1.参考博文:
非常感谢这里的这位博主,让我对特征可视化有了进一步的了解:
https://blog.csdn.net/Forrest97/article/details/105895087
2.实验效果:
3.结果分析:
通过我画的一些图可以看出,随着卷积层数的加深,特征图变得越来越专业。像这张图片,图片中狗的眼睛和鼻子被提取的比较多;而浅层更多的是提取图像的边缘和轮廓。因此,随着网络的深入,图像的特征越来越专业化,同时也越来越难以理解。
4.代码实现:
import os
import matplotlib
import tensorflow
import numpy as np
import matplotlib.pyplot as plt
from keras.preprocessing import image
from tensorflow.keras.models import Model
from tensorflow.keras.applications.vgg16 import preprocess_input,decode_predictions
#加载模型
def load_model():
model_vgg16=tensorflow.keras.applications.vgg16.VGG16(weights='imagenet')
return model_vgg16
#查询卷积层的名称
def block_layers_16():
model_vgg16=load_model()
layer_outputs = [layer.output for layer in model_vgg16.layers[:20]] #前16层输出
model = Model(inputs=model_vgg16.input, outputs=layer_outputs) #构建能够输出前16层的模型
return model
def featureVisualiztion(img_path):
block_name=[layer.name for layer in load_model().layers[1:5]]
model_vgg16 = block_layers_16()
#将图像缩放到固定大小
img=image.load_img(img_path,target_size=(224,224))
#将图片转换为向量
img=image.img_to_array(img)
#对其维度进行扩充
img=np.expand_dims(img,axis=0)
#对输入到网络的图像进行处理
output_img=preprocess_input(img)
#预测图像
features=model_vgg16.predict(output_img)
feature1_shape=np.shape(features[0])
print(feature1_shape)
# model_vgg16.summary()
#显示前五层的特征图
for step,feature in enumerate(features[1:5]):
#特征图的通道数//16=画图的行数
rows=feature.shape[-1]//16
subplots=np.zeros((int(feature.shape[1]*rows),int(16*feature.shape[1])))
for row in range(rows):
for col in range(16):
#每一个通道的特征图
feature_image=feature[0,:,:,row*16+col]
subplots[row*feature.shape[1]:(row+1)*feature.shape[1],col*feature.shape[1]:(col+1)*feature.shape[1]]=feature_image
scale=1.0/feature.shape[1]
plt.figure(figsize=(scale*subplots.shape[1],scale*subplots.shape[0]))
plt.title(block_name[step])
plt.grid(False)
plt.imshow(subplots,aspect='auto',cmap='viridis')
plt.show()
if __name__ == '__main__':
print('Pycharm')
featureVisualiztion('images/dog.jpg')
文章出处登录后可见!
已经登录?立即刷新