PaviaU类似高光谱数据集

pavia_u_like/
├── README.md                           # 本文档
├── LICENSE                             # MIT许可证
├── pavia_u_visualization.png           # 可视化图表
│
├── pavia_u_urban.mat                   # 城市场景 (MATLAB格式)
├── pavia_u_urban.npy                   # 城市场景 (NumPy格式)
├── pavia_u_urban_wavelengths.txt       # 城市场景波长数据
├── pavia_u_urban_gt.mat                # 城市场景标注 (MATLAB格式)
├── pavia_u_urban_gt.npy                # 城市场景标注 (NumPy格式)
│
├── pavia_u_agriculture.mat             # 农业场景 (MATLAB格式)
├── pavia_u_agriculture.npy             # 农业场景 (NumPy格式)
├── pavia_u_agriculture_wavelengths.txt # 农业场景波长数据
├── pavia_u_agriculture_gt.mat          # 农业场景标注 (MATLAB格式)
├── pavia_u_agriculture_gt.npy          # 农业场景标注 (NumPy格式)
│
├── pavia_u_mixed.mat                   # 混合场景 (MATLAB格式)
├── pavia_u_mixed.npy                   # 混合场景 (NumPy格式)
├── pavia_u_mixed_wavelengths.txt       # 混合场景波长数据
├── pavia_u_mixed_gt.mat                # 混合场景标注 (MATLAB格式)
└── pavia_u_mixed_gt.npy                # 混合场景标注 (NumPy格式)

load('pavia_u_urban.mat')

% 可用变量:
% - data: 高光谱数据 (340 × 610 × 103)
% - wavelengths: 波长数组 (103 × 1)
% - shape: 数据形状 [340, 610, 103]
% - metadata: 元数据字符串

import numpy as np

# 加载数据
hsi_data = np.load('pavia_u_urban.npy')
# 形状: (340, 610, 103)
# 类型: float64

# 加载波长
wavelengths = np.loadtxt('pavia_u_urban_wavelengths.txt', skiprows=1)
# 形状: (103,)
# 单位: nm

Wavelength (nm)
430.00
434.27
438.54
...
860.00

import numpy as np
import scipy.io as sio

# 加载标注数据
gt = sio.loadmat('pavia_u_urban_gt.mat')
labels = gt['gt']  # (340, 610)
# 或
labels = np.load('pavia_u_urban_gt.npy')  # (340, 610)

import numpy as np
import scipy.io as sio
import matplotlib.pyplot as plt

# 加载MAT文件
mat_data = sio.loadmat('pavia_u_urban.mat')
hsi_data = mat_data['data']  # (340, 610, 103)
wavelengths = mat_data['wavelengths'].flatten()  # (103,)

print(f'数据形状: {hsi_data.shape}')
print(f'波长范围: {wavelengths[0]:.1f} - {wavelengths[-1]:.1f} nm')
print(f'反射率范围: {hsi_data.min():.4f} - {hsi_data.max():.4f}')

# RGB合成显示
def get_rgb_image(hsi_data, r_band=80, g_band=40, b_band=20):
    """从高光谱数据提取RGB图像"""
    rgb = np.zeros((hsi_data.shape[0], hsi_data.shape[1], 3))
    rgb[:, :, 0] = hsi_data[:, :, r_band]  # Red
    rgb[:, :, 1] = hsi_data[:, :, g_band]  # Green
    rgb[:, :, 2] = hsi_data[:, :, b_band]  # Blue
    
    # 归一化到 0-1
    rgb = (rgb - rgb.min()) / (rgb.max() - rgb.min())
    return rgb

rgb_img = get_rgb_image(hsi_data)
plt.figure(figsize=(10, 6))
plt.imshow(rgb_img)
plt.title('PaviaU Urban - RGB合成')
plt.axis('off')
plt.show()

# 显示光谱曲线
pixel_y, pixel_x = 170, 305  # 中心像素
spectrum = hsi_data[pixel_y, pixel_x, :]

plt.figure(figsize=(12, 4))
plt.plot(wavelengths, spectrum, linewidth=2)
plt.xlabel('波长 (nm)')
plt.ylabel('反射率')
plt.title(f'像素 ({pixel_x}, {pixel_y}) 光谱曲线')
plt.grid(True, alpha=0.3)
plt.show()

from sklearn.preprocessing import StandardScaler

# 加载数据
hsi_data = np.load('pavia_u_urban.npy')

# 1. 去除噪声波段（可选）
valid_bands = np.arange(10, 95)  # 去除前10和后8个波段
hsi_clean = hsi_data[:, :, valid_bands]

# 2. 归一化
scaler = StandardScaler()
h, w, b = hsi_clean.shape
hsi_normalized = scaler.fit_transform(
    hsi_clean.reshape(-1, b)
).reshape(h, w, b)

# 3. PCA降维
from sklearn.decomposition import PCA

n_components = 30
pca = PCA(n_components=n_components)
hsi_pca = pca.fit_transform(
    hsi_clean.reshape(-1, b)
).reshape(h, w, n_components)

print(f'原始数据: {hsi_data.shape}')
print(f'PCA后数据: {hsi_pca.shape}')
print(f'解释方差比: {pca.explained_variance_ratio_.sum():.2%}')

from sklearn.svm import SVC
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report, accuracy_score

# 加载数据
hsi_data = np.load('pavia_u_urban.npy')
h, w, b = hsi_data.shape

# 创建模拟标签（实际应用中应使用真实标签）
# 这里简单地将图像分为4个区域
labels = np.zeros((h, w), dtype=int)
labels[:h//2, :w//2] = 0  # 类别0
labels[:h//2, w//2:] = 1  # 类别1
labels[h//2:, :w//2] = 2  # 类别2
labels[h//2:, w//2:] = 3  # 类别3

# 准备训练数据
X = hsi_data.reshape(-1, b)
y = labels.flatten()

# 采样部分数据用于训练（加速）
sample_idx = np.random.choice(len(X), size=5000, replace=False)
X_sample = X[sample_idx]
y_sample = y[sample_idx]

# 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(
    X_sample, y_sample, test_size=0.3, random_state=42
)

# 训练SVM分类器
clf = SVC(kernel='rbf', random_state=42)
clf.fit(X_train, y_train)

# 预测
y_pred = clf.predict(X_test)

# 评估
print(f'准确率: {accuracy_score(y_test, y_pred):.4f}')
print('\n分类报告:')
print(classification_report(y_test, y_pred))

% 加载数据
load('pavia_u_urban.mat');

% 显示基本信息
disp(['数据形状: ', num2str(shape)]);
disp(['波长范围: ', num2str(wavelengths(1)), ' - ', num2str(wavelengths(end)), ' nm']);

% RGB合成显示
rgb_img = zeros(340, 610, 3);
rgb_img(:,:,1) = data(:,:,80);  % Red
rgb_img(:,:,2) = data(:,:,40);  % Green
rgb_img(:,:,3) = data(:,:,20);  % Blue

% 归一化
rgb_img = (rgb_img - min(rgb_img(:))) / (max(rgb_img(:)) - min(rgb_img(:)));

figure;
imshow(rgb_img);
title('PaviaU Urban - RGB合成');
colorbar;

% 显示光谱曲线
pixel_y = 170;
pixel_x = 305;
spectrum = squeeze(data(pixel_y, pixel_x, :));

figure;
plot(wavelengths, spectrum, 'LineWidth', 2);
xlabel('波长 (nm)');
ylabel('反射率');
title(['像素 (', num2str(pixel_x), ', ', num2str(pixel_y), ') 光谱曲线']);
grid on;

% PCA降维
load('pavia_u_urban.mat');
[h, w, b] = size(data);

% 重塑为2D矩阵
X = reshape(data, [h*w, b]);

% PCA
[coeff, score, latent] = pca(X, 'NumComponents', 30);

% 重塑回3D
data_pca = reshape(score, [h, w, 30]);

disp(['原始波段数: ', num2str(b)]);
disp(['PCA后波段数: ', num2str(30)]);
disp(['解释方差: ', num2str(sum(latent(1:30))/sum(latent)*100), '%']);

@dataset{pavia_u_like_2026,
  author = {HSI Generator Project},
  title = {PaviaU-Like Hyperspectral Dataset},
  year = {2026},
  version = {1.0},
  generator = {HSI Generator v4.0.0},
  url = {https://atomgit.com/hunyuan2026/ggp}
}