完整代码：GitHub
我的简书：Awesome_Tang的简书
更好的阅读体验可访问我的Kesci Lab：AwesomeTang的Kesci Lab
整个项目代码分为三部分：
Generrate_Captcha:- 生成验证码图片（训练集，验证集和测试集）；
- 读取图片数据和标签（标签即为图片文件名）；
cnn_model:卷积神经网络；driver:模型训练及评估。
Generate Captcha
配置项
class Config(object):
width = 160 # 验证码图片的宽
height = 60 # 验证码图片的高
char_num = 4 # 验证码字符个数
characters = range(10) # 数字[0,9]
test_folder = 'test' # 测试集文件夹，下同
train_folder = 'train'
validation_folder = 'validation'
tensorboard_folder = 'tensorboard' # tensorboard的log路径
generate_num = (5000, 500, 500) # 训练集，验证集和测试集数量
alpha = 1e-3 # 学习率
Epoch = 100 # 训练轮次
batch_size = 64 # 批次数量
keep_prob = 0.5 # dropout比例
print_per_batch = 20 # 每多少次输出结果
save_per_batch = 20 # 每多少次写入tensorboard
生成验证码（class Generate）
- 验证码图片示例：
check_path():检查文件夹是否存在，如不存在则创建。gen_captcha():生成验证码方法，写入之前检查是否以存在，如存在重新生成。
读取数据（classs ReadData）
-
read_data():返回图片数组（numpy.array格式）和标签（即文件名）； -
label2vec():将文件名转为向量；-
例：
label = '1327' label_vec = [0,1,0,0,0,0,0,0,0,0, 0,0,0,1,0,0,0,0,0,0, 0,0,1,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,1,0,0]
-
-
load_data():加载文件夹下所有图片，返回图片数组，标签和图片数量。
定义模型（cnn_model）
采用三层卷积，filter_size均为5，为避免过拟合，每层卷积后面均接dropout操作，最终将$16060$的图像转为$208$的矩阵。
- 大致结构如下：
训练&评估
next_batch()：迭代器，分批次返还数据；feed_data()：给模型“喂”数据；x：图像数组；y：图像标签；keep_prob：dropout比例；
evaluate()：模型评估，用于验证集和测试集。run_model()：训练&评估
目前效果
目前经过4000次迭代训练集准确率可达99%以上，测试集准确率93%，还是存在一点过拟合，不过现在模型是基于CPU训练的，完成一次训练耗费时间大约4个小时左右，后续调整了再进行更新。
Images for train ：10000, for validation : 1000, for test : 1000
Epoch : 1
Step 0, train_acc: 7.42%, train_loss: 1.43, val_acc: 9.85%, val_loss: 1.40, improved:*
Step 20, train_acc: 12.50%, train_loss: 0.46, val_acc: 10.35%, val_loss: 0.46, improved:*
Step 40, train_acc: 9.38%, train_loss: 0.37, val_acc: 10.10%, val_loss: 0.37, improved:
Step 60, train_acc: 7.42%, train_loss: 0.34, val_acc: 10.25%, val_loss: 0.34, improved:
Step 80, train_acc: 7.81%, train_loss: 0.33, val_acc: 9.82%, val_loss: 0.33, improved:
Step 100, train_acc: 12.11%, train_loss: 0.33, val_acc: 10.00%, val_loss: 0.33, improved:
Step 120, train_acc: 9.77%, train_loss: 0.33, val_acc: 10.07%, val_loss: 0.33, improved:
Step 140, train_acc: 8.98%, train_loss: 0.33, val_acc: 10.40%, val_loss: 0.33, improved:*
Epoch : 2
Step 160, train_acc: 8.20%, train_loss: 0.33, val_acc: 10.52%, val_loss: 0.33, improved:*
...
Epoch : 51
Step 7860, train_acc: 100.00%, train_loss: 0.01, val_acc: 92.37%, val_loss: 0.08, improved:
Step 7880, train_acc: 99.61%, train_loss: 0.01, val_acc: 92.28%, val_loss: 0.08, improved:
Step 7900, train_acc: 100.00%, train_loss: 0.01, val_acc: 92.42%, val_loss: 0.08, improved:
Step 7920, train_acc: 100.00%, train_loss: 0.00, val_acc: 92.83%, val_loss: 0.08, improved:
Step 7940, train_acc: 100.00%, train_loss: 0.01, val_acc: 92.77%, val_loss: 0.08, improved:
Step 7960, train_acc: 100.00%, train_loss: 0.01, val_acc: 92.68%, val_loss: 0.08, improved:
Step 7980, train_acc: 100.00%, train_loss: 0.00, val_acc: 92.63%, val_loss: 0.09, improved:
No improvement for over 1000 steps, auto-stopping....
Test accuracy: 93.00%, loss: 0.08
- Tensorboard
每次训练之前将Tensorboard路径下的文件删除，不然趋势图上会凌乱。- Accurracy
- loss
- Accurracy