多光谱遥感分类（二）：VGG微调

继上篇。

CNN模型

使用VGG16预训练模型。

from keras.preprocessing.image import ImageDataGenerator
from keras.callbacks import TensorBoard
from keras.applications import VGG16
from keras import layers,models,optimizers



train_pic_gen=ImageDataGenerator(rescale=1./255,rotation_range=20,width_shift_range=0.2,height_shift_range=0.2,
                                 shear_range=0.2,zoom_range=0.5,horizontal_flip=True,fill_mode='nearest')
test_pic_gen=ImageDataGenerator(rescale=1./255)
train_flow=train_pic_gen.flow_from_directory("data/1_train",(128,128),batch_size=128,class_mode='categorical')
test_flow=test_pic_gen.flow_from_directory("data/1_test",(128,128),batch_size=128,class_mode='categorical')

# print(test_flow.class_indices)
### {'169': 0, '26': 1, '6': 2}

conv_base=VGG16(weights='imagenet',include_top=False,input_shape=(128,128,3))
model = models.Sequential()
model.add(conv_base)
model.add(layers.Flatten())
model.add(layers.Dense(1024, activation='relu'))
model.add(layers.Dense(20, activation='softmax'))

trainable=False
for layer in conv_base.layers:
    if layer.name=='block4_conv1':
        trainable=True
    layer.trainable=trainable
model.summary()

model.compile(loss='categorical_crossentropy',optimizer=optimizers.adam(lr=1e-5),metrics=['acc'])

model.fit_generator(train_flow,steps_per_epoch=60,epochs=100,verbose=1,
                    validation_data=test_flow,validation_steps=20,callbacks=[TensorBoard(log_dir='output/logs/t')])

model.save("output/classify3_vgg16_pro.h5")

查看结果（混淆矩阵、kappa、分类报告）。test_in_te.py

from keras.models import load_model
import numpy as np
import cv2
import os

model=load_model("output/classify3_vgg16_pro.h5")
vec={'bf': 0, 'br': 1, 'cc': 2, 'cx_b': 3, 'cx_gw': 4, 'cx_rw': 5, 'dp': 6, 'gf': 7, 'gh': 8, 'gr': 9, 'greenh': 10, 'lt': 11, 'of': 12, 'ptc': 13, 'rf': 14, 'st': 15, 'wm': 16, 'wr': 17, 'wt': 18, 'xkc': 19}

#使用测试集数据验证
res=[]
true_label=[]

def get_inputs(src=[]):
    pre_x = []
    for s in src:
        input = cv2.imread(s)
        input = cv2.resize(input, (128, 128))
        input = cv2.cvtColor(input, cv2.COLOR_BGR2RGB)
        pre_x.append(input)  # input一张图片
    pre_x = np.array(pre_x) / 255.0
    return pre_x

org_dir="data/1_test/"
for dir in os.listdir(org_dir):
    print("deal:%s" % dir)
    label=dir
    for d in os.listdir(org_dir + dir):
        pre_x=get_inputs([org_dir+dir+"/"+d])
        pred=np.argmax(model.predict(pre_x)[0])
        pred=list(vec.keys())[pred]
        res.append(pred)
        true_label.append(label)

from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.metrics import cohen_kappa_score


print(confusion_matrix(true_label,res))
print()
print(classification_report(true_label,res))
print("kappa: ",cohen_kappa_score(true_label,res))
mat = confusion_matrix(true_label,res)
sns.heatmap(mat,annot=True,square=True,fmt="d",xticklabels=list(vec.keys()),yticklabels=list(vec.keys()))
plt.show()