另人问的问题,有点用,记录下来,是用matlab实现的。当数据比较多时,可以不通过for实现,用函数的形式,但有可能需要多一点的内存。
http://www.ilovematlab.cn/forum.php?mod=viewthread&tid=319551&pid=2592723&page=1&extra=#pid2592723
算法是帖子中提供的,
然后我通过两种方式实现的。思路如图
结果如图
clear all
clc
D=[1 1.1 1.2 1.1 1.05 1.0 0.997 0.995 0.8 0.7 0.6 0.5 0.67 0.72 ];
n_D=length(D);
% peaks
findpeaks(D)
% neg peaks
-findpeaks(-D)
meps=0.04;
%% for version
gu=[];
for i=2:(n_D-1)
jud1 = (D(i)-D(i-1))*(D(i+1)-D(i)) >= 0 ;
jud2 = abs(D(i+1)-2*D(i)+D(i-1))>meps ;
if jud1 & jud2 % monotone
gu=[gu;i D(i)];
end
end
%% function version
[ gu_D, gu_ind] = fun_mm_find_guaidian( D ,meps);
gu_D
%% output result
plot(D,'.-')
hold on
% for version
plot(gu(:,1),gu(:,2),'ro','MarkerSize',10)
% function version
plot(gu_ind,gu_D,'g*')
%% logs
% mod : 24-Feb-2015 09:33:35
函数文件:
function [ gu_D, gu_ind] = fun_mm_find_guaidian( D,meps )
%UNTITLED2 Summary of this function goes here
% Detailed explanation goes here
n_D=length(D);
delt1=diff(D);
delt2=diff(delt1);
judm1=delt1(1:(n_D-2)).*delt1(2:n_D-1)>=0;
judm2=abs(delt2)>meps;
gu_ind = find(judm1 & judm2)+1;
gu_D = D(gu_ind);
end