机器学习-哺乳动物识别

机器学习——哺乳动物识别

（一）选题背景：

哺乳动物是动物世界中形态结构最高等、生理机能最完善的动物。与其他动物相比，哺乳动物最突出的特征在于胎生以及其幼崽由母体分泌的乳汁喂养长大。所有哺乳动物都长有毛发，以保持体温的恒定，适应各种复杂的生存环境；哺乳动物具有比较发达的大脑，因而能产生比其他动物更为复杂的行为，并能不断地改变自己的行为，以适应外界环境的变化。

（二）选题原因

最开始接触是手写，到后面开始学习到猫狗，我的想法就是写一个比较偏向于动物类型的，就写了一个关于猴子的识别，图片量大概在10000左右，但是为觉得猴子的判定太单调了，就加大类型选择到了哺乳动物

（三）实现步骤

1.导入对应包

#导入需要用到的库
import numpy as np
import pandas as pd
import os
import tensorflow as tf
import matplotlib.pyplot as plt
from pathlib import Path
from sklearn.model_selection import train_test_split
from keras.models import Sequential
from keras.layers import Activation
from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPooling2D
from keras.applications.resnet import preprocess_input

from keras_preprocessing.image import ImageDataGenerator
from keras.models import load_model
from keras.utils import image_utils
from keras import optimizers

from tensorflow.python.keras import layers, models
from keras.preprocessing.image import ImageDataGenerator as IDG
import plotly.express as px
from keras.layers import GlobalAveragePooling2D as GAP, Dense, Dropout

from keras.callbacks import EarlyStopping, ModelCheckpoint
from tensorflow.keras.applications import ResNet50V2, InceptionV3, Xception, ResNet152V2
from keras.callbacks import EarlyStopping, ModelCheckpoint

2.获取文件夹

# 训练集
root_path = './animals10/raw-img/'

# 获取文件夹的名字
class_names = sorted(os.listdir(root_path))
n_classes = len(class_names)

print(f"Class Names : {class_names}")
print(f"Number of Classes  : {n_classes}")
# 每个动物的个数
class_dis = [len(os.listdir(root_path + name)) for name in class_names]
class_dis
# Visualization
fig = px.pie(names=class_names, values=class_dis, title="哺乳动物种类训练集分布")
fig.update_layout({'title':{'x':0.45}})
fig.show()

# 测试集
pred_root_path = './animals10/test/'

# 获取文件夹的名字
testclass_names = sorted(os.listdir(pred_root_path))
n_testclasses = len(testclass_names)

print(f"Class Names : {testclass_names}")
print(f"Number of Classes  : {n_testclasses}")

# 动物的分布
testclass_dis = [len(os.listdir(pred_root_path + name)) for name in class_names]
testclass_dis

# Visualization
figtest = px.pie(names=testclass_names, values=testclass_dis, title="哺乳动物种类训练集分布")
figtest.update_layout({'title':{'x':0.45}})
figtest.show()

3.预处理

# 预处理
train_gen = IDG(rescale=1./255, rotation_range=10, horizontal_flip=True, validation_split=0.1)
test_gen = IDG(rescale=1./255)

# 加载数据
train_ds = train_gen.flow_from_directory(root_path,
                                         target_size=(256,256), 
                                         shuffle=True, 
                                         batch_size=32, 
                                         subset='training', 
                                         class_mode='binary')
valid_ds = train_gen.flow_from_directory(root_path, 
                                         target_size=(256,256), 
                                         shuffle=True,
                                         batch_size=32, 
                                         subset='validation', 
                                         class_mode='binary')
test_ds = train_gen.flow_from_directory(pred_root_path, 
                                        target_size=(256,256), 
                                        shuffle=True, 
                                        batch_size=32, 
                                        class_mode='binary')

4.图片显示函数

def show_images(data, GRID=[6,5], size=(20,25), model=None, class_names=class_names):

    # 行列
    n_rows, n_cols = GRID
    n_images = n_rows * n_cols

    # 大小
    plt.figure(figsize=size)

    # 数据
    images, labels = next(iter(data))
    
    # Do not repeat id
    ids = []

    # 数据
    for i in range(1, n_images+1):
        
        #选择随机的图片
        id = np.random.randint(len(images))
        if id in ids:
            id = np.random.randint(len(images))
        ids.append(id)
        image, label = images[id], class_names[int(labels[id])]

        
        plt.subplot(n_rows, n_cols, i)
        plt.imshow(image)
        plt.axis('off')

        # 选择标题
        if model is None:
            title = f"Class : {label}"
        else:
            pred = class_names[np.argmax(model.predict(image[np.newaxis,...]))]
            title = f"Class : {label} \nPred : {pred}"
        plt.title(title)


    # Show Plot
    plt.show()
	
	
#运行测试
show_images(data=train_ds)

5.训练模型

# 模型编译
base_model = ResNet50V2(input_shape=(256,256,3), include_top=False)
base_model.trainable = False

# 模型的名字
name = "animals"

model = Sequential([
    base_model,
    GAP(),
    Dense(256, activation='relu', kernel_initializer='he_normal'),
    Dropout(0.2),
    Dense(10, activation='softmax')
], name=name)

# 编译模型
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

# Callbacks 
cbs = [EarlyStopping(patience=3, restore_best_weights=True), ModelCheckpoint(name + ".h5", save_best_only=True)]

# 训练
h2=model.fit(train_ds, validation_data=valid_ds, callbacks=cbs, epochs=10)

6.绘制损失和精度曲线

#绘制精度和损失曲线
accuracy = h2.history['accuracy']
loss = h2.history['loss']
val_loss = h2.history['val_loss']
val_accuracy = h2.history['val_accuracy']
plt.figure(figsize=(17, 7))
plt.subplot(2, 2, 1)
plt.plot(range(7), accuracy,'bo', label='Training Accuracy')
plt.plot(range(7), val_accuracy, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Accuracy : Training vs. Validation ')
plt.subplot(2, 2, 2)
plt.plot(range(7), loss,'bo' ,label='Training Loss')
plt.plot(range(7), val_loss, label='Validation Loss')
plt.title('Loss : Training vs. Validation ')
plt.legend(loc='upper right')
plt.show()

7. 模型加载，评估，测试

# 模型加载
model_path = './animals.h5'
model = load_model(model_path)
model.summary()

# 模型评估
model.evaluate(test_ds)

# 结果预测
show_images(data=test_ds, model=model, GRID=[6,5], size=(25,25))

（四）全部代码

#导入需要用到的库
import numpy as np
import pandas as pd
import os
import tensorflow as tf
import matplotlib.pyplot as plt
from pathlib import Path
from sklearn.model_selection import train_test_split
from keras.models import Sequential
from keras.layers import Activation
from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPooling2D
from keras.applications.resnet import preprocess_input

from keras_preprocessing.image import ImageDataGenerator
from keras.models import load_model
from keras.utils import image_utils
from keras import optimizers

from tensorflow.python.keras import layers, models
from keras.preprocessing.image import ImageDataGenerator as IDG
import plotly.express as px
from keras.layers import GlobalAveragePooling2D as GAP, Dense, Dropout

from keras.callbacks import EarlyStopping, ModelCheckpoint
from tensorflow.keras.applications import ResNet50V2, InceptionV3, Xception, ResNet152V2
from keras.callbacks import EarlyStopping, ModelCheckpoint

# 训练集
root_path = './animals10/raw-img/'

# 获取文件夹的名字
class_names = sorted(os.listdir(root_path))
n_classes = len(class_names)

print(f"Class Names : {class_names}")
print(f"Number of Classes  : {n_classes}")
# 每个动物的个数
class_dis = [len(os.listdir(root_path + name)) for name in class_names]
class_dis
# Visualization
fig = px.pie(names=class_names, values=class_dis, title="哺乳动物种类训练集分布")
fig.update_layout({'title':{'x':0.45}})
fig.show()

# 测试集
pred_root_path = './animals10/test/'

# 获取文件夹的名字
testclass_names = sorted(os.listdir(pred_root_path))
n_testclasses = len(testclass_names)

print(f"Class Names : {testclass_names}")
print(f"Number of Classes  : {n_testclasses}")

# 动物的分布
testclass_dis = [len(os.listdir(pred_root_path + name)) for name in class_names]
testclass_dis

# Visualization
figtest = px.pie(names=testclass_names, values=testclass_dis, title="哺乳动物种类训练集分布")
figtest.update_layout({'title':{'x':0.45}})
figtest.show()

# 预处理
train_gen = IDG(rescale=1./255, rotation_range=10, horizontal_flip=True, validation_split=0.1)
test_gen = IDG(rescale=1./255)

# Load the data
train_ds = train_gen.flow_from_directory(root_path,
                                         target_size=(256,256), 
                                         shuffle=True, 
                                         batch_size=32, 
                                         subset='training', 
                                         class_mode='binary')
valid_ds = train_gen.flow_from_directory(root_path, 
                                         target_size=(256,256), 
                                         shuffle=True,
                                         batch_size=32, 
                                         subset='validation', 
                                         class_mode='binary')
test_ds = train_gen.flow_from_directory(pred_root_path, 
                                        target_size=(256,256), 
                                        shuffle=True, 
                                        batch_size=32, 
                                        class_mode='binary')

def show_images(data, GRID=[6,5], size=(20,25), model=None, class_names=class_names):

    # 行列
    n_rows, n_cols = GRID
    n_images = n_rows * n_cols

    # 大小
    plt.figure(figsize=size)

    # 数据
    images, labels = next(iter(data))
    
    # Do not repeat id
    ids = []

    # 数据
    for i in range(1, n_images+1):
        
        # 选择随机的图片
        id = np.random.randint(len(images))
        if id in ids:
            id = np.random.randint(len(images))
        ids.append(id)
        image, label = images[id], class_names[int(labels[id])]

        
        plt.subplot(n_rows, n_cols, i)
        plt.imshow(image)
        plt.axis('off')

        # 选择标题
        if model is None:
            title = f"Class : {label}"
        else:
            pred = class_names[np.argmax(model.predict(image[np.newaxis,...]))]
            title = f"Class : {label} \nPred : {pred}"
        plt.title(title)


    # Show Plot
    plt.show()

#运行测试
show_images(data=train_ds)

# 模型编译
base_model = ResNet50V2(input_shape=(256,256,3), include_top=False)
base_model.trainable = False

# 模型的名字
name = "animals"

model = Sequential([
    base_model,
    GAP(),
    Dense(256, activation='relu', kernel_initializer='he_normal'),
    Dropout(0.2),
    Dense(10, activation='softmax')
], name=name)

# 编译模型
model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy'])

# Callbacks 
cbs = [EarlyStopping(patience=3, restore_best_weights=True), ModelCheckpoint(name + ".h5", save_best_only=True)]

# 训练
h2=model.fit(train_ds, validation_data=valid_ds, callbacks=cbs, epochs=10)

#绘制精度和损失曲线
accuracy = h2.history['accuracy']
loss = h2.history['loss']
val_loss = h2.history['val_loss']
val_accuracy = h2.history['val_accuracy']
plt.figure(figsize=(17, 7))
plt.subplot(2, 2, 1)
plt.plot(range(7), accuracy,'bo', label='Training Accuracy')
plt.plot(range(7), val_accuracy, label='Validation Accuracy')
plt.legend(loc='lower right')
plt.title('Accuracy : Training vs. Validation ')
plt.subplot(2, 2, 2)
plt.plot(range(7), loss,'bo' ,label='Training Loss')
plt.plot(range(7), val_loss, label='Validation Loss')
plt.title('Loss : Training vs. Validation ')
plt.legend(loc='upper right')
plt.show()

# 模型加载
model_path = './animals.h5'
model = load_model(model_path)
model.summary()

# 模型评估
model.evaluate(test_ds)

# 结果预测
show_images(data=test_ds, model=model, GRID=[6,5], size=(25,25))