SAT1.4.1.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "gpus = tf.config.list_physical_devices(device_type='GPU')\n",
    "for gpu in gpus:\n",
    "    tf.config.experimental.set_memory_growth(device=gpu, enable=True)   \n",
    "#GPU内存管理"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dir = 'sat2/train'\n",
    "test_dir = 'sat2/val'\n",
    "\n",
    "im_size = 224\n",
    "batch_size = 32"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_images = ImageDataGenerator(rescale = 1/255,horizontal_flip=True)\n",
    "test_images = ImageDataGenerator(rescale = 1/255)\n",
    "#归一化"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 3150 images belonging to 5 classes.\n"
     ]
    }
   ],
   "source": [
    "train_gen = train_images.flow_from_directory(directory=train_dir,\n",
    "                                             batch_size=batch_size,\n",
    "                                             shuffle=True,\n",
    "                                             target_size=(im_size, im_size),\n",
    "                                             class_mode='categorical')\n",
    "#按路径载入图片，批处理大小，随机，尺寸，读热编码"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 350 images belonging to 5 classes.\n"
     ]
    }
   ],
   "source": [
    "val_gen = test_images.flow_from_directory(directory=test_dir,\n",
    "                                          batch_size=batch_size,\n",
    "                                          shuffle=False,\n",
    "                                          target_size=(im_size, im_size),\n",
    "                                          class_mode='categorical')\n",
    "#按路径载入图片，批处理大小，随机，尺寸，读热编码"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def Inception (ch1x1, ch3x3red, ch3x3, ch5x5red, ch5x5, pool_proj, input_):\n",
    "    inputs = tf.keras.layers.Input(shape=input_.shape[1:])\n",
    "    x1 = tf.keras.layers.Conv2D(ch1x1, kernel_size=1, activation=\"relu\")(inputs)\n",
    "    \n",
    "    x21 = tf.keras.layers.Conv2D(ch3x3red, kernel_size=1, activation=\"relu\")(inputs)\n",
    "    x22 =  tf.keras.layers.Conv2D(ch3x3, kernel_size=3, padding=\"same\", activation=\"relu\")(x21)\n",
    "    \n",
    "    x31 = tf.keras.layers.Conv2D(ch5x5red, kernel_size=1, activation=\"relu\")(inputs)\n",
    "    x32 = tf.keras.layers.Conv2D(ch5x5, kernel_size=5, padding=\"same\", activation=\"relu\")(x31)\n",
    "    \n",
    "    x41 = tf.keras.layers.MaxPool2D(pool_size=3, strides=1, padding=\"same\")(inputs) \n",
    "    x42 = tf.keras.layers.Conv2D(pool_proj, kernel_size=1, activation=\"relu\")(x41)\n",
    "    outputs = tf.concat((x1, x22, x32, x42),axis=-1)\n",
    "    \n",
    "    return tf.keras.Model(inputs=inputs, outputs=outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def InceptionAux (num_classes, input_):\n",
    "    inputs = tf.keras.layers.Input(shape=input_.shape[1:])\n",
    "    x = tf.keras.layers.AvgPool2D(pool_size=5, strides=3)(inputs)\n",
    "    x = tf.keras.layers.Conv2D(128, kernel_size=1, activation=\"relu\")(x)\n",
    "    \n",
    "    x = tf.keras.layers.Flatten()(x)\n",
    "    x = tf.keras.layers.Dropout(rate=0.7)(x)\n",
    "    x = tf.keras.layers.Dense(1024, activation=\"relu\")(x)\n",
    "    x = tf.keras.layers.Dropout(rate=0.7)(x)\n",
    "    x = tf.keras.layers.Dense(num_classes)(x)\n",
    "    \n",
    "    return tf.keras.Model(inputs=inputs, outputs=x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def GoogLeNet():\n",
    "    input_image = tf.keras.layers.Input(shape=(224, 224, 3))\n",
    "    x = tf.keras.layers.Conv2D(64, kernel_size=7, strides=2, padding=\"same\", activation=\"relu\")(input_image)\n",
    "    x = tf.keras.layers.MaxPool2D(pool_size=3, strides=2, padding=\"same\")(x)\n",
    "    x = tf.keras.layers.Conv2D(64, kernel_size=1, activation=\"relu\")(x)\n",
    "    x = tf.keras.layers.Conv2D(192, kernel_size=3, padding=\"same\", activation=\"relu\")(x)\n",
    "    x = tf.keras.layers.MaxPool2D(pool_size=3, strides=2, padding=\"same\")(x)\n",
    "    \n",
    "    #3a\n",
    "    x = Inception(64, 96, 128, 16, 32, 32, x)(x)\n",
    "    #3b\n",
    "    x = Inception(128, 128, 192, 32, 96, 64, x)(x)\n",
    "\n",
    "    x = tf.keras.layers.MaxPool2D(pool_size=3, strides=2, padding=\"same\")(x)\n",
    "    #4a\n",
    "    x = Inception(192, 96, 208, 16, 48, 64, x)(x)\n",
    "    #aux1\n",
    "    aux11 = InceptionAux(5,x)(x)\n",
    "    aux1 = tf.keras.layers.Softmax(name=\"aux_1\")(aux11)\n",
    "    #4b\n",
    "    x = Inception(160, 112, 224, 24, 64, 64, x)(x)\n",
    "    #4c\n",
    "    x = Inception(128, 128, 256, 24, 64, 64, x)(x)\n",
    "    #4d\n",
    "    x = Inception(112, 144, 288, 32, 64, 64, x)(x)\n",
    "    #aux2\n",
    "    aux22 = InceptionAux(5,x)(x)\n",
    "    aux2 = tf.keras.layers.Softmax(name=\"aux_2\")(aux22)\n",
    "    #4e\n",
    "    x = Inception(256, 160, 320, 32, 128, 128, x)(x)\n",
    "    x = tf.keras.layers.MaxPool2D(pool_size=3, strides=2, padding=\"same\")(x)\n",
    "    #5a\n",
    "    x = Inception(256, 160, 320, 32, 128, 128, x)(x)\n",
    "    #5b\n",
    "    x = Inception(384, 192, 384, 48, 128, 128, x)(x)\n",
    "    x = tf.keras.layers.AvgPool2D(pool_size=7, strides=1)(x)\n",
    "\n",
    "    x = tf.keras.layers.Flatten()(x)\n",
    "\n",
    "    x = tf.keras.layers.Dropout(rate=0.4)(x)\n",
    "    x = tf.keras.layers.Dense(5)(x)\n",
    "    #aux3\n",
    "    aux3 = tf.keras.layers.Softmax(name=\"aux_3\")(x)\n",
    "\n",
    "    model = tf.keras.models.Model(inputs=input_image, outputs=[aux1, aux2, aux3])\n",
    "#     model = tf.keras.models.Model(inputs=input_image, outputs=aux3)\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = GoogLeNet()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_11\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            [(None, 224, 224, 3) 0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d (Conv2D)                 (None, 112, 112, 64) 9472        input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D)    (None, 56, 56, 64)   0           conv2d[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_1 (Conv2D)               (None, 56, 56, 64)   4160        max_pooling2d[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_2 (Conv2D)               (None, 56, 56, 192)  110784      conv2d_1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2D)  (None, 28, 28, 192)  0           conv2d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "model (Functional)              (None, 28, 28, 256)  163696      max_pooling2d_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "model_1 (Functional)            (None, 28, 28, 480)  388736      model[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2D)  (None, 14, 14, 480)  0           model_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "model_2 (Functional)            (None, 14, 14, 512)  376176      max_pooling2d_4[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "model_4 (Functional)            (None, 14, 14, 512)  449160      model_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "model_5 (Functional)            (None, 14, 14, 512)  510104      model_4[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "model_6 (Functional)            (None, 14, 14, 528)  605376      model_5[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "model_8 (Functional)            (None, 14, 14, 832)  868352      model_6[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_10 (MaxPooling2D) (None, 7, 7, 832)    0           model_8[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "model_9 (Functional)            (None, 7, 7, 832)    1043456     max_pooling2d_10[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "model_10 (Functional)           (None, 7, 7, 1024)   1444080     model_9[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "average_pooling2d_2 (AveragePoo (None, 1, 1, 1024)   0           model_10[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "flatten_2 (Flatten)             (None, 1024)         0           average_pooling2d_2[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "dropout_4 (Dropout)             (None, 1024)         0           flatten_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "model_3 (Functional)            (None, 5)            2168965     model_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "model_7 (Functional)            (None, 5)            2171013     model_6[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dense_4 (Dense)                 (None, 5)            5125        dropout_4[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "aux_1 (Softmax)                 (None, 5)            0           model_3[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "aux_2 (Softmax)                 (None, 5)            0           model_7[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "aux_3 (Softmax)                 (None, 5)            0           dense_4[0][0]                    \n",
      "==================================================================================================\n",
      "Total params: 10,318,655\n",
      "Trainable params: 10,318,655\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.0005),\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=['acc'])\n",
    "#学习率设置"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "99/99 [==============================] - 12s 70ms/step - loss: 4.7388 - aux_1_loss: 1.5717 - aux_2_loss: 1.5815 - aux_3_loss: 1.5857 - aux_1_acc: 0.2274 - aux_2_acc: 0.2195 - aux_3_acc: 0.2041 - val_loss: 3.7580 - val_aux_1_loss: 1.2492 - val_aux_2_loss: 1.2531 - val_aux_3_loss: 1.2557 - val_aux_1_acc: 0.4657 - val_aux_2_acc: 0.4457 - val_aux_3_acc: 0.4229\n",
      "Epoch 2/20\n",
      "99/99 [==============================] - 5s 51ms/step - loss: 3.8507 - aux_1_loss: 1.2815 - aux_2_loss: 1.2896 - aux_3_loss: 1.2796 - aux_1_acc: 0.3853 - aux_2_acc: 0.3747 - aux_3_acc: 0.3706 - val_loss: 3.7841 - val_aux_1_loss: 1.2400 - val_aux_2_loss: 1.2657 - val_aux_3_loss: 1.2784 - val_aux_1_acc: 0.4257 - val_aux_2_acc: 0.4086 - val_aux_3_acc: 0.3886\n",
      "Epoch 3/20\n",
      "99/99 [==============================] - 5s 51ms/step - loss: 3.7424 - aux_1_loss: 1.2328 - aux_2_loss: 1.2562 - aux_3_loss: 1.2534 - aux_1_acc: 0.4059 - aux_2_acc: 0.3904 - aux_3_acc: 0.3893 - val_loss: 3.7003 - val_aux_1_loss: 1.2147 - val_aux_2_loss: 1.2281 - val_aux_3_loss: 1.2576 - val_aux_1_acc: 0.4400 - val_aux_2_acc: 0.4000 - val_aux_3_acc: 0.3800\n",
      "Epoch 4/20\n",
      "99/99 [==============================] - 5s 50ms/step - loss: 3.7260 - aux_1_loss: 1.2137 - aux_2_loss: 1.2452 - aux_3_loss: 1.2672 - aux_1_acc: 0.4279 - aux_2_acc: 0.3974 - aux_3_acc: 0.3682 - val_loss: 3.4652 - val_aux_1_loss: 1.1266 - val_aux_2_loss: 1.1544 - val_aux_3_loss: 1.1842 - val_aux_1_acc: 0.4886 - val_aux_2_acc: 0.4429 - val_aux_3_acc: 0.3971\n",
      "Epoch 5/20\n",
      "99/99 [==============================] - 5s 51ms/step - loss: 3.5130 - aux_1_loss: 1.1445 - aux_2_loss: 1.1705 - aux_3_loss: 1.1980 - aux_1_acc: 0.4807 - aux_2_acc: 0.4516 - aux_3_acc: 0.4031 - val_loss: 3.4835 - val_aux_1_loss: 1.1264 - val_aux_2_loss: 1.1554 - val_aux_3_loss: 1.2017 - val_aux_1_acc: 0.4514 - val_aux_2_acc: 0.4486 - val_aux_3_acc: 0.4029\n",
      "Epoch 6/20\n",
      "99/99 [==============================] - 5s 50ms/step - loss: 3.3719 - aux_1_loss: 1.0419 - aux_2_loss: 1.1464 - aux_3_loss: 1.1837 - aux_1_acc: 0.5669 - aux_2_acc: 0.4433 - aux_3_acc: 0.4167 - val_loss: 2.6125 - val_aux_1_loss: 0.7977 - val_aux_2_loss: 0.8646 - val_aux_3_loss: 0.9502 - val_aux_1_acc: 0.6771 - val_aux_2_acc: 0.6657 - val_aux_3_acc: 0.5914\n",
      "Epoch 7/20\n",
      "99/99 [==============================] - 5s 51ms/step - loss: 2.8071 - aux_1_loss: 0.8995 - aux_2_loss: 0.9291 - aux_3_loss: 0.9785 - aux_1_acc: 0.6404 - aux_2_acc: 0.6325 - aux_3_acc: 0.6056 - val_loss: 2.5787 - val_aux_1_loss: 0.7841 - val_aux_2_loss: 0.8424 - val_aux_3_loss: 0.9522 - val_aux_1_acc: 0.7086 - val_aux_2_acc: 0.6714 - val_aux_3_acc: 0.6371\n",
      "Epoch 8/20\n",
      "99/99 [==============================] - 5s 51ms/step - loss: 2.5757 - aux_1_loss: 0.8280 - aux_2_loss: 0.8560 - aux_3_loss: 0.8918 - aux_1_acc: 0.6735 - aux_2_acc: 0.6932 - aux_3_acc: 0.6613 - val_loss: 1.8965 - val_aux_1_loss: 0.6125 - val_aux_2_loss: 0.6196 - val_aux_3_loss: 0.6644 - val_aux_1_acc: 0.7543 - val_aux_2_acc: 0.7686 - val_aux_3_acc: 0.7714\n",
      "Epoch 9/20\n",
      "99/99 [==============================] - 5s 52ms/step - loss: 2.0539 - aux_1_loss: 0.7100 - aux_2_loss: 0.6643 - aux_3_loss: 0.6796 - aux_1_acc: 0.7298 - aux_2_acc: 0.7674 - aux_3_acc: 0.7555 - val_loss: 2.8638 - val_aux_1_loss: 0.9851 - val_aux_2_loss: 0.9340 - val_aux_3_loss: 0.9447 - val_aux_1_acc: 0.6257 - val_aux_2_acc: 0.6371 - val_aux_3_acc: 0.6457\n",
      "Epoch 10/20\n",
      "99/99 [==============================] - 5s 52ms/step - loss: 1.9460 - aux_1_loss: 0.6731 - aux_2_loss: 0.6420 - aux_3_loss: 0.6309 - aux_1_acc: 0.7422 - aux_2_acc: 0.7629 - aux_3_acc: 0.7706 - val_loss: 1.8713 - val_aux_1_loss: 0.5799 - val_aux_2_loss: 0.6134 - val_aux_3_loss: 0.6780 - val_aux_1_acc: 0.7714 - val_aux_2_acc: 0.7914 - val_aux_3_acc: 0.7543\n",
      "Epoch 11/20\n",
      "99/99 [==============================] - 5s 51ms/step - loss: 1.8272 - aux_1_loss: 0.6394 - aux_2_loss: 0.5884 - aux_3_loss: 0.5994 - aux_1_acc: 0.7606 - aux_2_acc: 0.7940 - aux_3_acc: 0.7837 - val_loss: 1.8349 - val_aux_1_loss: 0.5677 - val_aux_2_loss: 0.6028 - val_aux_3_loss: 0.6644 - val_aux_1_acc: 0.7971 - val_aux_2_acc: 0.8029 - val_aux_3_acc: 0.7743\n",
      "Epoch 12/20\n",
      "99/99 [==============================] - 5s 49ms/step - loss: 1.5562 - aux_1_loss: 0.5528 - aux_2_loss: 0.5032 - aux_3_loss: 0.5001 - aux_1_acc: 0.7942 - aux_2_acc: 0.8182 - aux_3_acc: 0.8198 - val_loss: 1.5192 - val_aux_1_loss: 0.4648 - val_aux_2_loss: 0.5037 - val_aux_3_loss: 0.5507 - val_aux_1_acc: 0.8486 - val_aux_2_acc: 0.8429 - val_aux_3_acc: 0.8171\n",
      "Epoch 13/20\n",
      "99/99 [==============================] - 5s 48ms/step - loss: 1.3666 - aux_1_loss: 0.5079 - aux_2_loss: 0.4326 - aux_3_loss: 0.4262 - aux_1_acc: 0.8178 - aux_2_acc: 0.8561 - aux_3_acc: 0.8502 - val_loss: 1.3419 - val_aux_1_loss: 0.4321 - val_aux_2_loss: 0.4323 - val_aux_3_loss: 0.4775 - val_aux_1_acc: 0.8429 - val_aux_2_acc: 0.8600 - val_aux_3_acc: 0.8486\n",
      "Epoch 14/20\n",
      "99/99 [==============================] - 5s 48ms/step - loss: 1.2212 - aux_1_loss: 0.4495 - aux_2_loss: 0.3910 - aux_3_loss: 0.3807 - aux_1_acc: 0.8412 - aux_2_acc: 0.8659 - aux_3_acc: 0.8668 - val_loss: 1.1511 - val_aux_1_loss: 0.4001 - val_aux_2_loss: 0.3715 - val_aux_3_loss: 0.3795 - val_aux_1_acc: 0.8514 - val_aux_2_acc: 0.8743 - val_aux_3_acc: 0.8714\n",
      "Epoch 15/20\n",
      "99/99 [==============================] - 5s 50ms/step - loss: 1.1597 - aux_1_loss: 0.4350 - aux_2_loss: 0.3712 - aux_3_loss: 0.3535 - aux_1_acc: 0.8514 - aux_2_acc: 0.8732 - aux_3_acc: 0.8726 - val_loss: 1.1095 - val_aux_1_loss: 0.3936 - val_aux_2_loss: 0.3435 - val_aux_3_loss: 0.3724 - val_aux_1_acc: 0.8571 - val_aux_2_acc: 0.8914 - val_aux_3_acc: 0.8800\n",
      "Epoch 16/20\n",
      "99/99 [==============================] - 5s 48ms/step - loss: 1.0379 - aux_1_loss: 0.3860 - aux_2_loss: 0.3345 - aux_3_loss: 0.3173 - aux_1_acc: 0.8620 - aux_2_acc: 0.8882 - aux_3_acc: 0.8933 - val_loss: 1.0656 - val_aux_1_loss: 0.3731 - val_aux_2_loss: 0.3421 - val_aux_3_loss: 0.3504 - val_aux_1_acc: 0.8743 - val_aux_2_acc: 0.8686 - val_aux_3_acc: 0.8829\n",
      "Epoch 17/20\n",
      "99/99 [==============================] - 5s 48ms/step - loss: 1.0243 - aux_1_loss: 0.3854 - aux_2_loss: 0.3248 - aux_3_loss: 0.3142 - aux_1_acc: 0.8664 - aux_2_acc: 0.8973 - aux_3_acc: 0.8919 - val_loss: 1.1732 - val_aux_1_loss: 0.4353 - val_aux_2_loss: 0.3675 - val_aux_3_loss: 0.3704 - val_aux_1_acc: 0.8514 - val_aux_2_acc: 0.8743 - val_aux_3_acc: 0.8800\n",
      "Epoch 18/20\n",
      "99/99 [==============================] - 5s 48ms/step - loss: 1.0823 - aux_1_loss: 0.4202 - aux_2_loss: 0.3408 - aux_3_loss: 0.3213 - aux_1_acc: 0.8542 - aux_2_acc: 0.8909 - aux_3_acc: 0.8913 - val_loss: 0.9446 - val_aux_1_loss: 0.3305 - val_aux_2_loss: 0.2949 - val_aux_3_loss: 0.3192 - val_aux_1_acc: 0.8943 - val_aux_2_acc: 0.8857 - val_aux_3_acc: 0.8914\n",
      "Epoch 19/20\n",
      "99/99 [==============================] - 5s 48ms/step - loss: 0.9642 - aux_1_loss: 0.3791 - aux_2_loss: 0.2989 - aux_3_loss: 0.2862 - aux_1_acc: 0.8724 - aux_2_acc: 0.9117 - aux_3_acc: 0.9057 - val_loss: 0.8487 - val_aux_1_loss: 0.3175 - val_aux_2_loss: 0.2604 - val_aux_3_loss: 0.2708 - val_aux_1_acc: 0.8943 - val_aux_2_acc: 0.9229 - val_aux_3_acc: 0.9086\n",
      "Epoch 20/20\n",
      "99/99 [==============================] - 5s 49ms/step - loss: 0.8453 - aux_1_loss: 0.3251 - aux_2_loss: 0.2735 - aux_3_loss: 0.2467 - aux_1_acc: 0.8885 - aux_2_acc: 0.9086 - aux_3_acc: 0.9095 - val_loss: 1.0314 - val_aux_1_loss: 0.3490 - val_aux_2_loss: 0.3243 - val_aux_3_loss: 0.3581 - val_aux_1_acc: 0.8971 - val_aux_2_acc: 0.9000 - val_aux_3_acc: 0.8857\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(train_gen,epochs=20,validation_data=val_gen)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7faf70252280>]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.epoch,history.history.get('aux_3_acc'))\n",
    "plt.plot(history.epoch,history.history.get('val_aux_3_acc'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "11/11 [==============================] - 0s 41ms/step - loss: 1.0314 - aux_1_loss: 0.3490 - aux_2_loss: 0.3243 - aux_3_loss: 0.3581 - aux_1_acc: 0.8971 - aux_2_acc: 0.9000 - aux_3_acc: 0.8857\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1.0314477682113647,\n",
       " 0.34904995560646057,\n",
       " 0.32426247000694275,\n",
       " 0.35813525319099426,\n",
       " 0.8971428275108337,\n",
       " 0.8999999761581421,\n",
       " 0.8857142925262451]"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.evaluate(val_gen)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 保存模型\n",
    "model.save('sat75.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}