這寫的也太複雜了…
注意細節

import pickle
import numpy as np
import matplotlib.pyplot as plt
from keras.models import Model
from keras.layers import Dropout, Dense, Conv2D, Flatten, MaxPooling2D,Input,BatchNormalization,Activation
from keras.callbacks import ModelCheckpoint
from keras import utils  # 歸一化變數
from sklearn.
 
model_selection import train_test_split
from keras.layers.merge import concatenate
from keras.utils import plot_model
from keras.optimizers  import Adam
from keras import regularizers

f1 = open('test_image.p', 'rb')
X = pickle.load(f1)
f1.close()

f1 = open('y_test.p', 'rb')
y = pickle.load(f1)
f1.
 
close()
Y = np.array(y)

y1 = utils.to_categorical(Y, 8)
X_train, X_test, Y_train, Y_test = train_test_split(X, y1, test_size=0.2,random_state=0)

f = []
for i in range(8):
    tmp = X_train[:,:,:,i]
    tmp = np.expand_dims(tmp,axis=-1)
    f.append(tmp)
    
f1= []
for i in range(8):
    tmp = X_test[
 
:,:,:,i]
    tmp = np.expand_dims(tmp,axis=-1)
    f1.append(tmp)

inputA = Input(shape=(64,64,1))
inputB = Input(shape=(64,64,1))
inputC = Input(shape=(64,64,1))
inputD = Input(shape=(64,64,1))
inputE = Input(shape=(64,64,1))
inputF = Input(shape=(64,64,1))
inputG = Input(shape=(64,64,1))
inputH = Input(shape=(64,64,1))

weight_decay = 0.005 

def tmp_model(input_data):
    x = Conv2D(32,(3,3),kernel_regularizer=regularizers.l2(weight_decay))(input_data)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = MaxPooling2D((2, 2))(x)
    x = Conv2D(64,(3,3),kernel_regularizer=regularizers.l2(weight_decay))(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = MaxPooling2D((2, 2))(x)
    x = Conv2D(128,(3,3),kernel_regularizer=regularizers.l2(weight_decay))(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = MaxPooling2D((2, 2))(x)
    x = Conv2D(256,(3,3),kernel_regularizer=regularizers.l2(weight_decay))(x)
    x = BatchNormalization()(x)
    x = Activation('relu')(x)
    x = MaxPooling2D((2, 2))(x)
    x = Flatten()(x)
    M = Model(inputs=input_data,outputs=x)
    return M

m1 = tmp_model(inputA)
m2 = tmp_model(inputB)
m3 = tmp_model(inputC)
m4 = tmp_model(inputD)
m5 = tmp_model(inputE)
m6 = tmp_model(inputF)
m7 = tmp_model(inputG)
m8 = tmp_model(inputH)

com = concatenate([m1.output,m2.output,m3.output,m4.output,m5.output,m6.output,m7.output,m8.output],axis=-1)
z = Dense(1024,activation='relu')(com)
z = Dropout(0.5)(z)
z = Dense(8,activation='softmax')(z)

model = Model(inputs=[m1.input,m2.input,m3.input,m4.input,m5.input,m6.input,m7.input,m8.input],outputs=z)

filepath = "best_8.hdf5"
# 只把準確率最高的一次，儲存下來
checkpoint = ModelCheckpoint(filepath
                             , monitor='val_accuracy'
                             , verbose=0, save_best_only=True, mode='max')

callbacks_list = [checkpoint]
model.compile(loss='categorical_crossentropy'
              , optimizer='adam', metrics=['accuracy'])

BATCH_SIZE = 64
EPOCHS = 100
history = model.fit([f[0],f[1],f[2],f[3],f[4],f[5],f[6],f[7]], Y_train, batch_size=BATCH_SIZE
                    , epochs=EPOCHS
                    , shuffle=True
                    , callbacks=callbacks_list, validation_data=([f1[0],f1[1],f1[2],f1[3],f1[4],f1[5],f1[6],f1[7]],Y_test), verbose=1)


# 儲存一些訓練集和測試集上的accuracy和loss
t_a = history.history['accuracy']
v_a = history.history['val_accuracy']


t_l = history.history['loss']
v_l = history.history['val_loss']

# 把accuracy變化的曲線存到txt中
np.savetxt('acc_curve8.txt',(t_a,v_a))

plt.plot(t_a)
plt.plot(v_a)
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='best')
plt.show()

# summarize history for loss
np.savetxt('loss_curve8.txt',(t_l,v_l))

plt.plot(t_l)
plt.plot(v_l)
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='best')
plt.show()
s = model.evaluate([f1[0],f1[1],f1[2],f1[3],f1[4],f1[5],f1[6],f1[7]], Y_test)
print(s[1])