實戰七：手把手教你用TensorFlow進行驗證碼識別（下）

實戰七：手把手教你用Tensorflow進行驗證碼識別（下）

github下載地址

目錄

• 準備模型開發環境
• 生成驗證碼數據集
• 輸入與輸出數據處理
• 模型結構設計
• 模型損失函數設計
• 模型訓練過程分析
• 模型部署與效果演示

七、模型部署與效果演示

1.數據-模型-服務流水線

2.啓動一個flask服務

將下述代碼另存爲hello.py，然後本地python hello.py運行

from flask import Flask
app = Flask(__name__)

@app.route('/')
def hello_world():
    return 'Hello World!'

if __name__ == '__main__':
    app.run()

訪問127.0.0.1:5000，便可看到結果

3.部署驗證碼識別服務

將下述代碼另外存app.py，啓動flask來加載app.py文件

import base64

import numpy as np
import tensorflow as tf

from io import BytesIO
from flask import Flask, request, jsonify
from keras.models import load_model
from PIL import Image

NUMBER = ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
LOWERCASE = ['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u',
            'v', 'w', 'x', 'y', 'z']
UPPERCASE = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U',
           'V', 'W', 'X', 'Y', 'Z']

CAPTCHA_CHARSET = NUMBER   # 驗證碼字符集
CAPTCHA_LEN = 4            # 驗證碼長度
CAPTCHA_HEIGHT = 60        # 驗證碼高度
CAPTCHA_WIDTH = 160        # 驗證碼寬度

# 10 個 Epochs 訓練的模型
MODEL_FILE = './pre-trained/model/captcha_rmsprop_binary_crossentropy_bs_100_epochs_10.h5'

def vec2text(vector):
    if not isinstance(vector, np.ndarray):
        vector = np.asarray(vector)
    vector = np.reshape(vector, [CAPTCHA_LEN, -1])
    text = ''
    for item in vector:
        text += CAPTCHA_CHARSET[np.argmax(item)]
    return text

def rgb2gray(img):
    # Y' = 0.299 R + 0.587 G + 0.114 B 
    # https://en.wikipedia.org/wiki/Grayscale#Converting_color_to_grayscale
    return np.dot(img[...,:3], [0.299, 0.587, 0.114])

app = Flask(__name__) # 創建 Flask 實例

# 測試 URL
@app.route('/ping', methods=['GET', 'POST'])
def hello_world():
    return 'pong'

# 驗證碼識別 URL
@app.route('/predict', methods=['POST'])
def predict():
    response = {'success': False, 'prediction': '', 'debug': 'error'}
    received_image= False
    if request.method == 'POST':
        if request.files.get('image'): # 圖像文件
            image = request.files['image'].read()
            received_image = True
            response['debug'] = 'get image'
        elif request.get_json(): # base64 編碼的圖像文件
            encoded_image = request.get_json()['image']
            image = base64.b64decode(encoded_image)
            received_image = True
            response['debug'] = 'get json'
        if received_image:
            image = np.array(Image.open(BytesIO(image)))
            image = rgb2gray(image).reshape(1, 60, 160, 1).astype('float32') / 255
            with graph.as_default():
                pred = model.predict(image)
            response['prediction'] = response['prediction'] + vec2text(pred)
            response['success'] = True
            response['debug'] = 'predicted'
    else:
        response['debug'] = 'No Post'
    return jsonify(response)

model = load_model(MODEL_FILE) # 加載模型
graph = tf.get_default_graph() # 獲取 TensorFlow 默認數據流圖

if __name__ == "__main__":
    app.run()

3.使用Flask啓動驗證碼識別服務

a.啓動flask服務

b.打開遊覽器訪問測試URL（http://localhost:5000/ping）

c.訪問驗證碼識別服務

利用curl進行圖像上傳

curl -X POST -F [email protected] "http://localhost:5000/predict"

另開cmd終端，運行上述命令

可以看出圖片真實值爲0044，預測結果爲0244