卫星图片的Classification_model

Tensorflow版本：2.6.0
使用的是CNN神经网络，网络结构在最后给出
飞机和湖泊的卫星图片二分类网络
数据集请点击链接：https://www.kaggle.com/datasets/yo7oyo/lake-plane-binaryclass
数据集的构成：airplane: 700 张， lake 700 张

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd

import glob
%matplotlib inline

# 获取数据路径
# 'D:\\datasets\\dset\\dataset1\\卫星图像识别数据\\2_class\\airplane\\airplane_427.jpg 路径格式
img_path = glob.glob(r'D:\datasets\dset\dataset1\卫星图像识别数据\2_class/*/*.jpg')
import random
random.shuffle(img_path)
img_path[1].split('\\')[-2]

output: 'airplane'

#得到标签值label
#path: 
dict_img = {'lake':1, 'airplane':0}
dict_label = dict((v, k) for k, v in dict_img.items())
label = [dict_img.get(img.split('\\')[-2]) for img in img_path]
print(dict_label)

读取图片:

"""
读取： x = tf.io.read_file(path) 读取二进制数据
解码： x = tf.image.decode_jpeg(x) 返回Tensor张量 支持多种图片解码
类型转化 tf.cast(x, tf.float64)
"""
def load_image(path):
    # 读取图片
    img = tf.io.read_file(path)
    # 解码数据
    img = tf.image.decode_jpeg(img, channels=3)    # 转换图片大小 统一大小为 256 X 256
    img = tf.image.resize(img, [256, 256])
    # 数据转化
    img = tf.cast(img, tf.float32)
    # 归一化
    img /= 255.0
    return img
#%%
p, l = img_path[20], label[20]
img_tensor = load_image(p)
plt.title(dict_label.get(l))
plt.imshow(img_tensor)

# 创建dataset
ds_img = tf.data.Dataset.from_tensor_slices(img_path)
ds_img = ds_img.map(load_image)

# 标签的dataset
ds_label = tf.data.Dataset.from_tensor_slices(label)

print(ds_img, '\n', ds_label)

# 使用zip函数将数据进行合并
datas = tf.data.Dataset.zip((ds_img, ds_label))
print(datas)

all_count = len(img_path)
BATCH_SIZE = 16

# 划分训练集和测试集
train_ds = datas.take(int(all_count * 0.8))
test_ds = datas.skip(int(all_count * 0.8))

# 设置训练集和测试集
"""
测试集无需乱序 和 重复
"""
train_ds = train_ds.repeat().shuffle(100).batch(BATCH_SIZE)
test_ds = test_ds.batch(BATCH_SIZE)
print(train_ds, test_ds)

构建模型

model = tf.keras.Sequential([
    layers.Conv2D(64, (3, 3), input_shape=(256, 256, 3), activation='relu'),
    layers.Conv2D(64, (3, 3), activation='relu'),
    layers.MaxPool2D(),
    # layers.Dropout(0.5),
    layers.Conv2D(128, (3, 3), activation='relu'),
    layers.Conv2D(128, (3, 3), activation='relu'),
    layers.MaxPool2D(),
    # layers.Dropout(0.5),
    layers.Conv2D(256, (3, 3), activation='relu'),
    layers.Conv2D(256, (3, 3), activation='relu'),
    layers.MaxPool2D(),
    # layers.Dropout(0.5),
    layers.Conv2D(512, (3, 3), activation='relu'),
    layers.Conv2D(512, (3, 3), activation='relu'),
    layers.MaxPool2D(),
    layers.Conv2D(512, (3, 3), activation='relu'),
    layers.Conv2D(512, (3, 3), activation='relu'),
    layers.Conv2D(512, (3, 3), activation='relu'),
    layers.MaxPool2D(),
    layers.GlobalAveragePooling2D(),
    layers.Dense(1024, activation='relu'),
    layers.Dense(256, activation='relu'),
    layers.Dense(1, activation='sigmoid')
])

model.summary()

model.compile(
    optimizer=tf.keras.optimizers.Adam(learning_rate=0.0001),
    loss=tf.keras.losses.BinaryCrossentropy(),
    metrics=['acc']
)

steps_pre_epoch = int(all_count*0.8) // BATCH_SIZE
val_steps = int(all_count * 0.2) // BATCH_SIZE

history = model.fit(
    train_ds,
    epochs=10,
    steps_per_epoch=steps_pre_epoch,
    validation_data=test_ds,
    validation_steps=val_steps
)

plt.plot(history.epoch, history.history.get('loss'), label='loss')
plt.plot(history.epoch, history.history.get('val_loss'), label='val_loss')
plt.legend()

plt.plot(history.epoch, history.history.get('acc'), label='acc')
plt.plot(history.epoch, history.history.get('val_acc'), label='val_acc')
plt.legend()

# 使用模型进行预测
plt.figure(figsize=(20,20))
for i in range(10):
    plt.subplot(2,5,i+1)
    index = random.randint(0, 1400)
    plt.imshow(load_image(img_path[index]))
    img = load_image(img_path[index])
    plt.title(f'Predict:{dict_label.get(float(model.predict(tf.expand_dims(img, axis=0))[0]) > 0.5)} \nTrue:{dict_label.get(label[index])}')
plt.show()

加入BatchNormalization()的批量归一化，防止出现梯度消失的梯度爆炸并加快神经网络的传递和更新

构建模型

model = tf.keras.Sequential([
    layers.Conv2D(64, (3, 3), input_shape=(256, 256, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.Conv2D(64, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    
    layers.MaxPool2D(),
    # layers.Dropout(0.5),
    layers.Conv2D(128, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.Conv2D(128, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.MaxPool2D(),
    # layers.Dropout(0.5),
    layers.Conv2D(256, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.Conv2D(256, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.MaxPool2D(),
    # layers.Dropout(0.5),
    layers.Conv2D(512, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.Conv2D(512, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.MaxPool2D(),
    layers.Conv2D(512, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.Conv2D(512, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.Conv2D(512, (3, 3)),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.MaxPool2D(),
    layers.GlobalAveragePooling2D(),
    layers.Dense(1024),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.Dense(256),
    layers.BatchNormalization(),
    layers.Activation('relu'),
    layers.Dense(1, activation='sigmoid')
])

model.summary()
# 训练模型
model.compile(
    optimizer=tf.keras.optimizers.Adam(learning_rate=0.0001),
    loss=tf.keras.losses.BinaryCrossentropy(),
    metrics=['acc']
)
steps_pre_epoch = int(all_count*0.8) // BATCH_SIZE
val_steps = int(all_count * 0.2) // BATCH_SIZE
history = model.fit(
    train_ds,
    epochs=10,
    steps_per_epoch=steps_pre_epoch,
    validation_data=test_ds,
    validation_steps=val_steps
)

未加入批量归一化的网络结构

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 254, 254, 64)      1792      
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 252, 252, 64)      36928     
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 126, 126, 64)      0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 124, 124, 128)     73856     
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 122, 122, 128)     147584    
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 61, 61, 128)       0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 59, 59, 256)       295168    
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 57, 57, 256)       590080    
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 28, 28, 256)       0         
_________________________________________________________________
conv2d_6 (Conv2D)            (None, 26, 26, 512)       1180160   
_________________________________________________________________
conv2d_7 (Conv2D)            (None, 24, 24, 512)       2359808   
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 12, 12, 512)       0         
_________________________________________________________________
conv2d_8 (Conv2D)            (None, 10, 10, 512)       2359808   
_________________________________________________________________
conv2d_9 (Conv2D)            (None, 8, 8, 512)         2359808   
_________________________________________________________________
conv2d_10 (Conv2D)           (None, 6, 6, 512)         2359808   
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 3, 3, 512)         0         
_________________________________________________________________
global_average_pooling2d (Gl (None, 512)               0         
_________________________________________________________________
dense (Dense)                (None, 1024)              525312    
_________________________________________________________________
dense_1 (Dense)              (None, 256)               262400    
_________________________________________________________________
dense_2 (Dense)              (None, 1)                 257       
=================================================================
Total params: 12,552,769
Trainable params: 12,552,769
Non-trainable params: 0