tensorflow学习018——200中鸟类图片分类实例

数据链接：https://pan.baidu.com/s/1zxa2KnkW5nFYFhF_bzfxkg

提取码：eii7

需要注意的是将里面所有的路径改成自己下载的数据所在的路径

 

from tensorflow import keras
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
import glob

#1 获取图片和标签
imgs_path = glob.glob(r"E:\WORK\tensorflow\dataset\日月光华_birds分类竞赛数据\birds_train\*\*.jpg")
all_labels_name = [img_p.split("\\")[-2].split(".")[1] for img_p in imgs_path]
label_names = np.unique(all_labels_name)
label_to_index = dict((name, i) for i, name in enumerate(label_names))
index_to_label = dict((v,k) for k, v in label_to_index.items())

#2 创建dataset 读取图片
all_labels = [label_to_index.get(name) for name in all_labels_name]
np.random.seed(2021)
random_index = np.random.permutation(len(imgs_path)) #返回一个乱序，可以对img和label同时进行乱序操作
imgs_path = np.array(imgs_path)[random_index]
all_labels = np.array(all_labels)[random_index]
train_count = int(len(imgs_path)*0.8)
train_path = imgs_path[:train_count]
train_labels = all_labels[:train_count]
test_path = imgs_path[train_count:]
test_labels = all_labels[train_count:]
train_ds = tf.data.Dataset.from_tensor_slices((train_path,train_labels))
test_ds = tf.data.Dataset.from_tensor_slices((test_path,test_labels))
def load_img(path, label):
    image = tf.io.read_file(path)
    image = tf.image.decode_jpeg(image, channels=3)
    image = tf.image.resize(image, [256,256])
    image = tf.cast(image,tf.float32)
    image = image / 255
    return image, label

#3 模型和损失函数

AUTOTUNE = tf.data.experimental.AUTOTUNE
train_ds = train_ds.map(load_img, num_parallel_calls=AUTOTUNE) #会根据CPU的情况开启多线程
test_ds = test_ds.map(load_img, num_parallel_calls=AUTOTUNE)
BATCH_SIZE = 8
train_ds = train_ds.repeat().shuffle(100).batch(BATCH_SIZE)
test_ds = test_ds.batch(BATCH_SIZE)
model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(64,(3,3),input_shape=(256,256,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.Conv2D(64,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.MaxPooling2D(),
    tf.keras.layers.Conv2D(128,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.Conv2D(128,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.MaxPooling2D(),
    tf.keras.layers.Conv2D(256,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.Conv2D(256,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.MaxPooling2D(),
    tf.keras.layers.Conv2D(512,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.Conv2D(512,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.MaxPooling2D(),
    tf.keras.layers.Conv2D(512,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.Conv2D(512,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.Conv2D(512,(3,3)),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.GlobalAveragePooling2D(),
    tf.keras.layers.Dense(1024),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.Dense(512),
    tf.keras.layers.BatchNormalization(),
    tf.keras.layers.Activation('relu'),
    tf.keras.layers.Dense(200) #激活放在损失函数中
])
model.compile(optimizer=tf.keras.optimizers.Adam(0.0001),
              loss = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),#并进行激活
              metrics=['acc']
              )

# 4训练和预测
train_count = len(train_path)
test_count = len(test_path)
steps_per_epoch = train_count // BATCH_SIZE
validation_steps = test_count // BATCH_SIZE
history = model.fit(
    train_ds,epochs=10,
    steps_per_epoch=steps_per_epoch,
    validation_data=test_ds,
    validation_steps=validation_steps
)

def load_and_preprocess_image(path):
    img_raw = tf.io.read_file(path)
    img_tensor = tf.image.decode_jpeg(img_raw, channels=3)  # tf.Tensor 每一个元素的大小都是[0,255] 是uint8
    img_tensor = tf.image.resize(img_tensor, [256, 256])  # 将所有图片统一大小
    # 转换为float类型
    img_tensor = tf.cast(img_tensor, tf.float32)
    # 归一化
    img_tensor = img_tensor / 255
    return img_tensor

test_img = r"E:\WORK\tensorflow\dataset\日月光华_birds分类竞赛数据\birds_test\0.jpg"
test_tensor = load_and_preprocess_image(test_img)
test_tensor = tf.expand_dims(test_tensor, axis=0)
pred = model.predict(test_tensor)
print(index_to_label.get(np.argmax(pred)))