如何在卷积神经网络(CNN)中可视化滤波器和特征映射图？

首先加载相关模块

import requests
import numpy as np 
import matplotlib.pyplot as plt 
import tensorflow as tf 
from tensorflow.keras.applications.vgg16 import VGG16, preprocess_input
from tensorflow.keras.utils import img_to_array, get_file
from tensorflow.keras.preprocessing.image import load_img
from skimage.transform import resize

加载标签

LABELS_PATH = './labels.txt'

with open(LABELS_PATH, 'rt') as f:
    labels = [label[:-1] for label in f.readlines()]

print('Num of Labels: %s' % len(labels))
print('Labels Sample: ', labels[:5])

Num of Labels: 1000

Labels Sample: ['tench, Tinca tinca', 'goldfish, Carassius auratus', 'great white shark, white shark, man-eater, man-eating shark, Carcharodon carcharias', 'tiger shark, Galeocerdo cuvieri', 'hammerhead, hammerhead shark']

预处理

INPUT_SHAPE = (224, 224, 3)
def load_and_preprocess(image_path):
    img = load_img(image_path)
    img = img_to_array(img)
    img = resize(img, INPUT_SHAPE)
    img = preprocess_input(img)
    return img

加载训练好的模型

vgg = VGG16(input_shape=INPUT_SHAPE, weights='imagenet', include_top=True)

vgg.summary()

Model: "vgg16"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_3 (InputLayer)        [(None, 224, 224, 3)]     0         
                                                                 
 block1_conv1 (Conv2D)       (None, 224, 224, 64)      1792      
                                                                 
 block1_conv2 (Conv2D)       (None, 224, 224, 64)      36928     
                                                                 
 block1_pool (MaxPooling2D)  (None, 112, 112, 64)      0         
                                                                 
 block2_conv1 (Conv2D)       (None, 112, 112, 128)     73856     
                                                                 
 block2_conv2 (Conv2D)       (None, 112, 112, 128)     147584    
                                                                 
 block2_pool (MaxPooling2D)  (None, 56, 56, 128)       0         
                                                                 
 block3_conv1 (Conv2D)       (None, 56, 56, 256)       295168    
                                                                 
 block3_conv2 (Conv2D)       (None, 56, 56, 256)       590080    
                                                                 
 block3_conv3 (Conv2D)       (None, 56, 56, 256)       590080    
                                                                 
 block3_pool (MaxPooling2D)  (None, 28, 28, 256)       0         
                                                                 
 block4_conv1 (Conv2D)       (None, 28, 28, 512)       1180160   
                                                                 
 block4_conv2 (Conv2D)       (None, 28, 28, 512)       2359808   
                                                                 
 block4_conv3 (Conv2D)       (None, 28, 28, 512)       2359808   
                                                                 
 block4_pool (MaxPooling2D)  (None, 14, 14, 512)       0         
                                                                 
 block5_conv1 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_conv2 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_conv3 (Conv2D)       (None, 14, 14, 512)       2359808   
                                                                 
 block5_pool (MaxPooling2D)  (None, 7, 7, 512)         0         
                                                                 
 flatten (Flatten)           (None, 25088)             0         
                                                                 
 fc1 (Dense)                 (None, 4096)              102764544 
                                                                 
 fc2 (Dense)                 (None, 4096)              16781312  
                                                                 
 predictions (Dense)         (None, 1000)              4097000   
                                                                 
=================================================================
Total params: 138,357,544
Trainable params: 138,357,544
Non-trainable params: 0
_________________________________________________________________

定义相关可视化函数

SELECT_LAYERS = [1, 3, 5, 7, 8, 10]

def predict(x):
  X = np.expand_dims(x, axis=0)
  y_preds = vgg.predict(X, verbose=False)
  idx = np.argmax(y_preds[0])
  return labels[idx]

def get_layers_output(img):
  label = predict(img)
  print('Image predicted to be: %s' % label)
  input_layer = vgg.layers[0]
  x = input_layer(np.expand_dims(img, axis=0))
  outputs = [x[0]]
  for layer in vgg.layers[1:]:
    x = layer(x)
    outputs.append(x[0])
  
  outputs = np.array(outputs)
  return outputs[SELECT_LAYERS]

def plot_layer_output(feature_maps, layer_idx, first=3):
  fig = plt.figure(figsize=(first * 3, 4))
  fig.suptitle('Layer %s' % layer_idx)
  for idx in range(first):
    sub_img = feature_maps[:, : , idx]
    ax = plt.subplot(1, first, idx + 1)
    ax.imshow(sub_img) 
    ax.set_title(feature_maps.shape)
    ax.axis('off')
  plt.show()

def plot_feature_maps(layers_output):
  for idx, output in zip(SELECT_LAYERS, layers_output):
    plot_layer_output(output, idx, first=10)
  
def plot_original_img(path):
  img = load_img(path)
  img = img_to_array(img)
  plt.figure(figsize=(6, 9))
  plt.title(path.split('.')[-2])
  plt.imshow(img / 255.0)
  plt.axis('off')
  plt.show()

在已知类别上测试VGG

def get_inside(path):
    img = load_and_preprocess(path)
    plot_original_img(path)
    # Get feature maps
    outputs = get_layers_output(img)
    plot_feature_maps(outputs)

get_inside('kite.jpg')

 

 

 

 

 

 

mbd.pub/o/GeBENHAGEN

擅长现代信号处理(改进小波分析系列，改进变分模态分解，改进经验小波变换，改进辛几何模态分解等等)，改进机器学习，改进深度学习，机械故障诊断，改进时间序列分析(金融信号，心电信号，振动信号等)