機器學(xué)習(xí)方法在遙感圖像處理中的應(yīng)用研究
[Abstract]:With the development of remote sensing technology, the number of remote sensing images that we can obtain not only increases explosively in quantity, but also increases the dimension of spectral features. However, manual calibration of remote sensing images is time-consuming and laborious, so it is necessary to process images automatically by means of machine learning. The research work in this paper mainly includes classification algorithm design and spectral feature learning. The domain adjusted (Domain adaptation) method mainly deals with the problem that the distribution of the original (source domain) data and the target domain (target domain) data is different and interrelated. At the same time, it is required that the classification tasks of the two domains are the same. In the field of remote sensing image classification, this phenomenon is very common. For example, remote sensing data collected in the same area may be different in time, weather conditions, atmospheric environment, etc., or data collected by the same remote sensing detector. They may also come from different locations (even if they have similar geomorphological features). To deal with this kind of data, the classification model based on the assumption of independent same distribution is directly used, and the result is generally not satisfactory. In this paper, we propose a domain adjustment algorithm based on input-output space consistency hypothesis. We select the half-mark points with high confidence weights from the target domain and delete the original domain training points which do not conform to the distribution of the target domain data. The iterative retraining classification model is proposed. Therefore, the algorithm is named as the input and output consistency domain adjustment algorithm (input-consistent-output domain adaptation, ICODA). ICODA algorithm) on two actual hyperspectral datasets (Botswana and KSC), which is used for verification and evaluation. The experimental results show that the final classification accuracy of ICODA algorithm is much higher than that of conventional classifier. In addition, on the basis of the spectral characteristics, the spectral derivative characteristics can be obtained by simple calculation, which can easily reflect the changing trend of the spectral curves. The existing research works adopt different methods to integrate this feature into the original spectral feature and apply it to the classification of remote sensing data. This paper studies data preprocessing without other means (e.g. In the case of dimensionality reduction and feature mixing, the condition that the spectral derivative feature is effective to the traditional classifier. On the basis of a large number of experiments, we conclude that under the following two conditions, the original spectral feature incorporating the first derivative feature can greatly improve the classification accuracy ratio: 1) the training set is relatively small and the quality of the training set is relatively small, and the training set is greatly affected by noise. At the same time, a large number of free download remote sensing data can be used for in-depth learning of unsupervised feature learning. In this paper, the depth confidence network (DBN) learning algorithm is used to detect the oil spill region on the NASA AVIRIS oil spill data in the Gulf of Mexico. The experimental classification results are in good agreement with the color map derived from the RGB band.
【學(xué)位授予單位】:復(fù)旦大學(xué)
【學(xué)位級別】:碩士
【學(xué)位授予年份】:2014
【分類號】:TP751
【相似文獻】
相關(guān)期刊論文 前10條
1 李熙;;基于應(yīng)用導(dǎo)向的遙感圖像處理實踐教學(xué)[J];科技資訊;2013年30期
2 單勇兵,趙軍;數(shù)學(xué)形態(tài)學(xué)在遙感圖像處理中的應(yīng)用[J];三晉測繪;2002年Z1期
3 李瑋;袁運能;朱博勤;;自適應(yīng)方法在遙感圖像處理中的應(yīng)用研究[J];遙感信息;2006年01期
4 馬偉鋒;岑崗;李君;沈占鋒;;高性能遙感圖像處理與空間信息網(wǎng)格建模[J];計算機工程;2006年05期
5 嚴紅萍;俞兵;;主成分分析在遙感圖像處理中的應(yīng)用[J];資源環(huán)境與工程;2006年02期
6 陳燕;齊清文;楊桂山;;基于遙感圖像處理的圖譜單元提取與分析[J];水土保持學(xué)報;2006年02期
7 秦軍;;鐵路工程地質(zhì)信息的遙感圖像處理[J];鐵道工程學(xué)報;2006年S1期
8 王楠;鄧靈博;何洪峰;;遙感圖像處理及其在軍事目標定位中的應(yīng)用[J];電腦知識與技術(shù)(學(xué)術(shù)交流);2007年18期
9 李黨輝;肖志豪;;企業(yè)級遙感圖像處理與應(yīng)用簡介[J];科技創(chuàng)新導(dǎo)報;2008年03期
10 馬江林;趙忠明;楊建;楊俊;汪承義;;基于設(shè)計模式的遙感圖像處理平臺設(shè)計與實現(xiàn)[J];計算機工程;2008年03期
相關(guān)會議論文 前7條
1 匡鴻海;況明生;;多媒體數(shù)據(jù)壓縮技術(shù)在遙感圖像處理中的應(yīng)用[A];中國圖象圖形學(xué)會第十屆全國圖像圖形學(xué)術(shù)會議(CIG’2001)和第一屆全國虛擬現(xiàn)實技術(shù)研討會(CVR’2001)論文集[C];2001年
2 武冰;周石琳;;遙感圖像處理中的數(shù)學(xué)形態(tài)學(xué)應(yīng)用[A];2005年中國智能自動化會議論文集[C];2005年
3 李瑋;朱博勤;袁運能;;自適應(yīng)方法在遙感圖像處理中的應(yīng)用研究[A];第十五屆全國遙感技術(shù)學(xué)術(shù)交流會論文摘要集[C];2005年
4 苑馨方;賈媛媛;李子揚;;蘆山地震遙感圖像處理與應(yīng)用[A];中國地震學(xué)會空間對地觀測專業(yè)委員會2013年學(xué)術(shù)研討會論文摘要集[C];2013年
5 李樹甫;陳亮;;偵察遙感圖像處理中的地球物理問題初探[A];國家安全工程地球物理研究——第二屆國家安全地球物理學(xué)術(shù)研討會論文集[C];2006年
6 馬成龍;李文科;;基于“3S”技術(shù)的干旱預(yù)警監(jiān)測研究[A];GPS在水利行業(yè)應(yīng)用研討會論文集[C];2007年
7 王華林;王東雷;王紀強;杜立英;劉霞;;淄博、濰坊地區(qū)遙感圖像處理與地質(zhì)構(gòu)造解譯[A];第二屆“測繪科學(xué)前沿技術(shù)論壇”論文精選[C];2010年
相關(guān)重要報紙文章 前1條
1 本報記者 許泳;PCI Geomatics推出新款遙感圖像處理產(chǎn)品[N];計算機世界;2009年
相關(guān)博士學(xué)位論文 前2條
1 董延華;超光譜遙感圖像處理關(guān)鍵技術(shù)研究[D];哈爾濱理工大學(xué);2006年
2 毛成林;復(fù)小波變換在遙感圖像處理方面的應(yīng)用研究[D];中國科學(xué)技術(shù)大學(xué);2013年
相關(guān)碩士學(xué)位論文 前10條
1 尹波;基于對象的遙感圖像處理平臺應(yīng)用研究[D];上海交通大學(xué);2008年
2 劉金亨;基于小波變換的遙感圖像處理研究[D];重慶大學(xué);2010年
3 陳燁;隨機共振在遙感圖像處理中的應(yīng)用研究[D];天津大學(xué);2009年
4 婁珊珊;基于偏微分方程的遙感圖像處理方法[D];大連海事大學(xué);2012年
5 謝永華;基于小波變換的遙感圖像處理的應(yīng)用研究[D];南京氣象學(xué)院;2002年
6 曹照清;遙感圖像處理的若干關(guān)鍵技術(shù)研究[D];南京航空航天大學(xué);2013年
7 陶靖琦;基于證據(jù)理論的無人機遙感圖像處理研究[D];湖北工業(yè)大學(xué);2012年
8 張德祥;方向小波變換在遙感圖像處理中的應(yīng)用研究[D];安徽大學(xué);2006年
9 陳超;可擴展遙感圖像處理平臺的研究與開發(fā)[D];中國地質(zhì)大學(xué)(北京);2014年
10 宋衛(wèi)艷;RANSAC算法及其在遙感圖像處理中的應(yīng)用[D];華北電力大學(xué)(北京);2011年
,本文編號:2215262
本文鏈接:http://sikaile.net/guanlilunwen/gongchengguanli/2215262.html