diff --git a/mlearn.py b/mlearn.py
new file mode 100644
index 0000000..8b40327
--- /dev/null
+++ b/mlearn.py
@@ -0,0 +1,57 @@
+# coding: utf-8
+import pathlib
+
+import cv2
+import numpy as np
+
+
+def load_data():
+    data = np.load('texts.npz')
+    texts, labels = data['texts'], data['labels']
+    n = int(texts.shape[0] * 0.9)   # 90%用于训练，10%用于测试
+    return (texts[:n], labels[:n]), (texts[n:], labels[n:])
+
+
+def main():
+    from keras import models
+    from keras import layers
+    (train_x, train_y), (test_x, test_y) = load_data()
+    _, train_x = cv2.threshold(train_x, 220, 1, cv2.THRESH_BINARY)
+    _, test_x = cv2.threshold(test_x, 220, 1, cv2.THRESH_BINARY)
+    model = models.Sequential([
+        layers.Flatten(),
+        layers.Dense(500, activation='relu'),
+        layers.Dense(80, activation='softmax'),
+    ])
+    model.compile(optimizer='adam',
+                  loss='sparse_categorical_crossentropy',
+                  metrics=['accuracy'])
+    model.fit(train_x, train_y, epochs=50)
+    print(model.evaluate(test_x, test_y))
+    model.save('model.h5')
+
+
+def predict():
+    from keras import models
+    model = models.load_model('model.h5')
+    texts = np.load('data.npy')
+    _, texts = cv2.threshold(texts, 220, 1, cv2.THRESH_BINARY)
+    labels = model.predict(texts)
+    np.save('labels.npy', labels)
+
+
+def show():
+    texts = np.load('data.npy')
+    labels = np.load('labels.npy')
+    labels = labels.argmax(axis=1)
+    pathlib.Path('classify').mkdir(exist_ok=True)
+    for idx, (text, label) in enumerate(zip(texts, labels)):
+        # 使用聚类结果命名
+        fn = f'classify/{label}.{idx}.jpg'
+        cv2.imwrite(fn, text)
+
+
+if __name__ == '__main__':
+    main()
+    predict()
+    show()