# 小型CNN模型搭建和测试
import picture_read
import numpy as np
import tensorflow as tf
import time
def data_processing(train_document1,train_document2,test_document1,test_document2, #(训练集数据1，训练集数据2，测试集数据1，测试集数据2)
pic_size,label1,label2,): #(图片缩放大小，标签1，标签2)
train_data_1,train_label_1 = picture_read.picture_document_deal(train_document1, pic_size, label1) #sxs标签为0
train_data_2,train_label_2 = picture_read.picture_document_deal(train_document2, pic_size, label2) #xsx标签为1
test_data_1,test_label_1 = picture_read.picture_document_deal(test_document1, pic_size, label1) #sxs标签为0
test_data_2,test_label_2 = picture_read.picture_document_deal(test_document2, pic_size, label2) #xsx标签为1
# numpy:vstack列上合并两个矩阵,hstack在行上合并两个矩阵
train_data = np.vstack((train_data_1,train_data_2))
train_labels = np.hstack((train_label_1,train_label_2))
test_data = np.vstack((test_data_1,test_data_2))
test_labels = np.hstack((test_label_1,test_label_2))
print('train_data = ',train_data.shape) #train_data,
# print('train_labels = ', train_labels, train_labels.shape)
# 二、 归一化,将每个点的像素值归一化到[0, 1]之间
training_pics = train_data.reshape(train_data.shape[0], train_data.shape[1], train_data.shape[2], 1)
test_pics = test_data.reshape(test_data.shape[0], test_data.shape[1], test_data.shape[2], 1)
training_pics = training_pics / 255.0
test_pics = test_pics / 255.0
print('training_pics = ',training_pics.shape) #training_pics,
# print('test_pics = ',test_pics,test_pics.shape)
print(training_pics.shape[1], training_pics.shape[2], training_pics.shape[3])
print("training_pics = ", training_pics.shape)
print("training_labels = ", train_labels.shape)
return training_pics,test_pics,train_labels,test_labels
def model_structure(shape,classification ): #（图片的格式，最终分类的结果）
print('shape = ',shape)
# 第一种写法
# 定义模型，按顺序创建模型
model_def = tf.keras.models.Sequential([
tf.keras.layers.Conv2D(6, (3, 3), activation='relu', input_shape=(shape[1], shape[2], shape[3])),
tf.keras.layers.MaxPooling2D(2, 2),
tf.keras.layers.ReLU(),
tf.keras.layers.Conv2D(16, (3, 3), activation='relu'),
tf.keras.layers.MaxPooling2D(2, 2),
tf.keras.layers.ReLU(),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(120, activation='relu'),
tf.keras.layers.Dense(84, activation='relu'),
tf.keras.layers.Dense(classification, activation='softmax')
])
# # 第二种写法
# inputs = tf.keras.layers.Input(shape = shape[1:])
# x = tf.keras.layers.Conv2D(filters=6, kernel_size=3, strides=1,
# activation='relu', padding='same')(inputs)
# x = tf.keras.layers.MaxPool2D(pool_size=2, strides=1, padding='same')(x)
# x = tf.keras.layers.ReLU()(x)
# x = tf.keras.layers.Conv2D(filters=16, kernel_size=3, strides=1,
# activation='relu', padding='same')(x)
# x = tf.keras.layers.MaxPool2D(pool_size=2, strides=1, padding='same')(x)
# x = tf.keras.layers.ReLU()(x)
# x = tf.keras.layers.Flatten()(x)
# x = tf.keras.layers.Dense(120,activation='relu')(x)
# x = tf.keras.layers.Dense(84,activation='relu')(x)
# outputs = tf.keras.layers.Dense(classification,activation='softmax')(x) # 二分类
# model_def = tf.keras.Model(inputs = inputs, outputs=outputs)
# 编译模型，确定优化方法，损失函数等
model_def.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])
return model_def
def model_train_save(model,model_file): #(模型保存文件的名称)
# 四、训练模型,并将每轮训练的历史信息保存在变量 history 中
history = model.fit(x=training_images, y=train_label, batch_size = 32 ,
epochs = 20,validation_split=0.2) # validation_split = 0.1 validation_data=(test_images, test_labels)
# 计算损失
test_loss = model.evaluate(test_images, test_label)
# print('test_loss = ',test_loss)
# 模型保存
model.save(model_file) # 保存全部
# 删除模型
del model
def model_predict_labels(model_file,test_pic,test_labels): #测试集有标签时使用
# 四、恢复模型并预测
# print("模型读回")
model = tf.keras.models.load_model(model_file)
# 测试模型
loss, acc = model.evaluate(test_pic, test_labels)
# 五、预测
predict = model.predict(test_pic)
# print('predict = ',predict)
predict = np.array(tf.argmax(predict, 1))
print('predict = ',predict) #tf.argmax根据axis取值的不同返回每行或者每列最大值的索引,axis = 1时取每行最大值的索引位置
print('test_labels = ', test_labels)
print("Restored model, accuracy:{:5.2f}%".format(100 * acc))
return predict
def model_predict_nolabels(model_file,test_pic): #测试集无标签时使用
# 四、恢复模型
# print("模型读回")
model = tf.keras.models.load_model(model_file)
# 五、预测
predict = model.predict(test_pic)
# print('predict = ',predict)
predict = np.array(tf.argmax(predict, 1))
print('predict = ',predict) #tf.argmax根据axis取值的不同返回每行或者每列最大值的索引,axis = 1时取每行最大值的索引位置
return predict
if __name__ == '__main__':
start_time = time.time()
# 一、获取数据并处理数据
im_size = (28, 28) #图片放缩的大小
train_document_1 = 'D:\\python_work\\model_project\\dataset\\train\\sxs'
train_document_2 = 'D:\\python_work\\model_project\\dataset\\train\\xsx'
test_document_1 = 'D:\\python_work\\model_project\\dataset\\test\\sxs'
test_document_2 = 'D:\\python_work\\model_project\\dataset\\test\\xsx'
training_images, test_images, train_label, test_label = data_processing(train_document_1,train_document_2,
test_document_1,test_document_2, #(训练集数据1，训练集数据2，测试集数据1，测试集数据2)
im_size,0,1,)
# 二、模型搭建:
model_name = model_structure(training_images.shape,classification = 2)
# 三、训练神经网络模型进行图片分类:
model_train_save(model_name,'my_model.h5')
# 四、模型读回及预测（有对比标签）
predict_labels = model_predict_labels('my_model.h5',test_images,test_label)
# # 五、模型读回及预测（无对比标签）
# predict_labels = model_predict_nolabels('my_model.h5',test_images)
end_time = time.time()
print('程序运行时间：',end_time-start_time)
