使用卷积神经网络构建的模型识别手写数字并实现了GUI如何将每个数字的识别率显示出来

小诺爷

1. from tensorflow.keras import datasets, layers, optimizers, Sequential,callbacks,losses
   
2. import os
   
3. import numpy as np
   
4. 
   
5. np.set_printoptions(threshold=np.inf)
   
6. 
   
7. (x_train, y_train), (x_test, y_test) = datasets.mnist.load_data("mnist.pkl")
   
8. x_train, x_test = x_train / 255.0, x_test / 255.0
   
9. 
   
10. model = Sequential([
    
11.     layers.InputLayer(input_shape=(28, 28)),
    
12.     layers.Reshape((28, 28, 1)),
    
13.     layers.Conv2D(kernel_size=3, strides=1, filters=16,padding='same', activation='relu', name='layer_conv1'),
    
14.     layers.MaxPooling2D(pool_size=2, strides=2),
    
15.     layers.BatchNormalization(),
    
16.     layers.Conv2D(kernel_size=3, strides=1, filters=36,padding='same', activation='relu', name='layer_conv2'),
    
17.     layers.MaxPooling2D(pool_size=2, strides=2),
    
18.     layers.BatchNormalization(),
    
19.     layers.Flatten(),
    
20.     layers.Dense(units=128, activation='selu'),
    
21.     layers.BatchNormalization(),
    
22.     layers.Dense(units=10, activation='softmax')
    
23. ])
    
24. 
    
25. 
    
26. model.compile(optimizer=optimizers.Adam(lr=1e-3),loss=losses.SparseCategoricalCrossentropy(from_logits=False),metrics=['sparse_categorical_accuracy'])
    
27. 
    
28. checkpoint_save_path = "./checkpoint/mnist.ckpt"
    
29. if os.path.exists(checkpoint_save_path + '.index'):
    
30.     print('-------------load the model-----------------')
    
31.     model.load_weights(checkpoint_save_path)
    
32. 
    
33. cp_callback = callbacks.ModelCheckpoint(filepath=checkpoint_save_path,
    
34.                                                  save_weights_only=True,
    
35.                                                  save_best_only=True)
    
36. model.fit(x=x_train, y=y_train,batch_size=32,epochs=5,validation_data=(x_test, y_test), validation_freq=1,callbacks=[cp_callback])
    
37. 
    
38. model.summary()
    
39. file = open('weights.txt', 'w')
    
40. for v in model.trainable_variables:
    
41.     file.write(str(v.name) + '\n')
    
42.     file.write(str(v.shape) + '\n')
    
43.     file.write(str(v.numpy()) + '\n')
    
44. file.close()

复制代码

hhiop

这个有点复杂，估计小甲鱼都不知道

dadan1

山理工？

Victor0321

卷i起来 哈哈哈

Levin-e

首先你得实现test()吧，输入你的测试数据然后计算每个数字的准确率，然后就是想办法再GUI里显示出来了，主要工作量是在计算准确率那边，顺便也可以算一手召回率，然后求个F1