用Keras对URP教务验证码的识别训练
python3环境
pip install -r requirements.txt
python3 model_test.py模型及训练集均来源于网上，请勿用于非法用途
来源于网上，具体链接忘记了。大约8W张标记好的训练集,如需要，可以ISSUE里留言
有一种可行的方案是观察验证码的特征，模拟生成。本人水平有限，就不尝试了。
参考项目https://github.com/JasonLiTW/simple-railway-captcha-solver
参考https://github.com/JasonLiTW/simple-railway-captcha-solver (参考项目是6字符的)
model.summary()
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) (None, 56, 176, 3) 0
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, 56, 176, 32) 896 input_1[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, 54, 174, 32) 9248 conv2d_1[0][0]
__________________________________________________________________________________________________
batch_normalization_1 (BatchNor (None, 54, 174, 32) 128 conv2d_2[0][0]
__________________________________________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None, 27, 87, 32) 0 batch_normalization_1[0][0]
__________________________________________________________________________________________________
dropout_1 (Dropout) (None, 27, 87, 32) 0 max_pooling2d_1[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, 27, 87, 64) 18496 dropout_1[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 25, 85, 64) 36928 conv2d_3[0][0]
__________________________________________________________________________________________________
batch_normalization_2 (BatchNor (None, 25, 85, 64) 256 conv2d_4[0][0]
__________________________________________________________________________________________________
max_pooling2d_2 (MaxPooling2D) (None, 12, 42, 64) 0 batch_normalization_2[0][0]
__________________________________________________________________________________________________
dropout_2 (Dropout) (None, 12, 42, 64) 0 max_pooling2d_2[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 12, 42, 128) 73856 dropout_2[0][0]
__________________________________________________________________________________________________
conv2d_6 (Conv2D) (None, 10, 40, 128) 147584 conv2d_5[0][0]
__________________________________________________________________________________________________
batch_normalization_3 (BatchNor (None, 10, 40, 128) 512 conv2d_6[0][0]
__________________________________________________________________________________________________
max_pooling2d_3 (MaxPooling2D) (None, 5, 20, 128) 0 batch_normalization_3[0][0]
__________________________________________________________________________________________________
dropout_3 (Dropout) (None, 5, 20, 128) 0 max_pooling2d_3[0][0]
__________________________________________________________________________________________________
conv2d_7 (Conv2D) (None, 3, 18, 256) 295168 dropout_3[0][0]
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None, 3, 18, 256) 1024 conv2d_7[0][0]
__________________________________________________________________________________________________
max_pooling2d_4 (MaxPooling2D) (None, 1, 9, 256) 0 batch_normalization_4[0][0]
__________________________________________________________________________________________________
flatten_1 (Flatten) (None, 2304) 0 max_pooling2d_4[0][0]
__________________________________________________________________________________________________
dropout_4 (Dropout) (None, 2304) 0 flatten_1[0][0]
__________________________________________________________________________________________________
dense_1 (Dense) (None, 36) 82980 dropout_4[0][0]
__________________________________________________________________________________________________
dense_2 (Dense) (None, 36) 82980 dropout_4[0][0]
__________________________________________________________________________________________________
dense_3 (Dense) (None, 36) 82980 dropout_4[0][0]
__________________________________________________________________________________________________
dense_4 (Dense) (None, 36) 82980 dropout_4[0][0]
==================================================================================================
Total params: 916,016
Trainable params: 915,056
Non-trainable params: 960
(这里是照搬的图，实际用的时候是4字符的)
可以考虑使用RNN https://github.com/ypwhs/captcha_break
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent