光谱特征选择之UVE算法

光谱特征选择之UVE算法

from sklearn.cross_decomposition import PLSRegression
from scipy import signal
import matplotlib.pyplot as plt
import numpy as np
from pandas import read_csv
import warnings
warnings.filterwarnings("ignore")

data = read_csv(r'C:\Users\Lenovo\Desktop\scores\tree leaf N-train-30ge.csv', header=None)  # 数据导入
x = np.array(data.loc[:, 1:])
y = np.array(data.loc[:, 0])
wl = np.linspace(866, 1701, 228)
#wl = np.linspace(0, 700, 700)


def uve(m, n):
    """
    :param m:  校正集样本数量
    :param n: 光谱特征数目
    :return:
    """
    noise = np.random.normal(loc=0, scale=0.01, size=(m, n))  # 产生一个(m,n)的噪音矩阵
    #noise = np.random.uniform(-0.0000000001, 0.0000000001, size=(m, n))
    new_x = np.append(x, noise, axis=1)  # 将噪音矩阵与光谱矩阵合并
    # 留一交叉验证，得到n个PLS回归系数组成的矩阵(n,2m)
    PLScoef = np.zeros((new_x.shape[0], new_x.shape[1]))
    for i in range(m):
        new_x_delete = np.delete(new_x, i, axis=0)  # 删除第i行，留一交叉验证
        y_delete = np.delete(y, i, axis=0)
        pls = PLSRegression(n_components=25)  # 第一步进行PLS回归的最佳主成分数
        pls.fit(new_x_delete, y_delete)
        #PLScoef[i, :] = pls.coef_.T.reshape(1, -1)[0]
        PLScoef[i, :] = pls.coef_.reshape(1, -1)[0]
    meancoef = np.mean(PLScoef, axis=0)
    stdcoef = np.std(PLScoef, axis=0)
    h = meancoef / stdcoef  # 按列计算矩阵的标准偏差和平均值向量，然后计算h(i)=me(i)/s(i)
    h_select = h[n:]
    h_max = np.max(abs(h_select))
    index = []
    for j in range(0, n):
        if h_max <= abs(h[j]):  # 保留绝对值大于h_max的变量
            index.append(j)
    selected_wave = index
    print(selected_wave)
    selected_wavelength = x[:, index]
    print(selected_wavelength.shape)
    return selected_wave


selected_wave1 = uve(120, 228)
plt.figure(figsize=(12, 8), dpi=100)
plt.plot(wl, x[2, :].T)
plt.scatter(wl[selected_wave1], x[2, :][selected_wave1], marker='*', color='red')
plt.show()