如何绘制 MNIST 随机图像的频率分析？

Question

我想加载 MNIST 并计算一些随机图像的功率谱的振幅。我收到的与我预期的非常不同。如何计算功率谱的振幅然后将它们并排绘制（我不知道该怎么做——所以我只绘制光谱）

这是我的尝试：

import torch
import numpy as np
import matplotlib.pyplot as plt
from torchvision import datasets, transforms

# Define the batch size
batch_size = 32

# Define the transform to normalize the images
transform = transforms.Compose([transforms.ToTensor(), 
                                transforms.Normalize((0.5,), (0.5,))])

# Load the dataset
trainset = datasets.MNIST(root='./data', train=True, download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True)

# The Fourier transform function
def fourier_transform(image):
  
    image = image.numpy()[0] # Convert the tensor to a numpy array
    f = np.fft.fft2(image) # Compute the 2D Fourier transform
    fshift = np.fft.fftshift(f) # Shift the zero-frequency component to the center of the spectrum
    magnitude_spectrum = np.log(np.abs(fshift)) # Compute the magnitude spectrum and take the logarithm for visualization
    
    return magnitude_spectrum


X,y = next(iter(trainloader))
images, labels = next(iter(trainloader))
fig,ax = plt.subplots(3,6,figsize=(10,6))

for (i,ax) in enumerate(ax.flatten()):

  # extract that image
  magnitude_spectrum = fourier_transform(images[i])
  
  
  # and its label
  label = trainset.classes[y[i]]

  # and show
  ax.imshow(magnitude_spectrum)
  ax.text(14,0,label,ha='center',fontweight='bold',color='k',backgroundcolor='y')
  ax.axis('off')

plt.tight_layout()
plt.show()

我的情节是这样的：

Answer 1

你需要取傅里叶变换的绝对值并平方。在您的

fourier_transform

函数中，替换此行：

magnitude_spectrum = np.abs(fshift) ** 2 

这条线：

magnitude_spectrum = np.abs(fshift) **