本文關鍵詞：LAMOST拼接異常光譜的分析與處理　出處：《山東大學》2017年碩士論文　論文類型：學位論文

  更多相關文章： 光譜拼接異常 分段擬合 流量差 異常分級 高性能計算

【摘要】：2008年10月,我國的大天區(qū)面積多目標光纖光譜天文望遠鏡(LAMOST)落成典禮在河北興隆觀測基地舉行,該望遠鏡于2011年10月開始進行先導觀測,目的是檢查設備的性能和評估巡天計劃的可行性。2012年9月LAMOST正式進行觀測,到次年6月,共計拍攝約4,149,500條目標光譜,其中包括先導巡天的1,338,750條光譜。2013年9月,LAMOST正式對外發(fā)布DR1數(shù)據(jù),其中共計2,204,860 條光譜。LAMOST觀測產(chǎn)生了大量光譜,我們注意到對外釋放的達到質(zhì)量要求的光譜數(shù)據(jù),只占所有觀測數(shù)據(jù)的70%。即使在這些已發(fā)布的數(shù)據(jù)中,也存在著質(zhì)量較差的光譜。其中,拼接異常光譜是質(zhì)量較差光譜的一種,本文主要研究拼接異常光譜的分析與處理,目的是在海量光譜數(shù)據(jù)中挖掘拼接異常光譜。本文的研究內(nèi)容包括:(1)課題相關技術介紹。分為三個部分,第一部分介紹python以及基于python的天文數(shù)據(jù)處理技術,主要介紹python處理光譜數(shù)據(jù)時依賴的包以及其對光譜處理的便捷性。第二部分介紹高性能計算平臺的架構和基本原理。第三部分介紹基于python的高性能計算平臺的并行計算技術。(2)拼接異常光譜的識別和異常分級方法。這是本文的核心部分,主要介紹方法的原理,閾值的確定和自定義異常識別函數(shù)。本文根據(jù)大量的實驗數(shù)據(jù),提取光譜的數(shù)學統(tǒng)計特性,針對這些特性,確定了異常光譜的各項閾值,通過實驗分析,提出了一個光譜異常分級的評價函數(shù),通過該函數(shù)的評價得分將異常光譜分為三個等級,可以為光譜研究提供不同質(zhì)量等級的光譜。(3)介紹高性能平臺下拼接異常光譜識別的并行處理,并與單機環(huán)境下的運行效率進行比較。本部分研究串行化的異常光譜識別方法的并行化實現(xiàn),以便在高性能計算平臺運行,并介紹了我校高性能計算平臺的使用方法。拼接異常是光譜在紅藍兩端拼接區(qū)域表現(xiàn)出的光譜連續(xù)性差的一種現(xiàn)象。在LAMOST的光譜處理中,儀器的穩(wěn)定性、觀測條件以及獲得的響應函數(shù)等問題都是造成拼接異常的原因。光譜拼接是否正常對于光譜發(fā)布等后續(xù)工作的質(zhì)量有重要影響。本文提出一種拼接異常光譜的自動檢測方法,有效地提高了工作效率。本文的研究可以為LAMOST數(shù)據(jù)提供一個自動的標記,來評價拼接質(zhì)量,也可以為用戶提供一個使用數(shù)據(jù)時的選擇。本文中的方法首先將待測光譜進行流量歸一化、去除鈉線等預處理,并將其分為紅藍兩端;然后對紅藍兩端分別進行擬合;最后對兩條擬合曲線,選取一系列等波長間隔的點,計算在這些點處的流量差值,得到所有流量差值的均值,標準差,并且計算兩條曲線積分面積的差值;基于上述統(tǒng)計量,我們提出了一個判斷光譜是否異常及其異常程度的評價函數(shù)。大量的實驗證明,該方法可以將拼接異常光譜準確識別出來。同時本文研究了在高性能計算平臺上拼接異常光譜的識別與異常分級方法,效率相比單機有了很大的提升。
[Abstract]:On October 2008, China's large sky area multi-target optical fiber spectral astronomical telescope (LAMOST) was inaugurated at Xinglong observation Base in Hebei Province. The telescope began conducting pilot observations on October 2011 to check the performance of the equipment and assess the feasibility of the survey plan. LAMOST officially observed it on September 2012 and until June the following year. A total of 4,149,500 target spectra were taken, including 1,338,750 spectra of pilot surveys. In September 2013, LAMOST officially released DR1 data. A total of 2,204,860 spectra. LAMOST observations produced a large number of spectra, and we note the emission of spectral data that meet quality requirements. Only 70 of all observed data. Even in these published data, there is a spectrum of poor quality. Among them, the spliced abnormal spectrum is one of the poor quality spectra. This paper mainly studies the analysis and processing of splicing abnormal spectrum. The purpose of this paper is to mine the splicing abnormal spectrum from the massive spectral data. The research content of this paper includes the introduction of the related technology of the subject: 1). It is divided into three parts. The first part introduces python and astronomical data processing technology based on python. This paper mainly introduces the package that python depends on when processing spectral data and its convenience to spectral processing. The second part introduces the architecture and basic principle of high performance computing platform. The third part introduces the structure and basic principle of high performance computing platform based on python. Parallel computing technology of high performance computing platform based on. 2) the method of identifying and classifying abnormal spectrum is the core of this paper. This paper mainly introduces the principle of the method, the determination of threshold and the self-defined anomaly recognition function. Based on a large number of experimental data, the mathematical and statistical characteristics of the spectrum are extracted. In view of these characteristics, the threshold values of the abnormal spectrum are determined. Through experimental analysis, an evaluation function of spectral anomaly classification is proposed, and the abnormal spectrum is divided into three grades by the evaluation score of the function. It can provide spectrum with different quality level for spectral research.) it can introduce parallel processing of splicing abnormal spectrum recognition under high performance platform. In this part, the parallel realization of serialized anomaly spectrum recognition method is studied in order to run on the high performance computing platform. The paper also introduces the application method of our high performance computing platform. The splicing anomaly is a phenomenon of spectral continuity difference in the region of red and blue splicing. It is used in the spectral processing of LAMOST. The stability of the instrument. The observation conditions and the response function obtained are all the causes of the abnormal splicing. Whether the spectrum splicing is normal or not has an important effect on the quality of the subsequent work, such as the spectral distribution. This paper proposes a kind of self-stitching abnormal spectrum. Dynamic detection method. The research in this paper can provide an automatic mark for LAMOST data to evaluate the quality of stitching. The method in this paper firstly normalizes the flow rate of the spectrum to be measured, removes the sodium line, and divides it into red and blue ends. Then the red and blue ends were fitted. Finally, for two fitting curves, a series of points with equal wavelength spacing are selected, and the flow difference at these points is calculated. The mean value and standard deviation of all the flow differences are obtained, and the difference of integral area between the two curves is calculated. Based on the above statistics, we propose an evaluation function to judge whether the spectrum is abnormal or not and the degree of anomaly. This method can accurately identify the spliced abnormal spectrum. At the same time, this paper studies the method of identifying and classifying the abnormal spectrum on the high performance computing platform. The efficiency of this method is greatly improved compared with that of the single machine.
【學位授予單位】：山東大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：P111;O433.4

【相似文獻】

相關期刊論文 前9條

1 熊宇虹,溫志渝,陳剛,黃儉,徐溢;基于小波包分析的光譜識別方法研究[J];光譜學與光譜分析;2005年08期

2 曹暉;周延;;多種群精英共享遺傳算法在異常光譜識別中的應用[J];光譜學與光譜分析;2011年07期

3 施德恒;光譜識別型激光警戒技術和激光警戒裝置的研究與發(fā)展現(xiàn)狀[J];應用光學;1994年06期

4 熊宇虹;溫志渝;王命延;徐少平;王煒立;肖建;;基于神經(jīng)網(wǎng)絡的光譜識別系統(tǒng)的設計與分析[J];光譜學與光譜分析;2007年01期

5 施德恒,郭峰;光譜識別型激光警戒系統(tǒng)述評[J];激光與紅外;1999年01期

6 白鵬;王建華;王宏柯;張發(fā)啟;劉君華;;基于SVM回歸模型的混合氣體組分種類光譜識別方法[J];光子學報;2008年04期

7 施德恒;光譜識別型激光警戒裝置的研究與發(fā)展[J];紅外與激光技術;1995年02期

8 潘沈元;蠶繭近紅外反射(NIR)光譜的模式識別II.光譜識別中特征值選取方法的探討[J];生物物理學報;1998年02期

9 ;激光軍事應用[J];中國光學與應用光學文摘;1999年05期

相關會議論文 前1條

1 鄧巍;趙春江;何雄奎;張瑞瑞;許剛;陳天恩;;玉米苗期雜草光譜識別三種分類模型的比較[A];第十六屆全國分子光譜學學術會議論文集[C];2010年

相關碩士學位論文 前4條

1 馬帥;基于深層信念網(wǎng)絡的太赫茲光譜識別研究[D];昆明理工大學;2015年

2 孟凡龍;LAMOST拼接異常光譜的分析與處理[D];山東大學;2017年

3 武樹斌;激光偵察中的光譜識別與信號檢測技術[D];西安電子科技大學;2000年

4 王莉峰;幾種經(jīng)濟作物的光譜識別研究及應用[D];新疆大學;2014年

，

本文編號：1396623