基于資源3號影像的陽澄湖圍網(wǎng)區(qū)自動(dòng)提取算法研究
發(fā)布時(shí)間：2019-01-12 09:15
【摘要】：隨著中國水產(chǎn)漁業(yè)的快速發(fā)展,長江流域的淺水湖泊圍網(wǎng)養(yǎng)殖業(yè)極速擴(kuò)張,極大地增加了經(jīng)濟(jì)效益,但超高密度的圍網(wǎng)養(yǎng)殖也對湖泊生態(tài)造成了嚴(yán)重破壞。在此背景下,合理科學(xué)地規(guī)劃并控制湖泊圍網(wǎng)區(qū)已被湖泊管理部門列為湖泊水質(zhì)提升和優(yōu)化的重點(diǎn)工作,而獲取并掌握湖泊圍網(wǎng)養(yǎng)殖的時(shí)空分布信息是湖泊生態(tài)管理部門在科學(xué)管理圍網(wǎng)養(yǎng)殖區(qū)時(shí)制定措施的依據(jù)。由于遙感技術(shù)具有實(shí)時(shí)、大范圍和客觀等優(yōu)勢,遙感監(jiān)測逐漸成為獲取湖泊圍網(wǎng)養(yǎng)殖區(qū)空間分布信息的主流方法。而目前對湖泊圍網(wǎng)養(yǎng)殖區(qū)的遙感識別方法主要是人工解譯湖泊的遙感影像,解譯時(shí)間長,受人工影響大。為進(jìn)一步提高解譯效率,部分學(xué)者開始研究機(jī)器解譯圍網(wǎng)區(qū)的方法,但大多數(shù)方法只使用遙感圖像的紋理特征,且識別時(shí)需要人為確定特征閾值,方法自動(dòng)化程度低。本文以長江流域典型的圍網(wǎng)型湖泊——陽澄湖為研究區(qū),基于高分辨率的資源3號衛(wèi)星(ZY-3)影像,運(yùn)用模式識別理論,設(shè)計(jì)了圍網(wǎng)養(yǎng)殖區(qū)自動(dòng)識別方案。該方案主要包括遙感影像預(yù)處理,邊界信息增強(qiáng)、圖像噪聲濾除、閾值分割、特征提取、分類識別6個(gè)步驟。主要研究內(nèi)容如下:(1)基于圖像處理的影像增強(qiáng)和分割算法。首先,分析了圍網(wǎng)邊界增強(qiáng)的必要性,并利用梯度變換和灰度變換進(jìn)行圍網(wǎng)區(qū)邊界信息增強(qiáng),結(jié)果表明,該方法有效的增強(qiáng)了圍網(wǎng)區(qū)的邊界信息;然后,描述了自適應(yīng)濾波算法的濾波原理和濾波流程,并運(yùn)用自適應(yīng)濾波算法濾除了遙感影像中的椒鹽噪聲和高斯噪聲;最后,研究了閾值分割的基本原理,使用了小波變換精確尋找到進(jìn)行閾值分割的閾值,結(jié)果表明,當(dāng)閾值為130和180時(shí),圖像能很好的完成閾值分割任務(wù)。(2)基于傅里葉變換的圍網(wǎng)區(qū)特征提取。首先,通過分析圍網(wǎng)區(qū)的光譜信息,明確了陽澄湖圍網(wǎng)區(qū)和水體最具可分性的波段是近紅外波段;然后,基于資源3號陽澄湖影像的近紅外波段,利用二維離散傅里葉變換對湖泊遙感影像中的每一小塊區(qū)域進(jìn)行變換,選取了以直流信號為中心的頻率譜特征;最后對二維頻率譜進(jìn)行奇異值分解,并將得到的奇異值重新組成特征向量,達(dá)到對原特征向量降維的目的。(3)基于最近鄰分類的圍網(wǎng)區(qū)識別。首先,通過分析不同分類方法的優(yōu)缺點(diǎn),明確了最近鄰分類法作為陽澄湖圍網(wǎng)區(qū)識別方法;然后研究了最近鄰分類法的基本原理,分析了最近鄰分類法的基本操作步驟,并使用最近鄰分類法對陽澄湖圍網(wǎng)養(yǎng)殖區(qū)進(jìn)行了分類識別;最后,使用混淆矩陣,以人工目視解譯的圍網(wǎng)區(qū)為評價(jià)標(biāo)準(zhǔn)對算法的分類結(jié)果進(jìn)行了評價(jià),提取精度為86%。
[Abstract]:With the rapid development of aquatic fishery in China, the shallow water lake seine aquaculture industry in the Yangtze River valley has expanded extremely rapidly, which has greatly increased the economic benefits, but the super-high density seine culture has also caused serious damage to the lake ecology. In this context, the rational and scientific planning and control of the lake seine area has been listed as the key task of lake water quality improvement and optimization by the lake management department. Therefore, obtaining and mastering the spatial and temporal distribution information of lake seine culture is the basis for the lake ecological management department to formulate measures in the scientific management of seine culture area. Because remote sensing technology has the advantages of real-time, large-scale and objective, remote sensing monitoring has gradually become the mainstream method to obtain spatial distribution information in lake seine culture area. At present, the method of remote sensing recognition of lake seine culture area is mainly artificial interpretation of lake remote sensing image, interpretation time is long, and is greatly affected by artificial. In order to further improve the interpretation efficiency, some scholars have begun to study the method of machine interpretation of the seine area, but most of the methods only use the texture features of remote sensing images, and the recognition needs to determine the threshold of features artificially, so the automation of the method is low. In this paper, Yangcheng Lake, a typical seine lake in the Yangtze River Basin, is used as the study area. Based on the high resolution ZY-3 image, the automatic identification scheme of the seine culture area is designed by using the pattern recognition theory. The scheme mainly includes six steps: remote sensing image preprocessing, edge information enhancement, image noise filtering, threshold segmentation, feature extraction and classification and recognition. The main contents are as follows: (1) Image enhancement and segmentation algorithm based on image processing. Firstly, the necessity of the enhancement of the seine boundary is analyzed, and the edge information of the seine is enhanced by gradient transformation and gray transformation. The results show that the method can effectively enhance the boundary information of the seine. Then, the filtering principle and filtering flow of adaptive filtering algorithm are described, and the pepper and salt noise and Gao Si noise in remote sensing image are filtered by adaptive filtering algorithm. Finally, the basic principle of threshold segmentation is studied, and the threshold of threshold segmentation is accurately found by wavelet transform. The results show that when the threshold is 130 and 180, The image can accomplish the task of threshold segmentation well. (2) the feature extraction of seine area based on Fourier transform. Firstly, by analyzing the spectral information of the seine area, it is clear that the most divisible band between Yangcheng Lake seine area and water body is near infrared band. Then, based on the near infrared band of Yangcheng Lake image of Resource-3, every small region of lake remote sensing image is transformed by two-dimensional discrete Fourier transform, and the frequency spectrum characteristic centered on DC signal is selected. Finally, the singular value of the two-dimensional frequency spectrum is decomposed, and the singular value is recomposed into the eigenvector to reduce the dimension of the original eigenvector. (3) the identification of the seine area based on the nearest neighbor classification. Firstly, by analyzing the advantages and disadvantages of different classification methods, the nearest neighbor classification method is defined as the identification method of Yangcheng Lake seine area. Then the basic principle of nearest neighbor classification is studied, the basic operation steps of nearest neighbor classification are analyzed, and the most nearest neighbor classification method is used to classify and identify the seine culture area of Yangcheng Lake. Finally, the classification results of the algorithm are evaluated by using the confusion matrix and the artificial visual interpretation of the seine area as the evaluation criteria. The extraction accuracy is 86%.
【學(xué)位授予單位】：安徽大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP751
本文編號：2407609
[Abstract]:With the rapid development of aquatic fishery in China, the shallow water lake seine aquaculture industry in the Yangtze River valley has expanded extremely rapidly, which has greatly increased the economic benefits, but the super-high density seine culture has also caused serious damage to the lake ecology. In this context, the rational and scientific planning and control of the lake seine area has been listed as the key task of lake water quality improvement and optimization by the lake management department. Therefore, obtaining and mastering the spatial and temporal distribution information of lake seine culture is the basis for the lake ecological management department to formulate measures in the scientific management of seine culture area. Because remote sensing technology has the advantages of real-time, large-scale and objective, remote sensing monitoring has gradually become the mainstream method to obtain spatial distribution information in lake seine culture area. At present, the method of remote sensing recognition of lake seine culture area is mainly artificial interpretation of lake remote sensing image, interpretation time is long, and is greatly affected by artificial. In order to further improve the interpretation efficiency, some scholars have begun to study the method of machine interpretation of the seine area, but most of the methods only use the texture features of remote sensing images, and the recognition needs to determine the threshold of features artificially, so the automation of the method is low. In this paper, Yangcheng Lake, a typical seine lake in the Yangtze River Basin, is used as the study area. Based on the high resolution ZY-3 image, the automatic identification scheme of the seine culture area is designed by using the pattern recognition theory. The scheme mainly includes six steps: remote sensing image preprocessing, edge information enhancement, image noise filtering, threshold segmentation, feature extraction and classification and recognition. The main contents are as follows: (1) Image enhancement and segmentation algorithm based on image processing. Firstly, the necessity of the enhancement of the seine boundary is analyzed, and the edge information of the seine is enhanced by gradient transformation and gray transformation. The results show that the method can effectively enhance the boundary information of the seine. Then, the filtering principle and filtering flow of adaptive filtering algorithm are described, and the pepper and salt noise and Gao Si noise in remote sensing image are filtered by adaptive filtering algorithm. Finally, the basic principle of threshold segmentation is studied, and the threshold of threshold segmentation is accurately found by wavelet transform. The results show that when the threshold is 130 and 180, The image can accomplish the task of threshold segmentation well. (2) the feature extraction of seine area based on Fourier transform. Firstly, by analyzing the spectral information of the seine area, it is clear that the most divisible band between Yangcheng Lake seine area and water body is near infrared band. Then, based on the near infrared band of Yangcheng Lake image of Resource-3, every small region of lake remote sensing image is transformed by two-dimensional discrete Fourier transform, and the frequency spectrum characteristic centered on DC signal is selected. Finally, the singular value of the two-dimensional frequency spectrum is decomposed, and the singular value is recomposed into the eigenvector to reduce the dimension of the original eigenvector. (3) the identification of the seine area based on the nearest neighbor classification. Firstly, by analyzing the advantages and disadvantages of different classification methods, the nearest neighbor classification method is defined as the identification method of Yangcheng Lake seine area. Then the basic principle of nearest neighbor classification is studied, the basic operation steps of nearest neighbor classification are analyzed, and the most nearest neighbor classification method is used to classify and identify the seine culture area of Yangcheng Lake. Finally, the classification results of the algorithm are evaluated by using the confusion matrix and the artificial visual interpretation of the seine area as the evaluation criteria. The extraction accuracy is 86%.
【學(xué)位授予單位】：安徽大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP751
【參考文獻(xiàn)】
相關(guān)期刊論文 前10條
1 李永生;武鵬飛;;基于MODIS數(shù)據(jù)的艾比湖湖面變化研究[J];水資源與水工程學(xué)報(bào);2008年05期
2 王靜;高俊峰;;基于對應(yīng)分析的湖泊圍網(wǎng)養(yǎng)殖范圍提取[J];遙感學(xué)報(bào);2008年05期
3 馬明國;宋怡;王雪梅;;1973-2006年新疆若羌湖泊群遙感動(dòng)態(tài)監(jiān)測研究[J];冰川凍土;2008年02期
4 曹榮龍;李存軍;劉良云;王紀(jì)華;閻廣建;;基于水體指數(shù)的密云水庫面積提取及變化監(jiān)測[J];測繪科學(xué);2008年02期
5 張同尊;邵俊松;方勇杰;;一種基于離散傅里葉變換的頻率測量算法[J];電力系統(tǒng)自動(dòng)化;2007年22期
6 韓祥珍;厲恩華;袁龍義;李偉;;圍網(wǎng)養(yǎng)殖對水生植被和沉積物再懸浮的影響[J];湖北農(nóng)業(yè)科學(xué);2007年04期
7 高連如;張兵;張霞;申茜;;基于局部標(biāo)準(zhǔn)差的遙感圖像噪聲評估方法研究[J];遙感學(xué)報(bào);2007年02期
8 韓芳;李興華;高拉云;;內(nèi)蒙古達(dá)里諾爾湖泊濕地動(dòng)態(tài)的遙感監(jiān)測[J];內(nèi)蒙古農(nóng)業(yè)大學(xué)學(xué)報(bào)(自然科學(xué)版);2007年01期
9 瞿鈞;甘嵐;;梯度Hough變換在圓檢測中的應(yīng)用[J];華東交通大學(xué)學(xué)報(bào);2007年01期
10 李淑霞;王汝霖;李春梅;許亮;李國新;;基于噪聲方差估計(jì)的小波閾值圖像去噪新方法[J];計(jì)算機(jī)應(yīng)用研究;2007年01期
，本文編號：2407609
本文鏈接：http://sikaile.net/guanlilunwen/gongchengguanli/2407609.html
最近更新
教材專著
