💥1 概述
从高光谱图像中提取纯端体是目标检测、分类和解混应用中非常重要的步骤。利用凸几何的概念，提出了一种新的端部提取算法。该算法使用凸多边形最大化来确定一个凸集，该凸集根据测量员的公式给出最大的凸多边形面积。所提算法的并行实现有助于找到更有效的独特像素。通过合成数据证明了所提算法在存在噪声时的鲁棒性。利用真实高光谱数据进行的仿真结果表明，所提算法将光谱角度误差（SAE）和光谱信息发散误差（SID）降低了2.4–8.8%。所提算法在丰度映射中的有效性也使用均方根误差（RMSE）进行了验证。所提算法的RMSE也提高了1.7–7.6%。
📚2 运行结果
部分代码：
%% demo_CPM clc; close all; clear all; addpath('../data'); %% Image Read s=load('Cuprite.mat'); % link for data source : https://rslab.ut.ac.ir/data p=s.nRow; q=s.nCol; Bands=188; Y=s.Y; x=hyperConvert3d(Y,p,q,Bands); %% Virtual Dimension VD=12; %% CPM algorithm [endmemberindex] = CPM(Y,VD); endmemberindex_CPM=change_index(endmemberindex,p,q); %% VCA algorithm [U_VCA,e_index,snrEstimate]=hyperVca(Y,VD); endmemberindex_VCA=change_index(e_index,p,q); %% gt compare t1=load('groundTruth_Cuprite_nEnd12.mat'); gt=t1.M; n1=gt(3:103,:); n2=gt(114:147,:); n3=gt(168:220,:); gt=[n1;n2;n3]; [gt_m,gt_n]=size(gt); for i=1:gt_n for j=1:Bands extracted_VCA(j,i)=x(endmemberindex_VCA(i,1),endmemberindex_VCA(i,2),j); extracted_CPM(j,i)=x(endmemberindex_CPM(i,1),endmemberindex_CPM(i,2),j); end end %% SAM Calculation ex_n=gt_n; store_VCA=[0,0]; store_CPM=[0,0]; sam_VCA=0; sam_CPM=0; sam_total_VCA=0; sam_total_CPM=0; for i=1:gt_n for j=1:ex_n Mat_SAM_CPM(i,j)=real(acos(dot(gt(:,i),extracted_CPM(:,j))/(norm(gt(:,i)*norm(extracted_CPM(:,j)))))); Mat_SAM_VCA(i,j)=real(acos(dot(gt(:,i),extracted_VCA(:,j))/(norm(gt(:,i)*norm(extracted_VCA(:,j)))))); end end for i=1:gt_n %CPM [max_value1,mrow]=min(Mat_SAM_CPM); [max_value,col_CPM]=min(max_value1); sam_total_CPM=sam_total_CPM+max_value; sam_CPM=[sam_CPM;max_value]; row_CPM=mrow(col_CPM); s1=[row_CPM,col_CPM]; store_CPM=[store_CPM;s1]; save_CPM(row_CPM)=max_value; Mat_SAM_CPM(row_CPM,:)=[100*ones]; Mat_SAM_CPM(:,col_CPM)=[100*ones]; %VCA [max_value1,mrow]=min(Mat_SAM_VCA); [max_value,col_VCA]=min(max_value1); sam_total_VCA=sam_total_VCA+max_value; sam_VCA=[sam_VCA;max_value]; row_VCA=mrow(col_VCA); s1=[row_VCA,col_VCA]; store_VCA=[store_VCA;s1]; save_VCA(row_VCA)=max_value; Mat_SAM_VCA(row_VCA,:)=[100*ones]; Mat_SAM_VCA(:,col_VCA)=[100*ones]; end rms_sae=[rms(save_CPM); rms(save_VCA)]; rms_sae = radtodeg(rms_sae); disp('RMSSAE of VCA'); disp(rms_sae(2)); disp('RMSSAE of CPM'); disp(rms_sae(1));
🌈3 Matlab代码实现
🎉4 参考文献
部分理论来源于网络，如有侵权请联系删除。
[1]Dharambhai Shah, Tanish Zaveri, Yogesh Trivedi (2020) Convex Polygon Maximization-Based Hyperspectral Endmember Extraction Algorithm
