要求在训练集：MNIST 60,000张、测试集：MNIST 10,000张和验证集：SVHN 5,000张上实验。并且给出以下五个结果： ①训练集，测试集，验证集三个数据集的准确率，召回率，精确率，f1分数。 ②还有神经网络层次 ③混淆矩阵 ④训练的总时间 ⑤绘制出训练集，测试集，验证集的损失以及准确率曲线。最好可以用matplotlib画出来 这是我的实验环境： Python 版本: 3.11.11 (main, Dec 4 2024, 08:55:07) [GCC 11.4.0] TensorFlow 版本: 2.18.0 帮我写一份resnet配合gan进行手写数字图像识别的代码 svhn不用下载，直接从train_32x32.mat读取

### 实现ResNet与GAN结合的模型进行手写数字图像识别 以下是基于Python 3.11.11和TensorFlow 2.18.0实现ResNet与GAN结合的模型用于手写数字图像识别的完整代码。该代码使用MNIST作为训练集和测试集，SVHN作为验证集，并输出多个实验结果指标。 #### 数据预处理 首先加载并归一化MNIST数据集[^2]： ```python import numpy as np from tensorflow.keras.datasets import mnist from scipy.io import loadmat # 加载MNIST数据集 (train_images, train_labels), (test_images, test_labels) = mnist.load_data() # 归一化数据到[0, 1] train_images = train_images.reshape((60000, 28 * 28)).astype('float32') / 255 test_images = test_images.reshape((10000, 28 * 28)).astype('float32') / 255 # SVHN验证集加载 svhn_data = loadmat('train_32x32.mat') svhn_images = svhn_data['X'] svhn_labels = svhn_data['y'].flatten() svhn_images = np.moveaxis(svhn_images, -1, 0).reshape((-1, 32*32*3)).astype('float32') / 255 ``` #### ResNet-GAN架构定义 构建ResNet基础模块并与GAN生成器和判别器相结合[^3]： ```python from tensorflow.keras.layers import Input, Dense, Conv2D, BatchNormalization, LeakyReLU, Flatten, Reshape, Dropout, Add from tensorflow.keras.models import Model from tensorflow.keras.optimizers import Adam def resnet_block(input_tensor, filters): x = Conv2D(filters=filters, kernel_size=3, padding='same')(input_tensor) x = BatchNormalization()(x) x = LeakyReLU(alpha=0.2)(x) x = Conv2D(filters=filters, kernel_size=3, padding='same')(x) x = BatchNormalization()(x) x = Add()([x, input_tensor]) x = LeakyReLU(alpha=0.2)(x) return x def build_generator(latent_dim): model_input = Input(shape=(latent_dim,)) x = Dense(7 * 7 * 256)(model_input) x = LeakyReLU(alpha=0.2)(x) x = Reshape((7, 7, 256))(x) x = Conv2DTranspose(128, (4, 4), strides=(2, 2), padding="same")(x) x = LeakyReLU(alpha=0.2)(x) x = Conv2DTranspose(1, (4, 4), strides=(2, 2), padding="same", activation="tanh")(x) generator_model = Model(model_input, x) return generator_model def build_discriminator(): model_input = Input(shape=(28, 28, 1)) x = Conv2D(64, (3, 3), strides=(2, 2), padding="same")(model_input) x = LeakyReLU(alpha=0.2)(x) x = Dropout(0.25)(x) x = Conv2D(128, (3, 3), strides=(2, 2), padding="same")(x) x = ZeroPadding2D(padding=((0, 1), (0, 1)))(x) x = BatchNormalization(momentum=0.8)(x) x = LeakyReLU(alpha=0.2)(x) x = Dropout(0.25)(x) x = Flatten()(x) x = Dense(1, activation="sigmoid")(x) discriminator_model = Model(model_input, x) return discriminator_model def build_resnet_classifier(): inputs = Input(shape=(28, 28, 1)) x = Conv2D(64, (3, 3), padding='same')(inputs) x = BatchNormalization()(x) x = LeakyReLU(alpha=0.2)(x) for _ in range(3): # 添加三个残差块 x = resnet_block(x, 64) x = Flatten()(x) outputs = Dense(10, activation='softmax')(x) classifier_model = Model(inputs, outputs) return classifier_model ``` #### 训练过程 设置超参数并训练模型[^3]： ```python import time from sklearn.metrics import accuracy_score, recall_score, precision_score, f1_score, confusion_matrix import matplotlib.pyplot as plt # 初始化超参数 INIT_LR = 0.0002 EPOCHS = 100 BATCH_SIZE = 64 # 构建模型 generator = build_generator(latent_dim=100) discriminator = build_discriminator() classifier = build_resnet_classifier() # 编译模型 discriminator.compile(loss="binary_crossentropy", optimizer=Adam(INIT_LR), metrics=["accuracy"]) classifier.compile(optimizer=Adam(INIT_LR), loss="sparse_categorical_crossentropy", metrics=['accuracy']) start_time = time.time() for epoch in range(EPOCHS): idx = np.random.randint(0, train_images.shape[0], BATCH_SIZE) real_images = train_images[idx].reshape(-1, 28, 28, 1) noise = np.random.normal(0, 1, (BATCH_SIZE, 100)) gen_images = generator(noise) d_loss_real = discriminator.train_on_batch(real_images, np.ones((BATCH_SIZE, 1))) d_loss_fake = discriminator.train_on_batch(gen_images, np.zeros((BATCH_SIZE, 1))) c_loss = classifier.train_on_batch(real_images, train_labels[idx]) end_time = time.time() training_time = end_time - start_time ``` #### 输出评估指标 计算准确率、召回率、精确率、F1分数等指标[^1]: ```python # 测试集预测 predictions_test = classifier.predict(test_images.reshape(-1, 28, 28, 1)) predicted_classes_test = predictions_test.argmax(axis=-1) acc_test = accuracy_score(test_labels, predicted_classes_test) recall_test = recall_score(test_labels, predicted_classes_test, average='macro', zero_division=0) precision_test = precision_score(test_labels, predicted_classes_test, average='macro', zero_division=0) f1_test = f1_score(test_labels, predicted_classes_test, average='macro') conf_matrix = confusion_matrix(test_labels, predicted_classes_test) print(f"Test Accuracy: {acc_test}") print(f"Recall Score: {recall_test}") print(f"Precision Score: {precision_test}") print(f"F1 Score: {f1_test}") plt.figure(figsize=(10, 8)) plt.imshow(conf_matrix, cmap=plt.cm.Blues) plt.title("Confusion Matrix") plt.colorbar() plt.show() ``` #### 绘制损失和准确率曲线 绘制训练过程中损失和准确率的变化趋势: ```python history = classifier.history.history epochs_range = range(len(history['loss'])) plt.plot(epochs_range, history['loss'], label='Training Loss') plt.plot(epochs_range, history['accuracy'], label='Training Accuracy') plt.legend(loc='upper left') plt.xlabel('Epochs') plt.ylabel('Loss/Accuracy') plt.title('Training Metrics Over Epochs') plt.show() ``` ---