From f20b79d0bb1ec96994bf5529736a5738b2708d85 Mon Sep 17 00:00:00 2001
From: zhai_pro <xin_soft@foxmail.com>
Date: Sun, 27 Jan 2019 12:31:08 +0800
Subject: [PATCH] =?UTF-8?q?=E8=AE=A9=E5=8D=B7=E7=A7=AF=E7=A5=9E=E7=BB=8F?=
 =?UTF-8?q?=E7=BD=91=E7=BB=9C=E7=94=A8=E7=BB=9F=E8=AE=A1=E5=AD=A6=E5=BE=97?=
 =?UTF-8?q?=E5=87=BA=E7=9A=84=E6=95=B0=E6=8D=AE=E9=9B=86=E6=9D=A5=E5=AD=A6?=
 =?UTF-8?q?=E4=B9=A0?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 mlearn.py           |  2 +-
 mlearn_for_image.py | 66 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 67 insertions(+), 1 deletion(-)
 create mode 100644 mlearn_for_image.py

diff --git a/mlearn.py b/mlearn.py
index 7fff3c1..9437788 100644
--- a/mlearn.py
+++ b/mlearn.py
@@ -61,7 +61,7 @@ def main():
                         validation_data=(test_x, test_y),
                         callbacks=[reduce_lr])
     savefig(history, start=10)
-    model.save('model.h5')
+    model.save('model.h5', include_optimizer=False)
 
 
 def load_data_v2():
diff --git a/mlearn_for_image.py b/mlearn_for_image.py
new file mode 100644
index 0000000..2fb483b
--- /dev/null
+++ b/mlearn_for_image.py
@@ -0,0 +1,66 @@
+# coding: utf-8
+import cv2
+import numpy as np
+from keras import models
+from keras import layers
+from keras import optimizers
+from keras.applications import VGG16
+from keras.callbacks import ReduceLROnPlateau
+from keras.preprocessing.image import ImageDataGenerator
+
+
+def load_data():
+    # 这是统计学专家提供的训练集
+    data = np.load('captcha.npz')
+    train_x, train_y = data['images'], data['labels']
+    train_x = train_x / 255.0
+    # 由于是统计得来的信息，所以在此给定可信度
+    sample_weight = train_y.max(axis=1) / np.sqrt(train_y.sum(axis=1))
+    sample_weight /= sample_weight.mean()
+    train_y = train_y.argmax(axis=1)
+
+    # 这是人工提供的验证集
+    data = np.load('captcha.test.npz')
+    test_x, test_y = data['images'], data['labels']
+    # resize
+    n = test_x.shape[0]
+    new_test_x = np.empty((n, 67, 67, 3), dtype=np.uint8)
+    for idx in range(n):
+        new_test_x[idx] = cv2.resize(test_x[idx], (67, 67))
+    test_x = new_test_x / 255.0
+    return (train_x, train_y, sample_weight), (test_x, test_y)
+
+
+def main():
+    (train_x, train_y, sample_weight), (test_x, test_y) = load_data()
+    datagen = ImageDataGenerator(horizontal_flip=True,
+                                 vertical_flip=True)
+    train_generator = datagen.flow(train_x, train_y, sample_weight=sample_weight)
+    _, h, w, c = train_x.shape
+    base = VGG16(weights='imagenet', include_top=False, input_shape=(h, w, c))
+    for layer in base.layers[:-4]:
+        layer.trainable = False
+    model = models.Sequential([
+        base,
+        layers.Conv2D(64, (3, 3), activation='relu', padding='same'),
+        layers.GlobalAveragePooling2D(),
+        layers.Dense(64, activation='relu'),
+        layers.Dropout(0.20),
+        layers.Dense(80, activation='softmax')
+    ])
+    model.compile(optimizer=optimizers.RMSprop(lr=1e-5),
+                  loss='sparse_categorical_crossentropy',
+                  metrics=['accuracy'])
+    model.summary()
+    reduce_lr = ReduceLROnPlateau(verbose=1)
+    model.fit_generator(train_generator, epochs=400,
+                        steps_per_epoch=100,
+                        validation_data=(test_x[:800], test_y[:800]),
+                        callbacks=[reduce_lr])
+    result = model.evaluate(test_x, test_y)
+    print(result)
+    model.save('12306.image.model.h5', include_optimizer=False)
+
+
+if __name__ == '__main__':
+    main()