【摘要】：近海海洋環(huán)境、海洋災(zāi)害和海上突發(fā)事件等通常具有快速動(dòng)態(tài)變化的特性(小時(shí)級(jí)快速變化),而每天觀測(cè)一次的太陽同步軌道(極軌)海洋衛(wèi)星難以滿足日變化監(jiān)測(cè)的需求。靜止軌道衛(wèi)星可以利用同一遙感器對(duì)感興趣的同一區(qū)域進(jìn)行連續(xù)觀測(cè),是開展高頻對(duì)地觀測(cè)的最佳手段。然而任何衛(wèi)星平臺(tái)都受到顫振的干擾,由于靜止軌道衛(wèi)星成像系統(tǒng)積分時(shí)間一般比較長,這些顫振將嚴(yán)重影響遙感圖像質(zhì)量。目前衛(wèi)星平臺(tái)的顫振測(cè)量計(jì)算和抑制技術(shù)都發(fā)展到了較高的水平。衛(wèi)星平臺(tái)抑制技術(shù)和相關(guān)設(shè)備可以消除大部分的平臺(tái)顫振,但是對(duì)于低頻的衛(wèi)星平臺(tái)的姿態(tài)漂移卻束手無策。靜止軌道海洋成像輻射計(jì)軌道高度是35800km,其探測(cè)波段為可見光至近紅外8個(gè)波段,角分辨率為7?rad,地面分辨率(星下點(diǎn))250m,采用2048×2048元硅CMOS面陣探測(cè)器LUPA4000。由于靜止衛(wèi)星海洋成像輻射計(jì)軌道高度高,海洋反射能量弱,地面分辨率高,因此在成像過程中對(duì)衛(wèi)星平臺(tái)的姿態(tài)漂移特別敏感,為了提高系統(tǒng)的信噪比可以使用多次累加的方法。靜止衛(wèi)星海洋成像輻射計(jì)可見光模塊最多采用16次累加。而在累加的過程中,衛(wèi)星平臺(tái)存在低頻的姿態(tài)漂移。根據(jù)掌握的衛(wèi)星平臺(tái)數(shù)據(jù),目前衛(wèi)星平臺(tái)的穩(wěn)定度是5×10-4?,因此16次累加過程中圖像最多偏移了1.8個(gè)像素。如果不進(jìn)行處理而直接進(jìn)行累加,必然導(dǎo)致圖像模糊,影響成像質(zhì)量。本論文根據(jù)化整為零的思路首次提出了遙感圖像亞像素配準(zhǔn)算法,首先根據(jù)最優(yōu)準(zhǔn)則把圖像分割成不同的子圖像,其次使用基于Ostu的Canny算法對(duì)圖像進(jìn)行邊緣分割提取,然后在提取后的圖像使用SURF算法提取特征點(diǎn),最后在關(guān)鍵點(diǎn)周圍進(jìn)行開窗,窗口大小為200×200像素。在窗口中使用矩陣乘法相位相關(guān)法來計(jì)算圖像亞像素偏移量。綜合所有的子圖像的偏移量最終獲得整幅圖像的亞像素偏移量。論文對(duì)遙感圖像亞像素配準(zhǔn)算法進(jìn)行了詳細(xì)的仿真,在仿真的基礎(chǔ)上,為了提高以后算法硬件實(shí)現(xiàn)的處理速度,本論文提出了改進(jìn)的香農(nóng)熵低信息量特征點(diǎn)剔除算法和改進(jìn)的SURF算法:改進(jìn)的香農(nóng)熵低信息量特征點(diǎn)剔除算法減少了參與匹配的特征點(diǎn)數(shù),提高了算法的執(zhí)行速度。改進(jìn)的SURF算法將特征向量描述符的維數(shù)由原來的64維減少到36維,這明顯可以提高特征點(diǎn)匹配速度,并且改進(jìn)的SURF算法可以多路并行處理。這些改進(jìn)將極大的提升算法的硬件執(zhí)行速度。本文的研究內(nèi)容和創(chuàng)新點(diǎn)有以下4個(gè)方面:1)由于遙感圖像尺寸2048×2048,那么處理過程中需要的存儲(chǔ)資源和計(jì)算資源將十分巨大,一般的硬件如FPGA和DSP等將無法處理,因此本論文提出的遙感圖像亞像素配準(zhǔn)算法對(duì)圖像進(jìn)行分塊并行處理,提高算法執(zhí)行速度的同時(shí)也解決了FPGA,DSP等硬件無法處理超大尺寸遙感圖像的問題。并且由于遙感圖像亞像素配準(zhǔn)算法采用了基于矩陣乘法的相位相關(guān)法,該算法對(duì)噪聲有明顯的抑制作用,因此即使遙感圖像存在噪聲,本論文算法仍然可以獲得較高精度的亞像素偏移量估計(jì)值。2)改進(jìn)的香農(nóng)熵低信息量特征點(diǎn)剔除:該算法不僅大量地減少了參與匹配的特征點(diǎn)數(shù),而且還提高了匹配的正確率。因此可以明顯提高算法的運(yùn)算速度。SURF算法能夠獲得眾多的特征點(diǎn),然而在圖像匹配過程中發(fā)現(xiàn)存在許多未配對(duì)點(diǎn),針對(duì)這種情況,本論文提出改進(jìn)香農(nóng)熵的低信息量特征點(diǎn)剔除算法。改進(jìn)主要體現(xiàn)在:不僅考慮特征區(qū)域的離散像素值,而且也考慮特征中心點(diǎn)和周圍其他像素的相互關(guān)系。3)改進(jìn)的SURF算法實(shí)現(xiàn)了主方向計(jì)算和特征向量描述符生成的并行計(jì)算,同時(shí)也把特征向量描述符從原來的64維減少到36維,這些改進(jìn)不僅可以明顯地提高算法的執(zhí)行速度,而且也提高了匹配的正確率:SURF算法的改進(jìn)主要體現(xiàn)在把特征點(diǎn)的梯度使用徑向梯度進(jìn)行替代,這種替代可以實(shí)現(xiàn)特征描述子的旋轉(zhuǎn)不變性。在特征向量描述符的生成過程中由于傳統(tǒng)的SURF算法采用的是正方形區(qū)域,區(qū)域大小為20S×20S(S是特征點(diǎn)所在的尺度空間的尺度)。取而代之,改進(jìn)的SURF算法使用半徑為20S的圓形區(qū)域,省略了坐標(biāo)系旋轉(zhuǎn)步驟。把20S的圓形區(qū)域劃分成9個(gè)特征子區(qū)域,每個(gè)子區(qū)域使用4個(gè)特征進(jìn)行描述,這樣一共生成了36維特征向量描述符,大大的減少了描述符的維數(shù)。4)在對(duì)遙感圖像亞像素配準(zhǔn)算法進(jìn)行詳細(xì)研究和實(shí)驗(yàn)驗(yàn)證的基礎(chǔ)上,詳細(xì)的介紹改進(jìn)SURF算法特征點(diǎn)提取算法的硬件架構(gòu)、矩陣乘法相位相關(guān)法亞像素偏移量估計(jì)的硬件架構(gòu)和基于回歸學(xué)習(xí)圖像插值放大算法的硬件架構(gòu)。在對(duì)算法的原理和步驟深入研究的基礎(chǔ)上,把算法進(jìn)行適應(yīng)于硬件硬件實(shí)現(xiàn)的細(xì)分,給后期的遙感圖像亞像素配準(zhǔn)算法的硬件實(shí)現(xiàn)提供基礎(chǔ)和指導(dǎo)。
[Abstract]:The offshore marine environment, marine disasters, and marine emergencies usually have the characteristics of fast dynamic changes (the fast change of the hour level), and the daily observed solar synchronous orbit (polar orbit) ocean satellite is difficult to meet the needs of diurnal change monitoring. The stationary orbit satellite can use the same remote sensor to connect the same area of interest to the same area. Continuous observation is the best means to carry out high frequency to earth observation. However, any satellite platform is disturbed by flutter. Because the integration time of the satellite imaging system of still orbit is generally long, these flutter will seriously affect the quality of remote sensing image. At present, the technique of measuring and inhibiting the flutter measurement and suppression of the satellite platform has developed to a higher level. The satellite platform suppression technology and related equipment can eliminate most of the platform flutter, but the attitude drift of the low frequency satellite platform is helpless. The orbit height of the stationary orbit ocean imaging radiometer is 35800km, its detection band is 8 wavelengths of visible to near infrared, the angular resolution is 7? Rad, and the ground resolution (below the star point) 250m, The 2048 x 2048 element silicon CMOS array detector LUPA4000. is highly sensitive to the attitude drift of the satellite platform in the imaging process because of the high orbit height of the stationary satellite ocean imaging radiometer, the weak reflection energy of the ocean and the high ground resolution, so that the signal to noise ratio of the system can be increased by multiple accumulating methods. The visible light module of the radiometer is 16 accumulation. In addition, the satellite platform has low frequency attitude drift. According to the data of the satellite platform, the stability of the satellite platform is 5 * 10-4? So the image is most offset by 1.8 pixels during the 16 accumulating process. If it is not processed directly, it must be added directly. In this paper, the sub pixel registration algorithm of remote sensing image is first proposed in this paper. Firstly, the image is divided into different sub images according to the optimal criterion. Secondly, the image is extracted by using the Canny algorithm based on Ostu, and then the extracted image is calculated by SURF. The feature points are extracted by the method, and the window size is 200 x 200 pixels around the key point. The sub pixel offset is calculated by matrix multiplication phase correlation method in the window. The sub pixel offset of the whole image is finally obtained by using the offset of all the sub images. On the basis of simulation, on the basis of simulation, in order to improve the processing speed of the future algorithm hardware, this paper proposes an improved Shannon entropy low information feature point elimination algorithm and an improved SURF algorithm: improved Shannon entropy low information quantity feature point elimination algorithm reduces the number of parameters and matches, and improves the speed of execution of the algorithm. The improved SURF algorithm reduces the dimension of the feature vector descriptor from the original 64 dimension to 36 dimension, which can obviously improve the matching speed of the feature points, and the improved SURF algorithm can multichannel parallel processing. These improvements will greatly enhance the hardware execution speed of the algorithm. The research content and innovation point of this paper are the following 4 aspects: 1) The size of remote sensing image is 2048 x 2048, so the storage and computing resources will be very huge in the process of processing. The general hardware such as FPGA and DSP will not be processed. Therefore, the sub pixel registration algorithm of remote sensing image is processed in block and parallel processing, and the speed of implementation of the algorithm is also solved as well as FPGA, DSP and so on. The problem of ultra large size remote sensing image can not be handled by hardware. And because of the phase correlation method based on matrix multiplication, the subpixel registration algorithm of remote sensing image has obvious suppression effect on noise. Therefore, even if the remote sensing image has noise, this algorithm can still obtain high precision estimation of sub pixel offset. 2) the improved Shannon entropy low information feature point elimination: this algorithm not only greatly reduces the number of feature points of the participation matching, but also improves the correct rate of matching. Therefore, it can obviously improve the arithmetic speed.SURF algorithm to obtain many feature points, however, there are many unpaired points in the image matching process. In this case, this paper proposes an improved Shannon entropy feature point elimination algorithm with low information quantity. The improvement is mainly reflected not only in the discrete pixel values of the characteristic regions but also on the relationship between the feature center point and the other pixels in the surrounding area.3). The improved SURF algorithm realizes the parallelism of the main direction calculation and the generation of feature vector descriptors. At the same time, the eigenvector descriptors are reduced from the original 64 dimension to 36 dimension. These improvements not only improve the execution speed of the algorithm, but also improve the accuracy of the matching. The improvement of the SURF algorithm is mainly reflected in the use of the gradient of the feature points in the radial gradient in the row substitution, which can implement the feature descriptor. Rotation invariance. In the generation of feature vector descriptors, the traditional SURF algorithm uses a square area, the size of the region is 20S x 20S (S is the scale of the scale space in which the feature is located). Instead, the improved SURF algorithm uses a circular region with a radius of 20S, omitting the rotation step of the coordinate system. The circular region of the 20S is taken. It is divided into 9 characteristic subregions, each subregion is described with 4 features, thus a 36 dimension eigenvector descriptor is generated, and the dimension.4 of the descriptor is greatly reduced. Based on the detailed study and experimental verification of the sub pixel registration algorithm for remote sensing images, the improved feature point extraction algorithm of the SURF algorithm is introduced in detail. The hardware architecture, the hardware architecture of the matrix multiplication phase correlation method subpixel offset estimation and the hardware architecture based on the regression learning image interpolation amplification algorithm. Based on the in-depth study of the principle and steps of the algorithm, the algorithm is adapted to the subdivision of hardware and hardware implementation, and the sub pixel registration algorithm of the later remote sensing image is given. Hardware implementation provides the basis and guidance.
【學(xué)位授予單位】：中國科學(xué)院大學(xué)(中國科學(xué)院上海技術(shù)物理研究所)
【學(xué)位級(jí)別】：博士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP751

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 張成;王昕;史健衛(wèi);韓凡石;;新型亞像素圓檢測(cè)算法[J];計(jì)算機(jī)工程與應(yīng)用;2009年31期

2 馬志剛;;圖像的亞像素化技術(shù)研究[J];醫(yī)療裝備;2012年01期

3 趙小明;張建奇;邵曉鵬;周筱媛;;彩色矩陣顯示器亞像素采樣混色錯(cuò)誤的分析與評(píng)價(jià)[J];光電子.激光;2012年05期

4 王靜;;平面運(yùn)動(dòng)亞像素位移算法探索[J];石家莊鐵路職業(yè)技術(shù)學(xué)院學(xué)報(bào);2007年03期

5 韓松峰;陳大海;郭雷;刁海南;孟衛(wèi)華;;指紋周期/頻率的亞像素精確算法[J];計(jì)算機(jī)工程與應(yīng)用;2009年05期

6 丁興號(hào);鄧善熙;李了了;;小波系數(shù)過零點(diǎn)的屋脊邊緣亞像素檢測(cè)算法[J];哈爾濱工業(yè)大學(xué)學(xué)報(bào);2009年09期

7 吉祥;郭雷;陳大海;孟衛(wèi)華;;指紋亞像素頻率的濾波方法[J];計(jì)算機(jī)工程與應(yīng)用;2011年12期

8 張億路;杰史;雨輝;劉延飛;鄭勇;;亞像素細(xì)分在光學(xué)自準(zhǔn)直系統(tǒng)觀測(cè)中的應(yīng)用[J];大地測(cè)量與地球動(dòng)力學(xué);2013年S2期

9 張藎尹,李波,姚春蓮;基于最優(yōu)位置計(jì)算的快速亞像素運(yùn)動(dòng)估計(jì)[J];計(jì)算機(jī)學(xué)報(bào);2005年09期

10 陳良;高成敏;;基于像素分割的亞像素圓模板算法[J];計(jì)算機(jī)工程與應(yīng)用;2007年16期

相關(guān)會(huì)議論文 前10條

1 鄭偉;于桂杰;楊松軍;張朝暉;;圖像位移檢測(cè)中的一種亞像素快速匹配算法[A];第16屆中國過程控制學(xué)術(shù)年會(huì)暨第4屆全國故障診斷與安全性學(xué)術(shù)會(huì)議論文集[C];2005年

2 董詩浩;趙曉波;李阿蒙;劉曉利;田近東;彭翔;;透鏡畸變模型下的亞像素對(duì)應(yīng)點(diǎn)匹配方法[A];高精度幾何量光電測(cè)量與校準(zhǔn)技術(shù)研討會(huì)論文集[C];2008年

3 陳世哲;浦昭邦;劉國棟;胡濤;屈玉福;;基于點(diǎn)Hough變換的圓形圖像亞像素檢測(cè)算法[A];2004全國光學(xué)與光電子學(xué)學(xué)術(shù)研討會(huì)、2005全國光學(xué)與光電子學(xué)學(xué)術(shù)研討會(huì)、廣西光學(xué)學(xué)會(huì)成立20周年年會(huì)論文集[C];2005年

4 屈玉福;浦昭邦;王亞愛;;視覺檢測(cè)系統(tǒng)中亞像素邊緣檢測(cè)技術(shù)的對(duì)比研究[A];中國儀器儀表學(xué)會(huì)第五屆青年學(xué)術(shù)會(huì)議論文集[C];2003年

5 姜立軍;熊志勇;李哲林;;基于亞像素特征點(diǎn)提取的螺紋檢測(cè)粗糙聚類[A];中國計(jì)算機(jī)圖形學(xué)進(jìn)展2008--第七屆中國計(jì)算機(jī)圖形學(xué)大會(huì)論文集[C];2008年

6 劉春;張燕;;基于曲線擬合的玻璃瓶邊緣亞像素檢測(cè)算法研究[A];第十七屆全國測(cè)控計(jì)量儀器儀表學(xué)術(shù)年會(huì)（MCMI'2007）論文集（下冊(cè)）[C];2007年

7 李欣;王睿;張廣軍;;方形靶標(biāo)邊緣直線的自動(dòng)提取方法[A];第十二屆全國圖象圖形學(xué)學(xué)術(shù)會(huì)議論文集[C];2005年

8 張鳳春;董增壽;劉明君;;基于局部方差均衡的遙感圖像增強(qiáng)方法[A];第六屆全國信息獲取與處理學(xué)術(shù)會(huì)議論文集（2）[C];2008年

9 鄧冰;林宗堅(jiān);彭曉東;;遙感圖像信息度量的原理與方法[A];《測(cè)繪通報(bào)》測(cè)繪科學(xué)前沿技術(shù)論壇摘要集[C];2008年

10 江興方;江鴻;何賢強(qiáng);;遙感圖像兩種半自動(dòng)拼接方法的研究[A];全國農(nóng)業(yè)遙感技術(shù)研討會(huì)論文集[C];2009年

相關(guān)重要報(bào)紙文章 前5條

1 蔣建科邋孫宏金 陳樹琛;傳回清晰遙感圖像[N];人民日?qǐng)?bào);2008年

2 記者 鄭千里;北京地區(qū)有了航空遙感圖像[N];科技日?qǐng)?bào);2000年

3 本報(bào)通訊員;煤航遙感院獲美國快鳥遙感圖像西部代理權(quán)[N];中煤地質(zhì)報(bào);2005年

4 王石;印度通過“快鳥”影像發(fā)現(xiàn)古墓地[N];中國測(cè)繪報(bào);2010年

5 記者 馬彥平 張桂敏;澳大利亞鉀礦鉆探啟動(dòng)[N];農(nóng)資導(dǎo)報(bào);2011年

相關(guān)博士學(xué)位論文 前10條

1 陳啟強(qiáng);高清圖像處理與顯示關(guān)鍵技術(shù)研究[D];蘭州大學(xué);2015年

2 張超;基于局部特征的圖像配準(zhǔn)算法及應(yīng)用研究[D];北京理工大學(xué);2015年

3 楊娟;醫(yī)學(xué)圖像配準(zhǔn)和四維磁共振成像相關(guān)技術(shù)研究[D];山東大學(xué);2015年

4 虞剛;在線自適應(yīng)放療若干關(guān)鍵問題的研究[D];東南大學(xué);2015年

5 朱光;基于遙感圖像的交通道路目標(biāo)識(shí)別方法研究[D];吉林大學(xué);2015年

6 祁友杰;基于SoC技術(shù)的遙感圖像快速匹配方法研究[D];東南大學(xué);2016年

7 霍麗君;基于變分的遙感圖像恢復(fù)算法研究[D];中國科學(xué)院長春光學(xué)精密機(jī)械與物理研究所;2017年

8 陳彥彤;基于局部不變特征的遙感圖像星上目標(biāo)識(shí)別技術(shù)研究[D];中國科學(xué)院長春光學(xué)精密機(jī)械與物理研究所;2017年

9 章學(xué)靜;像素級(jí)圖像增強(qiáng)及配準(zhǔn)算法研究[D];北京理工大學(xué);2014年

10 江興方;遙感圖像去云方法的研究及其應(yīng)用[D];南京理工大學(xué);2007年

相關(guān)碩士學(xué)位論文 前10條

1 趙洋;灰色關(guān)聯(lián)亞像素檢測(cè)理論在油井管套損檢測(cè)中的應(yīng)用研究[D];西安石油大學(xué);2015年

2 楊兵兵;亞像素圖像邊緣檢測(cè)方法研究[D];大連理工大學(xué);2015年

3 張稱稱;SLM視覺激光條紋亞像素提取和配準(zhǔn)方法研究[D];北京工業(yè)大學(xué);2015年

4 周子淳;基于亞像素分析的軸類零件自動(dòng)檢測(cè)技術(shù)[D];吉林大學(xué);2016年

5 李多;基于亞像素檢測(cè)算法的視覺測(cè)量技術(shù)研究[D];電子科技大學(xué);2016年

6 吳軍妮;基于機(jī)器視覺的圖像亞像素測(cè)量技術(shù)的應(yīng)用研究[D];西安科技大學(xué);2015年

7 崔宗會(huì);基于相關(guān)系數(shù)加權(quán)擬合求解亞像素位移數(shù)字圖像相關(guān)算法研究[D];中國科學(xué)技術(shù)大學(xué);2016年

8 滕鵬超;基于亞像素采樣的圖像顯示硬件系統(tǒng)設(shè)計(jì)[D];西安電子科技大學(xué);2015年

9 艾斌;亞像素掃描調(diào)制系統(tǒng)參數(shù)優(yōu)化研究[D];南昌航空大學(xué);2017年

10 劉雨航;基于機(jī)器視覺的工件尺寸測(cè)量系統(tǒng)的設(shè)計(jì)與實(shí)現(xiàn)[D];中國科學(xué)技術(shù)大學(xué);2017年

，

本文編號(hào)：2153949