T3 天气识别
- 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
- 🍖 原作者：K同学啊 | 接辅导、项目定制
- 🚀 文章来源：K同学的学习圈子
本文采用Tensorflow的框架，进行天气图像的检测识别，数据由K同学提供。电脑配置
Current time: 2023-11-10 13:47:34.951944
Python version: 3.6.13 |Anaconda, Inc.| (default, Jun 4 2021, 14:25:59)
[GCC 7.5.0]
Tensorflow version 2.6.2
Matplotlib version 3.3.4
Device: CPU
一 前期准备
1. 设置GPU
在anaconda prompt中新建名称为tensorflow_cpu的环境，并安装tensorflow，matplotlib的包。同时脚本代码为ipy文件，需要安装ipykernel。
conda create -n tensorflow_cpu python=3.6
conda activate tensorflow_cpu
pip install tensorflow
pip install matplotlib
pip install ipykernel
输出使用的电脑配置。
from datetime import datetime
import sys
import matplotlib
print("Current time:", datetime.today())
print("Python version:", sys.version)
print("Tensorflow version", tf.__version__)
print("Matplotlib version", matplotlib.__version__)
gpus = tf.config.list_physical_devices("GPU")
if gups:
gpu0 = gpus[0]
tf.config.experimental.set_memory_growth(gpu0, True)
tf.config.set_visible_devices([gpu0],"GPU")
print("Device: GPU")
else:
print("Device: CPU")
2. 导入数据并查看数据
导入数据并展示示例图片。
import os,PIL,pathlib
import matplotlib.pyplot as plt
import numpy as np
from tensorflow import keras
from tensorflow.keras import layers,models
data_dir = "./weather_photos/"
data_dir = pathlib.Path(data_dir)
image_count = len(list(data_dir.glob('*/*.jpg')))
print("图片总数为：",image_count)
#图片总数为： 1125
roses = list(data_dir.glob('sunrise/*.jpg'))
PIL.Image.open(str(roses[0]))
二、数据预处理
1. 加载数据
划分训练集和测试集，其中训练集数据为900张，测试集数据为225张。
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
data_dir,
validation_split=0.2,
subset="training",
seed=123,
image_size=(img_height, img_width),
batch_size=batch_size)
val_ds = tf.keras.preprocessing.image_dataset_from_directory(
data_dir,
validation_split=0.2,
subset="validation",
seed=123,
image_size=(img_height, img_width),
batch_size=batch_size)
2. 可视化
class_names = train_ds.class_names
print(class_names)
plt.figure(figsize=(20, 10))
for images, labels in train_ds.take(1):
for i in range(20):
ax = plt.subplot(5, 10, i + 1)
plt.imshow(images[i].numpy().astype("uint8"))
plt.title(class_names[labels[i]])
plt.axis("off")
3. 配置数据集
AUTOTUNE = tf.data.AUTOTUNE
train_ds = train_ds.cache().shuffle(1000).prefetch(buffer_size=AUTOTUNE)
val_ds = val_ds.cache().prefetch(buffer_size=AUTOTUNE)
二 构建网络模型
1. 搭建模型
构建三层卷积、两层全连接的神经网络。
num_classes = 4
"""
关于卷积核的计算不懂的可以参考文章：https://blog.youkuaiyun.com/qq_38251616/article/details/114278995
layers.Dropout(0.4) 作用是防止过拟合，提高模型的泛化能力。
在上一篇文章花朵识别中，训练准确率与验证准确率相差巨大就是由于模型过拟合导致的
关于Dropout层的更多介绍可以参考文章：https://mtyjkh.blog.youkuaiyun.com/article/details/115826689
"""
model = models.Sequential([
layers.experimental.preprocessing.Rescaling(1./255, input_shape=(img_height, img_width, 3)),
layers.Conv2D(16, (3, 3), activation='relu', input_shape=(img_height, img_width, 3)), # 卷积层1，卷积核3*3
layers.AveragePooling2D((2, 2)), # 池化层1，2*2采样
layers.Conv2D(32, (3, 3), activation='relu'), # 卷积层2，卷积核3*3
layers.AveragePooling2D((2, 2)), # 池化层2，2*2采样
layers.Conv2D(64, (3, 3), activation='relu'), # 卷积层3，卷积核3*3
layers.Dropout(0.3), # 让神经元以一定的概率停止工作，防止过拟合，提高模型的泛化能力。
layers.Flatten(), # Flatten层，连接卷积层与全连接层
layers.Dense(128, activation='relu'), # 全连接层，特征进一步提取
layers.Dense(num_classes) # 输出层，输出预期结果
])
model.summary() # 打印网络结构
模型结构如下所示
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
rescaling (Rescaling) (None, 180, 180, 3) 0
_________________________________________________________________
conv2d (Conv2D) (None, 178, 178, 16) 448
_________________________________________________________________
average_pooling2d (AveragePo (None, 89, 89, 16) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 87, 87, 32) 4640
_________________________________________________________________
average_pooling2d_1 (Average (None, 43, 43, 32) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 41, 41, 64) 18496
_________________________________________________________________
dropout (Dropout) (None, 41, 41, 64) 0
_________________________________________________________________
flatten (Flatten) (None, 107584) 0
_________________________________________________________________
dense (Dense) (None, 128) 13770880
_________________________________________________________________
dense_1 (Dense) (None, 4) 516
=================================================================
Total params: 13,794,980
Trainable params: 13,794,980
Non-trainable params: 0
三 编译
opt = tf.keras.optimizers.Adam(learning_rate=0.001)
model.compile(optimizer=opt,
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
四 训练模型
epochs = 30
history = model.fit(
train_ds,
validation_data=val_ds,
epochs=epochs
)
训练过程如下
29/29 [==============================] - 3s 120ms/step - loss: 0.0443 - accuracy: 0.9878 - val_loss: 1.0009 - val_accuracy: 0.7911
Epoch 2/30
29/29 [==============================] - 5s 159ms/step - loss: 0.1180 - accuracy: 0.9622 - val_loss: 0.7385 - val_accuracy: 0.7911
Epoch 3/30
29/29 [==============================] - 3s 116ms/step - loss: 0.1075 - accuracy: 0.9600 - val_loss: 0.4700 - val_accuracy: 0.8444
Epoch 4/30
29/29 [==============================] - 3s 113ms/step - loss: 0.0436 - accuracy: 0.9856 - val_loss: 0.5489 - val_accuracy: 0.8444
Epoch 5/30
29/29 [==============================] - 4s 146ms/step - loss: 0.0857 - accuracy: 0.9722 - val_loss: 0.6501 - val_accuracy: 0.8622
Epoch 6/30
29/29 [==============================] - 4s 127ms/step - loss: 0.1282 - accuracy: 0.9567 - val_loss: 0.5230 - val_accuracy: 0.8667
Epoch 7/30
29/29 [==============================] - 3s 111ms/step - loss: 0.0696 - accuracy: 0.9800 - val_loss: 0.5487 - val_accuracy: 0.8489
Epoch 8/30
29/29 [==============================] - 3s 117ms/step - loss: 0.0330 - accuracy: 0.9922 - val_loss: 0.5734 - val_accuracy: 0.8578
Epoch 9/30
29/29 [==============================] - 5s 160ms/step - loss: 0.0157 - accuracy: 0.9944 - val_loss: 0.5595 - val_accuracy: 0.8756
Epoch 10/30
29/29 [==============================] - 3s 115ms/step - loss: 0.0107 - accuracy: 0.9989 - val_loss: 0.5145 - val_accuracy: 0.8844
Epoch 11/30
29/29 [==============================] - 3s 114ms/step - loss: 0.0135 - accuracy: 0.9967 - val_loss: 0.5809 - val_accuracy: 0.8800
Epoch 12/30
29/29 [==============================] - 4s 129ms/step - loss: 0.0062 - accuracy: 0.9989 - val_loss: 0.5968 - val_accuracy: 0.8756
Epoch 13/30
29/29 [==============================] - 4s 140ms/step - loss: 0.0044 - accuracy: 1.0000 - val_loss: 0.6313 - val_accuracy: 0.8978
Epoch 14/30
29/29 [==============================] - 3s 113ms/step - loss: 0.0029 - accuracy: 0.9989 - val_loss: 0.7265 - val_accuracy: 0.8800
Epoch 15/30
29/29 [==============================] - 3s 110ms/step - loss: 0.0099 - accuracy: 0.9967 - val_loss: 0.6557 - val_accuracy: 0.8711
Epoch 16/30
29/29 [==============================] - 4s 140ms/step - loss: 0.0054 - accuracy: 1.0000 - val_loss: 0.6172 - val_accuracy: 0.8756
Epoch 17/30
29/29 [==============================] - 4s 131ms/step - loss: 0.0034 - accuracy: 1.0000 - val_loss: 0.6395 - val_accuracy: 0.8800
Epoch 18/30
29/29 [==============================] - 3s 112ms/step - loss: 0.0039 - accuracy: 0.9978 - val_loss: 0.6551 - val_accuracy: 0.8889
Epoch 19/30
29/29 [==============================] - 3s 111ms/step - loss: 0.0031 - accuracy: 1.0000 - val_loss: 0.6968 - val_accuracy: 0.8844
Epoch 20/30
29/29 [==============================] - 4s 150ms/step - loss: 0.0014 - accuracy: 1.0000 - val_loss: 0.6774 - val_accuracy: 0.8889
Epoch 21/30
29/29 [==============================] - 3s 121ms/step - loss: 6.1163e-04 - accuracy: 1.0000 - val_loss: 0.6914 - val_accuracy: 0.8933
Epoch 22/30
29/29 [==============================] - 3s 112ms/step - loss: 3.5274e-04 - accuracy: 1.0000 - val_loss: 0.7049 - val_accuracy: 0.8889
Epoch 23/30
29/29 [==============================] - 3s 114ms/step - loss: 2.6827e-04 - accuracy: 1.0000 - val_loss: 0.7202 - val_accuracy: 0.8889
Epoch 24/30
29/29 [==============================] - 5s 162ms/step - loss: 2.2787e-04 - accuracy: 1.0000 - val_loss: 0.7417 - val_accuracy: 0.8889
Epoch 25/30
29/29 [==============================] - 3s 112ms/step - loss: 1.8400e-04 - accuracy: 1.0000 - val_loss: 0.7201 - val_accuracy: 0.8889
Epoch 26/30
29/29 [==============================] - 3s 109ms/step - loss: 1.8860e-04 - accuracy: 1.0000 - val_loss: 0.7283 - val_accuracy: 0.8889
Epoch 27/30
29/29 [==============================] - 4s 141ms/step - loss: 1.5586e-04 - accuracy: 1.0000 - val_loss: 0.7083 - val_accuracy: 0.8933
Epoch 28/30
29/29 [==============================] - 4s 137ms/step - loss: 1.3365e-04 - accuracy: 1.0000 - val_loss: 0.7398 - val_accuracy: 0.8889
Epoch 29/30
29/29 [==============================] - 3s 109ms/step - loss: 1.3278e-04 - accuracy: 1.0000 - val_loss: 0.7296 - val_accuracy: 0.8889
Epoch 30/30
29/29 [==============================] - 3s 110ms/step - loss: 1.0332e-04 - accuracy: 1.0000 - val_loss: 0.7297 - val_accuracy: 0.8889
五 预测
plt.imshow(test_images[1])
import numpy as np
pre = model.predict(test_images)
print(class_names[np.argmax(pre[1])])
ship
plt.figure(figsize=(12, 4))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, acc, label='Training Accuracy')
plt.plot(epochs_range, val_acc, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss, label='Training Loss')
plt.plot(epochs_range, val_loss, label='Validation Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
可以看出，当训练epoch在20次以后，训练集精度已经到达1，训练集误差已经到0。
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
