LeNet图像检测代码以及注释

最新推荐文章于 2024-11-18 21:59:35 发布

我是花花花

最新推荐文章于 2024-11-18 21:59:35 发布

阅读量337

点赞数 5

文章标签： jupyter python 人工智能

本文链接：https://blog.youkuaiyun.com/m0_61604615/article/details/136730149

版权

本文介绍了如何使用TensorFlow库中的KerasAPI对MNIST手写数字数据集进行预处理、构建卷积神经网络模型，并通过训练和验证提升模型性能，最终保存模型以便后续使用。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

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

(train_images,train_labels),(test_images,test_labels)=tf.keras.datasets.mnist.load_data()

train_images=train_images.reshape(60000,28,28,1)
test_images=test_images.reshape(10000,28,28,1)

train_images=train_images/255
test_images=test_images/255

train_labels=np.array(pd.get_dummies(train_labels))
test_labels=np.array(pd.get_dummies(test_labels))

model=tf.keras.Sequential()

model.add(tf.keras.layers.Conv2D(filters=6,kernel_size=(5,5),input_shape=(28,28,1),padding='same',activation="sigmoid"))
model.add(tf.keras.layers.AveragePooling2D(pool_size=(2,2)))
model.add(tf.keras.layers.Conv2D(filters=16,kernel_size=(5,5),activation="sigmoid"))
model.add(tf.keras.layers.AveragePooling2D(pool_size=(2,2)))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(84,activation="sigmoid"))
model.add(tf.keras.layers.Dense(10,activation="softmax"))#因为是多分类问题

model.summary()

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d (Conv2D)             (None, 28, 28, 6)         156       
                                                                 
 average_pooling2d (Average  (None, 14, 14, 6)         0         
 Pooling2D)                                                      
                                                                 
 conv2d_1 (Conv2D)           (None, 10, 10, 16)        2416      
                                                                 
 average_pooling2d_1 (Avera  (None, 5, 5, 16)          0         
 gePooling2D)                                                    
                                                                 
 flatten (Flatten)           (None, 400)               0         
                                                                 
 dense (Dense)               (None, 84)                33684     
                                                                 
 dense_1 (Dense)             (None, 10)                850       
                                                                 
=================================================================
Total params: 37106 (144.95 KB)
Trainable params: 37106 (144.95 KB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['acc'])

model.compile()：配置模型的学习过程，包括选择优化器、损失函数和评估指标等
metrics=[‘acc’]:表示模型在训练和验证过程中将会监测并输出模型的准确率。

history=model.fit(train_images,train_labels,epochs=10,validation_data=(test_images,test_labels))

Epoch 1/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.7652 - acc: 0.7607 - val_loss: 0.2475 - val_acc: 0.9280
Epoch 2/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.2062 - acc: 0.9373 - val_loss: 0.1559 - val_acc: 0.9530
Epoch 3/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.1385 - acc: 0.9576 - val_loss: 0.1039 - val_acc: 0.9699
Epoch 4/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.1014 - acc: 0.9693 - val_loss: 0.0778 - val_acc: 0.9756
Epoch 5/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0813 - acc: 0.9751 - val_loss: 0.0649 - val_acc: 0.9798
Epoch 6/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0686 - acc: 0.9795 - val_loss: 0.0638 - val_acc: 0.9800
Epoch 7/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0593 - acc: 0.9822 - val_loss: 0.0546 - val_acc: 0.9838
Epoch 8/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0533 - acc: 0.9840 - val_loss: 0.0467 - val_acc: 0.9848
Epoch 9/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0477 - acc: 0.9855 - val_loss: 0.0463 - val_acc: 0.9851
Epoch 10/10
1875/1875 [==============================] - 9s 5ms/step - loss: 0.0432 - acc: 0.9867 - val_loss: 0.0515 - val_acc: 0.9834

model.evaluate(test_images,test_labels)

313/313 [==============================] - 1s 3ms/step - loss: 0.0515 - acc: 0.9834





[0.05148673057556152, 0.9833999872207642]

model.save('mnist.h5')

最后一步保存模型一边下次调用