(tài)學運算對分類后的圖像進行多值化處理,多值化后的目標邊緣信息有所損失但陰影特征得到了增強,這時將圖像通過新的網(wǎng)絡實現(xiàn)真假目標的判定從而判斷SAR圖像是否已經(jīng)受到干擾并標記出欺騙目標。4.從彈射式和欺騙式兩個角度提出了兩種主動式陰影消除的SAR欺騙式干擾方法。彈射式方面,SAR成像中目標與其陰影方位向位置相同,在距離向陰影有所延后,陰影這一位置特點與彈射式干擾效果類似。因此本文通過對干擾機和彈射點位置的計算,將背景彈射至目標陰影處,實現(xiàn)了對目標陰影的消除。欺騙式干擾方面,獲取雷達參數(shù)后欺騙式干擾可以實現(xiàn)在特定位置產(chǎn)生干擾目標。本文通過真實目標計算出目標陰影位置,根據(jù)陰影位置信息進行欺騙式干擾調(diào)制,產(chǎn)生干擾回撥,從而在真實目標的陰影處疊加背景,完成對目標陰影的消除。兩種干擾方法都是通過消除真實目標的陰影特征令其與欺騙式干擾產(chǎn)生的虛假目標類似,使得干擾目標更具欺騙性,真假混淆達到干擾對方SAR圖像解譯的目的。
[Abstract]:SAR all weather, all day and so on make it have the advantage that optical imaging system can not be replaced in the military field. With the improvement of the military role of SAR, the interference and anti-jamming technology for SAR have also become an important subject of the research of SAR,.SAR deception jamming through the jammer to simulate false echoes, forming a deception target on the SAR image. The reliability of SAR image interpretation is seriously affected. Since SAR is generally side view imaging, the feature of target shadow in SAR image is obvious. Therefore, this paper studies the SAR countermeasures based on the shadow feature for the SAR deception jamming. The specific content and innovation are as follows: 1. the basic principle of SAR, the realization of the SAR deception jamming and the SAR image are described. It shows that the essence of SAR deception jamming is the superposition of the interference echo and the real echo. It is difficult to realize the simulation of the shadow feature of the false target. Therefore, the SAR anti deception jamming can be carried out from the angle of the shadow recognition in the image domain. The basic principle of the neural network used in recognition is briefly outlined, and the parameters of the neural network are trained. The common reverse propagation algorithm is practiced to show that.2.'s convolution neural network model of LeNet-5 structure is improved to be used for SAR target classification. The convolution layer of convolution neural network of LeNet-5 structure and the activation function of all connected classification layer are changed to ReLU and softmax functions respectively. The pool layer adopts the maximum sampling, and the network parameters are adjusted and modified. The effect of good SAR target classification. Because there is no deceptive interference in the SAR open database, this paper studies the SAR imaging simulation based on electromagnetic simulation software. By calculating the electromagnetic flow distribution of the target surface under the radar radiation, the echo simulation and imaging processing, the complex target and the imaging results under the deception jamming are obtained. The study of anti deception jamming based on recognition provides a sample library.3. proposed based on the image domain SAR target classification and the anti deception jamming method. The traditional SAR anti deception jamming can not interfere with the deception jamming image through the complex modulation suppression jamming echo accumulation imaging of the transmitted signal. The image is classified by the convolution neural network. At this time, the image is classified by the convolution neural network. At this time, the image edge information is obvious, and the target type can be distinguished well. But the real target shadow features are relatively weak and difficult to be learned and distinguished. Then, the images after the sub class are multivalued and multivalued after SAR and morphologic operations are combined. The target edge information is lost, but the shadow feature is enhanced, then the image is judged by the new network to determine the true and false targets, and the SAR image has been disturbed and the deception target.4. is marked out of the two active shadow elimination SAR deception jamming method from the projectile and deception. In the aspect of shooting, the target of SAR imaging is the same as the position of the shadow orientation, and the shadow is similar to the projectile jamming effect after the distance to the shadow. Therefore, this paper, by calculating the position of the jammer and ejection point, ejected the background to the shadow of the target, and realized the elimination of the shadow of the target. After obtaining radar parameters, deception jamming can achieve the interference target in a specific position. This paper calculates the target shadow position through the real target, makes the deception jamming modulation according to the shadow location information, produces the interference back and dial, and then superposes the back scene in the shadow of the real target, and completes the elimination of the shadow of the target. Two kinds of interference parties are completed. By eliminating the shadow characteristics of the real target, the method is similar to the false target produced by deception jamming, making the interference target more deceptive, and the true and false confusion can interfere with the target of the SAR image interpretation.
【學位授予單位】：電子科技大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TN974

【參考文獻】

相關期刊論文 前10條

1 史洪印;賈寶京;齊兆龍;;基于壓縮感知的非均勻脈沖SAR欺騙性干擾抑制方法[J];儀器儀表學報;2016年03期

2 丁軍;劉宏偉;陳渤;馮博;王英華;;相似性約束的深度置信網(wǎng)絡在SAR圖像目標識別的應用[J];電子與信息學報;2016年01期

3 高君宇;楊小汕;張?zhí)熘?徐常勝;;基于深度學習的魯棒性視覺跟蹤方法[J];計算機學報;2016年07期

4 和小冬;李昀豪;唐斌;;UWB-SAR相位域多普勒頻率調(diào)制假目標欺騙干擾[J];信號處理;2015年02期

5 李帥;許悅雷;馬時平;倪嘉成;史鶴歡;;一種深度神經(jīng)網(wǎng)絡SAR遮擋目標識別方法[J];西安電子科技大學學報;2015年03期

6 李云濤;陳永光;賈鑫;鄧云凱;尹燦斌;;多載頻MIMO-SAR對消處理抗欺騙干擾[J];電子學報;2012年09期

7 劉國滿;鄭坤;高梅國;;一種欺騙式SAR干擾信號發(fā)生器的設計與實現(xiàn)[J];北京理工大學學報;2012年02期

8 朱玉鵬;張月輝;王宏強;黎湘;;基于調(diào)頻率微調(diào)的ISAR抗干擾技術研究[J];信號處理;2010年03期

9 劉f ;王建國;;雙基SAR欺騙式干擾性能研究[J];電子與信息學報;2010年01期

10 孫光才;周峰;邢孟道;保錚;;虛假場景SAR欺騙式干擾技術及實時性分析[J];西安電子科技大學學報;2009年05期

，

本文編號：2036660