机器学习GDA的高光谱遥感数据分类(标准英文版论文)

机器学习GDA的高光谱遥感数据分类(标准英文版 论文 政研论文下载论文大学下载论文大学下载关于长拳的论文浙大论文封面下载 ) Class Student ID Xi’an University Machine learning Subject Hyperspectral remote sensing data classification Academy Electronic Engineering Major Intelligence Science and Technology Student Name Instructor Name 基于GDA的高光谱遥感数据分类 (作者:你若不来) (西安电子科技大学 电子工程学院，陕西 西安 710126) Abstract: Hyperspectral remote sensing technology is a new earth observation technologies. This application of this new technology to the Indian Pines scene detection, hyperspectral data obtained IndianPine. This data is 145 * 145 size, there are 200 bands. . Then apply this term "machine learning" course taught by an efficient and simple algorithm - Gaussian discriminant analysis, IndianPine data classification, the last of points out of the 16 categories of data, respectively, to calculate the correct rate, and in the matlab environment to do the signature, which can display vivid and intuitive classification results. Keywords hyperspectral remote sensing ; Gaussian discriminant analysis classification; accuracy ;signature 摘要:高光谱遥感技术是一种新型的对地观测技术。 关键词 高光谱遥感技术; 高斯判别 分析 定性数据统计分析pdf销售业绩分析模板建筑结构震害分析销售进度分析表京东商城竞争战略分析 ; 分类 ; 正确率 ; 标记图; 引言: 高光谱遥感技术是一种新的地球观测技术。结合传统的遥感成像技术和物理光 谱路口站，在 检测 工程第三方检测合同工程防雷检测合同植筋拉拔检测方案传感器技术课后答案检测机构通用要求培训 对象空间的空间特征，对每一个像素，而形成约10-20nm的 连续光谱带的色散波长带宽的数十或数百个，所以得到的作为统一的高光谱遥感 图像的光谱进行了丰富的光谱信息，是非常有利的地形分类战场情报侦察，目标 识别，监测气候变化，植物生长监测，灾害监测，数字地球等领域中被广泛使用。 要求达到最小标记的样品每5级至10时准确的地形分类。测试数据库的描述如 下:Indian Pines Here you can find information over some public available hyperspectral scenes. All of then are Earth Observation images taken from airbornes or satellites. You can find more information about hyperspectral sensors and remote sensing here. This scene was gathered by AVs scene contains two-thirds agriculture, and one-third forest or other natural perennial vegetation. There are two major dual lane highways, a rail line, as well as some low density housing, other built structures, and smaller roads. Since the scene is taken in June some of the crops present, corn, soybeans, are in early stages of growth with less than 5% coverage. The ground truth available is designated into sixteen classes and is not all mutually exclusive. We have also reduced the number of bands to 200 by removing bands covering the region of water absorption: [104-108], [150-163], 220. Indian Pines data are available through Pursue’s univeristy MultiSpec site Here,wo just use this type of date to be a model with which we will just want to find a method to classify.Gaussian discriminant analysis as an efficient and easy method might be a good choice. 方法 快递客服问题件处理详细方法山木方法pdf计算方法pdf华与华方法下载八字理论方法下载 : 高斯判别分析 The first generative learning algorithm that we’ll look at is Gaussian discriminantanalysis(GDA). In this model, we’ll assume that p(x|y) is distributed according to multivariate normal distribution. Let’s talk briefly about the properties of multivariate normal distributions before moving on to the GDA model itself. The multivariate normal distribution: The multivariate normal distribution in n-dimensions, also called the multivariate Gaussian distribution, is parameterized by a mean vector and a covariance matrix , where is symmetric and positive semi-definite. Also written its density is given by: The Gaussian Discriminant Analysis model When we have a classification problem in which the input features x arecontinuous-valued random variables, we can then use the Gaussian Discriminant Analysis (GDA) model, which modelspxyusing a multivariate normal distribution. The model is:: Writing out the distribution, this is: The log- By maximizing function l with respect to the parameters, we find the maximum likelihood estimate of the parameters to be: mi mi Pictorially, what the algorithm is doing can be seen in as follows: Shown in the figure are the training set, as well as the contours of the two Gaussian distributions that have been fit to the data in each of the two classes. Note that the two Gaussians have contours that are the same shape and orientation, since they share a covariance matrix , but they have different means and . Also shown in the figure is the straight line giving the decision boundary at which p(y = 1|x) = 0.5. On one side of the boundary, we’ll predict y = 1 to be the most likely outcome, and on the other side, we’ll predict y = 0. 实验结果: Conclusions: Hyperspectral remote sensing technology as a new earth observation technology, its spectrum as one of hyperspectral remote sensing images carry a wealth of spectral information. This scenario for the Indian Pines hyperspectral data IndianPine discriminant analysis using Gaussian hyperspectral data classification, classification results are good, the recognition rate pricey. Hyperspectral technology combined with Gaussian discriminant analysis, I believe that in the future the prediction of surface vegetation, crops growing and so will play a guiding role. Under certain accuracy requirements, references in this paper has value. 1.A Gaussian discriminant analysis constraint is strong, but the number of training samples required for less. In case of insufficient number of samples is very convenient. 2. From the histogram of classification can be seen in certain categories, while the high accuracy, there will be a number of low accuracy, because the training sample selection is random, although for the Laplacian smoothing However, if certain types of samples is extremely small, only a worst case, this case will result in the recognition rate disparity between the different categories. 3. in the form of papers and English accomplish the job so I really exercise their own, very difficult chance. 高光谱遥感技术作为一种新型的对地观测技术，其谱像合一的高光谱遥感图像 携带了丰富的光谱信息。 1. 高斯判别分析约束性强，但对训练样本数目要求少。在样本数目不足的情况 下很是方便。 2. 从分类结果的柱状图可以看出，在某些类别正确率很高的同时，也会有一些 正确率很低，原因是训练样本的选择是随机的，虽然进行了拉普拉斯平滑，但是 如果某类样本极其少，最坏情况下只有一个，这种情况下就会造成不同类别间识 别率悬殊。 3. 以论文并且英文的形式完成大作业让我真正锻炼了自己，很难的一次机会。 Reference 1. 2. 3. ppt 关于艾滋病ppt课件精益管理ppt下载地图下载ppt可编辑假如ppt教学课件下载triz基础知识ppt making by Shuyuan Yang matlab 7.x Xidian university press Shuntisnlou Ruoyuyao CS229 Lecture notes Andrew Ng Appendix close all;clear;clc; load IndianaPines.mat; tic [qws,qzl]=size(pixels); Label=double(Label); ZsLabel = length(unique(Label))-1; labels=unique(Label); double(labels); pixels1=cat(1,pixels,Label’); register=find(pixels1(201,:)==0); register1=find(pixels1(201,:)~=0); pixels1(:,register)=[]; %È??ý0?êÇ? [ws,zl]=size(pixels1); mm=rand(1,zl); m=0; for i=1:zl if mm(1,i)<=0.1 m=m+1; train(:,m)=pixels1(:,i); %?úÉúѵÁ?Ñù?? end end for i=1:ZsLabel sery=find(pixels1(ws,:)==i); smooth(:,i)=pixels1(:,sery(1)); %À-ÆÕÀ-Ë?Æ??? end train=cat(2,train,smooth); fai = zeros(1,ZsLabel); u = zeros(ws-1,ZsLabel); for i=1:ZsLabel c = (train(201,:)==labels(i+1)); fai(i)=sum(c)/(m+16); %?ÆËã?ÎÊý u(:,i)=mean(train(1:200,c),2); end xfc=cov(train(1:200,:)’); xfc=inv(xfc); p = zeros(1,ZsLabel); logfai = log(fai); for i=1:zl for k=1:16 hy(k)=logfai(1,k)-0.5*(pixels1(1:200,i)-u(:,k))’*xfc*(pixels1(1:200,i)-u(:,k)); end %Ê??ð?ý?Ì [maxP,maxI] = max(hy); pixels1(202,i)=maxI; end cnt=0; for i=1:zl if pixels1(201,i)==pixels1(202,i) cnt=cnt+1; end %?ÆËãÊ??ðÂÊ end accurate=cnt/zl; Label(register,2)=0; Label(register1,2)=pixels1(202,:); %??Ô-Àà?ð?êÇ? cnt1=0; for i=1:qzl if Label(i,1)==Label(i,2) cnt1=cnt1+1; end end accurate=cnt1/qzl for k=1:16 Zcn=0;cnt=0; for i=1:21025 if Label(i,1)==k Zcn=Zcn+1; if Label(i,2)==Label(i,1) cnt=cnt+1; end end end fprintf(‘µÚ %2d Àà %f\n’,k,cnt/Zcn); sbjz(k,1)=cnt/Zcn; end kk=1:1:16; figure (1); stem(kk,sbjz,’fill’,’b’) toc; colorMap = rand(qzl,3); colorMap(1,:)=0; img = zeros(145,145); img(:) = Label(:,1); figure (2); subplot 121; imshow(img,colorMap); title(‘yushi image’); img(:)= Label(:,2); subplot 122; imshow(img,colorMap); title(‘shibie image’);