卷积神经网络

卷积神经网络

1.导包

import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

import sklearn

import tensorflow as tf

from tensorflow import keras



from sklearn.datasets import load_sample_image



china = load_sample_image("china.jpg") / 255

flower = load_sample_image("flower.jpg") / 255

plt.subplot(1,2,1)

plt.imshow(china)

plt.subplot(1,2,2)

plt.imshow(flower)



print("china.jpg的维度：",china.shape)

print("flower.jpg的维度：",flower.shape)



images = np.array([china,flower])

images_shape = images.shape



print("数据集的维度：",images_shape)

加载两个示例图像，并对它们进行了归一化处理。然后，显示了这两个图像，并打印了形状和数据集的形状。

1. 卷积层

u = 7 #卷积核边长

s = 1 #滑动步长

p = 5 #输出特征图数目

conv = keras.layers.Conv2D(filters= p, kernel_size=u, strides=s,padding="SAME", activation="relu", input_shape=images_shape)



image_after_conv = conv(images)



print("卷积后的张量大小：", image_after_conv.shape)

1. 汇聚层

3.1最大汇聚

pool_max = keras.layers.MaxPool2D(pool_size=2)

image_after_pool_max = pool_max(image_after_conv)

print("最大汇聚后的张量大小：",image_after_pool_max.shape)

3.2平均汇聚

pool_avg = keras.layers.AvgPool2D(pool_size=2)

image_after_pool_avg = pool_avg(image_after_conv)

print("平均汇聚后的张量大小：",image_after_pool_avg.shape)

3.3全局平均汇聚

pool_global_avg = keras.layers.GlobalAvgPool2D()

image_after_pool_global_avg = pool_global_avg(image_after_conv)

print("全局平均汇聚后的张量大小：",image_after_pool_global_avg.shape)

搭建卷积神经网络进行手写数字识别

path = "C:/Users/14919/OneDrive/Desktop/工坊/深度学习-工坊/MNIST/" #存放.csv的文件夹

train_Data = pd.read_csv( path+'mnist_train.csv', header = None) #训练数据

test_Data = pd.read_csv( path+'mnist_test.csv', header = None) #测试数据

X, y = train_Data.iloc[:,1:].values/255, train_Data.iloc[:,0].values #数据归一化



X_valid, X_train = X[:5000].reshape(5000,28,28) , X[5000:].reshape(55000,28,28) #验证集与训练集

y_valid, y_train = y[:5000], y[5000:]



X_test,y_test = test_Data.iloc[:,1:].values.reshape(10000,28,28)/255, test_Data.iloc[:,0].values #测试集



X_train = X_train[..., np.newaxis]

X_valid = X_valid[..., np.newaxis]

X_test = X_test[..., np.newaxis]



print(X_train.shape)

print(X_valid.shape)

print(X_test.shape)

处理MNIST数据集的常见流程，包括数据加载、归一化、分割成训练集、验证集和测试集，以及将二维图像数据转换为适合深度学习模型的格式。

# 搭建卷积神经网络

model_cnn_mnist = tf.keras.models.Sequential([

    tf.keras.layers.Conv2D(32, kernel_size=3, padding="same", activation="relu", input_shape=(28, 28, 1)),

    tf.keras.layers.Conv2D(64, kernel_size=3, padding="same", activation="relu"),

    tf.keras.layers.MaxPooling2D(pool_size=2),

    tf.keras.layers.Flatten(),

    tf.keras.layers.Dropout(0.25),

    tf.keras.layers.Dense(128, activation="relu"),

    tf.keras.layers.Dropout(0.5),

    tf.keras.layers.Dense(10, activation="softmax")

])

model_cnn_mnist.compile(loss="sparse_categorical_crossentropy", optimizer="nadam", metrics=["accuracy"])

model_cnn_mnist.fit(X_train, y_train, epochs=10, validation_data=(X_valid, y_valid))

model_cnn_mnist.evaluate(X_test, y_test, batch_size=1)

class ResidualUnit(keras.layers.Layer):

    def __init__(self, filters, strides=1, activation="relu"):

        super().__init__()

        self.activation = keras.activations.get(activation)

        

        self.main_layers = [

            keras.layers.Conv2D(filters, 3, strides=strides, padding = "SAME", use_bias = False),

            keras.layers.BatchNormalization(),

            self.activation,

            keras.layers.Conv2D(filters, 3, strides=1, padding = "SAME", use_bias = False),

            keras.layers.BatchNormalization()]

        

        # 当滑动步长s = 1时，残差连接直接将输入与卷积结果相加，skip_layers为空，即实线连接

        self.skip_layers = []

        

        # 当滑动步长s = 2时，残差连接无法直接将输入与卷积结果相加，需要对输入进行卷积处理，即虚线连接

        if strides > 1:

            self.skip_layers = [

                keras.layers.Conv2D(filters, 1, strides=strides, padding = "SAME", use_bias = False),

                keras.layers.BatchNormalization()]



    def call(self, inputs):

        Z = inputs

        for layer in self.main_layers:

            Z = layer(Z)

        skip_Z = inputs

        for layer in self.skip_layers:

            skip_Z = layer(skip_Z)

        return self.activation(Z + skip_Z)

搭建完整的ResNet-34神经网络

model = keras.models.Sequential()



model.add(keras.layers.Conv2D(64, 7, strides=2, padding = "SAME", use_bias = False))

model.add(keras.layers.BatchNormalization())

model.add(keras.layers.Activation("relu"))

model.add(keras.layers.MaxPool2D(pool_size=3, strides=2, padding="SAME"))



prev_filters = 64

for filters in [64] * 3 + [128] * 4 + [256] * 6 + [512] * 3:

    strides = 1 if filters == prev_filters else 2    #在每次特征图数目扩展时，设置滑动步长为2

    model.add(ResidualUnit(filters, strides=strides))

    prev_filters = filters

    

model.add(keras.layers.GlobalAvgPool2D())

model.add(keras.layers.Flatten())

model.add(keras.layers.Dense(10, activation="softmax"))



model_cnn_mnist.compile(loss="sparse_categorical_crossentropy", optimizer="nadam", metrics=["accuracy"])

model_cnn_mnist.fit(X_train, y_train, epochs=10, validation_data=(X_valid, y_valid))

model_cnn_mnist.evaluate(X_test, y_test, batch_size=1)