网络编程
位置:首页>> 网络编程>> Python编程>> Python+Tensorflow+CNN实现车牌识别的示例代码

Python+Tensorflow+CNN实现车牌识别的示例代码

作者:苹果の芳菲  发布时间:2021-06-30 08:44:05 

标签:Python,Tensorflow,CNN,车牌识别

一、项目概述

本次项目目标是实现对自动生成的带有各种噪声的车牌识别。在噪声干扰情况下,车牌字符分割较困难,此次车牌识别是将车牌7个字符同时训练,字符包括31个省份简称、10个阿拉伯数字、24个英文字母('O'和'I'除外),共有65个类别,7个字符使用单独的loss函数进行训练。
(运行环境:tensorflow1.14.0-GPU版)

二、生成车牌数据集


import os
import cv2 as cv
import numpy as np
from math import *
from PIL import ImageFont
from PIL import Image
from PIL import ImageDraw

index = {"京": 0, "沪": 1, "津": 2, "渝": 3, "冀": 4, "晋": 5, "蒙": 6, "辽": 7, "吉": 8, "黑": 9,
      "苏": 10, "浙": 11, "皖": 12, "闽": 13, "赣": 14, "鲁": 15, "豫": 16, "鄂": 17, "湘": 18, "粤": 19,
      "桂": 20, "琼": 21, "川": 22, "贵": 23, "云": 24, "藏": 25, "陕": 26, "甘": 27, "青": 28, "宁": 29,
      "新": 30, "0": 31, "1": 32, "2": 33, "3": 34, "4": 35, "5": 36, "6": 37, "7": 38, "8": 39,
      "9": 40, "A": 41, "B": 42, "C": 43, "D": 44, "E": 45, "F": 46, "G": 47, "H": 48, "J": 49,
      "K": 50, "L": 51, "M": 52, "N": 53, "P": 54, "Q": 55, "R": 56, "S": 57, "T": 58, "U": 59,
      "V": 60, "W": 61, "X": 62, "Y": 63, "Z": 64}

chars = ["京", "沪", "津", "渝", "冀", "晋", "蒙", "辽", "吉", "黑",
      "苏", "浙", "皖", "闽", "赣", "鲁", "豫", "鄂", "湘", "粤",
      "桂", "琼", "川", "贵", "云", "藏", "陕", "甘", "青", "宁",
      "新", "0", "1", "2", "3", "4", "5", "6", "7", "8",
      "9", "A", "B", "C", "D", "E", "F", "G", "H", "J",
      "K", "L", "M", "N", "P", "Q", "R", "S", "T", "U",
      "V", "W", "X", "Y", "Z"]

def AddSmudginess(img, Smu):
 """
 模糊处理
 :param img: 输入图像
 :param Smu: 模糊图像
 :return: 添加模糊后的图像
 """
 rows = r(Smu.shape[0] - 50)
 cols = r(Smu.shape[1] - 50)
 adder = Smu[rows:rows + 50, cols:cols + 50]
 adder = cv.resize(adder, (50, 50))
 img = cv.resize(img,(50,50))
 img = cv.bitwise_not(img)
 img = cv.bitwise_and(adder, img)
 img = cv.bitwise_not(img)
 return img

def rot(img, angel, shape, max_angel):
 """
 添加透视畸变
 """
 size_o = [shape[1], shape[0]]
 size = (shape[1]+ int(shape[0] * cos((float(max_angel ) / 180) * 3.14)), shape[0])
 interval = abs(int(sin((float(angel) / 180) * 3.14) * shape[0]))
 pts1 = np.float32([[0, 0], [0, size_o[1]], [size_o[0], 0], [size_o[0], size_o[1]]])
 if angel > 0:
   pts2 = np.float32([[interval, 0], [0, size[1]], [size[0], 0], [size[0] - interval, size_o[1]]])
 else:
   pts2 = np.float32([[0, 0], [interval, size[1]], [size[0] - interval, 0], [size[0], size_o[1]]])
 M = cv.getPerspectiveTransform(pts1, pts2)
 dst = cv.warpPerspective(img, M, size)
 return dst

def rotRandrom(img, factor, size):
 """
 添加放射畸变
 :param img: 输入图像
 :param factor: 畸变的参数
 :param size: 图片目标尺寸
 :return: 放射畸变后的图像
 """
 shape = size
 pts1 = np.float32([[0, 0], [0, shape[0]], [shape[1], 0], [shape[1], shape[0]]])
 pts2 = np.float32([[r(factor), r(factor)], [r(factor), shape[0] - r(factor)], [shape[1] - r(factor), r(factor)],
           [shape[1] - r(factor), shape[0] - r(factor)]])
 M = cv.getPerspectiveTransform(pts1, pts2)
 dst = cv.warpPerspective(img, M, size)
 return dst

def tfactor(img):
 """
 添加饱和度光照的噪声
 """
 hsv = cv.cvtColor(img,cv.COLOR_BGR2HSV)
 hsv[:, :, 0] = hsv[:, :, 0] * (0.8 + np.random.random() * 0.2)
 hsv[:, :, 1] = hsv[:, :, 1] * (0.3 + np.random.random() * 0.7)
 hsv[:, :, 2] = hsv[:, :, 2] * (0.2 + np.random.random() * 0.8)
 img = cv.cvtColor(hsv, cv.COLOR_HSV2BGR)
 return img

def random_envirment(img, noplate_bg):
 """
 添加自然环境的噪声, noplate_bg为不含车牌的背景图
 """
 bg_index = r(len(noplate_bg))
 env = cv.imread(noplate_bg[bg_index])
 env = cv.resize(env, (img.shape[1], img.shape[0]))
 bak = (img == 0)
 bak = bak.astype(np.uint8) * 255
 inv = cv.bitwise_and(bak, env)
 img = cv.bitwise_or(inv, img)
 return img

def GenCh(f, val):
 """
 生成中文字符
 """
 img = Image.new("RGB", (45, 70), (255, 255, 255))
 draw = ImageDraw.Draw(img)
 draw.text((0, 3), val, (0, 0, 0), font=f)
 img = img.resize((23, 70))
 A = np.array(img)
 return A

def GenCh1(f, val):
 """
 生成英文字符
 """
 img =Image.new("RGB", (23, 70), (255, 255, 255))
 draw = ImageDraw.Draw(img)
 draw.text((0, 2), val, (0, 0, 0), font=f)  # val.decode('utf-8')
 A = np.array(img)
 return A

def AddGauss(img, level):
 """
 添加高斯模糊
 """
 return cv.blur(img, (level * 2 + 1, level * 2 + 1))

def r(val):
 return int(np.random.random() * val)

def AddNoiseSingleChannel(single):
 """
 添加高斯噪声
 """
 diff = 255 - single.max()
 noise = np.random.normal(0, 1 + r(6), single.shape)
 noise = (noise - noise.min()) / (noise.max() - noise.min())
 noise *= diff
 # noise= noise.astype(np.uint8)
 dst = single + noise
 return dst

def addNoise(img):  # sdev = 0.5,avg=10
 img[:, :, 0] = AddNoiseSingleChannel(img[:, :, 0])
 img[:, :, 1] = AddNoiseSingleChannel(img[:, :, 1])
 img[:, :, 2] = AddNoiseSingleChannel(img[:, :, 2])
 return img

class GenPlate:
 def __init__(self, fontCh, fontEng, NoPlates):
   self.fontC = ImageFont.truetype(fontCh, 43, 0)
   self.fontE = ImageFont.truetype(fontEng, 60, 0)
   self.img = np.array(Image.new("RGB", (226, 70),(255, 255, 255)))
   self.bg = cv.resize(cv.imread("data\\images\\template.bmp"), (226, 70))  # template.bmp:车牌背景图
   self.smu = cv.imread("data\\images\\smu2.jpg")  # smu2.jpg:模糊图像
   self.noplates_path = []
   for parent, parent_folder, filenames in os.walk(NoPlates):
     for filename in filenames:
       path = parent + "\\" + filename
       self.noplates_path.append(path)

def draw(self, val):
   offset = 2
   self.img[0:70, offset+8:offset+8+23] = GenCh(self.fontC, val[0])
   self.img[0:70, offset+8+23+6:offset+8+23+6+23] = GenCh1(self.fontE, val[1])
   for i in range(5):
     base = offset + 8 + 23 + 6 + 23 + 17 + i * 23 + i * 6
     self.img[0:70, base:base+23] = GenCh1(self.fontE, val[i+2])
   return self.img

def generate(self, text):
   if len(text) == 7:
     fg = self.draw(text)  # decode(encoding="utf-8")
     fg = cv.bitwise_not(fg)
     com = cv.bitwise_or(fg, self.bg)
     com = rot(com, r(60)-30, com.shape,30)
     com = rotRandrom(com, 10, (com.shape[1], com.shape[0]))
     com = tfactor(com)
     com = random_envirment(com, self.noplates_path)
     com = AddGauss(com, 1+r(4))
     com = addNoise(com)
     return com

@staticmethod
 def genPlateString(pos, val):
   """
 生成车牌string,存为图片
   生成车牌list,存为label
   """
   plateStr = ""
   plateList=[]
   box = [0, 0, 0, 0, 0, 0, 0]
   if pos != -1:
     box[pos] = 1
   for unit, cpos in zip(box, range(len(box))):
     if unit == 1:
       plateStr += val
       plateList.append(val)
     else:
       if cpos == 0:
         plateStr += chars[r(31)]
         plateList.append(plateStr)
       elif cpos == 1:
         plateStr += chars[41 + r(24)]
         plateList.append(plateStr)
       else:
         plateStr += chars[31 + r(34)]
         plateList.append(plateStr)
   plate = [plateList[0]]
   b = [plateList[i][-1] for i in range(len(plateList))]
   plate.extend(b[1:7])
   return plateStr, plate

@staticmethod
 def genBatch(batchsize, outputPath, size):
   """
   将生成的车牌图片写入文件夹,对应的label写入label.txt
   :param batchsize: 批次大小
   :param outputPath: 输出图像的保存路径
   :param size: 输出图像的尺寸
   :return: None
   """
   if not os.path.exists(outputPath):
     os.mkdir(outputPath)
   outfile = open('data\\plate\\label.txt', 'w', encoding='utf-8')
   for i in range(batchsize):
     plateStr, plate = G.genPlateString(-1, -1)
     # print(plateStr, plate)
     img = G.generate(plateStr)
     img = cv.resize(img, size)
     cv.imwrite(outputPath + "\\" + str(i).zfill(2) + ".jpg", img)
     outfile.write(str(plate) + "\n")

if __name__ == '__main__':
 G = GenPlate("data\\font\\platech.ttf", 'data\\font\\platechar.ttf', "data\\NoPlates")
 G.genBatch(101, 'data\\plate', (272, 72))

生成的车牌图像尺寸尽量不要超过300,本次尺寸选取:272 * 72

生成车牌所需文件:

  • 字体文件:中文‘platech.ttf',英文及数字‘platechar.ttf'

  • 背景图:来源于不含车牌的车辆裁剪图片

  • 车牌(蓝底):template.bmp

  • 噪声图像:smu2.jpg

车牌生成后保存至plate文件夹,示例如下:

Python+Tensorflow+CNN实现车牌识别的示例代码

三、数据导入


from genplate import *
import matplotlib.pyplot as plt

# 产生用于训练的数据
class OCRIter:
 def __init__(self, batch_size, width, height):
   super(OCRIter, self).__init__()
   self.genplate = GenPlate("data\\font\\platech.ttf", 'data\\font\\platechar.ttf', "data\\NoPlates")
   self.batch_size = batch_size
   self.height = height
   self.width = width

def iter(self):
   data = []
   label = []
   for i in range(self.batch_size):
     img, num = self.gen_sample(self.genplate, self.width, self.height)
     data.append(img)
     label.append(num)
   return np.array(data), np.array(label)

@staticmethod
 def rand_range(lo, hi):
   return lo + r(hi - lo)

def gen_rand(self):
   name = ""
   label = list([])
   label.append(self.rand_range(0, 31))  #产生车牌开头32个省的标签
   label.append(self.rand_range(41, 65))  #产生车牌第二个字母的标签
   for i in range(5):
     label.append(self.rand_range(31, 65))  #产生车牌后续5个字母的标签
   name += chars[label[0]]
   name += chars[label[1]]
   for i in range(5):
     name += chars[label[i+2]]
   return name, label

def gen_sample(self, genplate, width, height):
   num, label = self.gen_rand()
   img = genplate.generate(num)
   img = cv.resize(img, (height, width))
   img = np.multiply(img, 1/255.0)
   return img, label    #返回的label为标签,img为车牌图像

'''
# 测试代码
O = OCRIter(2, 272, 72)
img, lbl = O.iter()
for im in img:
 plt.imshow(im, cmap='gray')
 plt.show()
print(img.shape)
print(lbl)
'''

四、CNN模型构建


import tensorflow as tf

def cnn_inference(images, keep_prob):
 W_conv = {
   'conv1': tf.Variable(tf.random.truncated_normal([3, 3, 3, 32],
                           stddev=0.1)),
   'conv2': tf.Variable(tf.random.truncated_normal([3, 3, 32, 32],
                           stddev=0.1)),
   'conv3': tf.Variable(tf.random.truncated_normal([3, 3, 32, 64],
                           stddev=0.1)),
   'conv4': tf.Variable(tf.random.truncated_normal([3, 3, 64, 64],
                           stddev=0.1)),
   'conv5': tf.Variable(tf.random.truncated_normal([3, 3, 64, 128],
                           stddev=0.1)),
   'conv6': tf.Variable(tf.random.truncated_normal([3, 3, 128, 128],
                           stddev=0.1)),
   'fc1_1': tf.Variable(tf.random.truncated_normal([5*30*128, 65],
                           stddev=0.01)),
   'fc1_2': tf.Variable(tf.random.truncated_normal([5*30*128, 65],
                           stddev=0.01)),
   'fc1_3': tf.Variable(tf.random.truncated_normal([5*30*128, 65],
                           stddev=0.01)),
   'fc1_4': tf.Variable(tf.random.truncated_normal([5*30*128, 65],
                           stddev=0.01)),
   'fc1_5': tf.Variable(tf.random.truncated_normal([5*30*128, 65],
                           stddev=0.01)),
   'fc1_6': tf.Variable(tf.random.truncated_normal([5*30*128, 65],
                           stddev=0.01)),
   'fc1_7': tf.Variable(tf.random.truncated_normal([5*30*128, 65],
                           stddev=0.01)),
   }

b_conv = {
   'conv1': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[32])),
   'conv2': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[32])),
   'conv3': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[64])),
   'conv4': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[64])),
   'conv5': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[128])),
   'conv6': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[128])),
   'fc1_1': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[65])),
   'fc1_2': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[65])),
   'fc1_3': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[65])),
   'fc1_4': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[65])),
   'fc1_5': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[65])),
   'fc1_6': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[65])),
   'fc1_7': tf.Variable(tf.constant(0.1, dtype=tf.float32,
                    shape=[65])),
   }

# 第1层卷积层
 conv1 = tf.nn.conv2d(images, W_conv['conv1'], strides=[1,1,1,1], padding='VALID')
 conv1 = tf.nn.bias_add(conv1, b_conv['conv1'])
 conv1 = tf.nn.relu(conv1)

# 第2层卷积层
 conv2 = tf.nn.conv2d(conv1, W_conv['conv2'], strides=[1,1,1,1], padding='VALID')
 conv2 = tf.nn.bias_add(conv2, b_conv['conv2'])
 conv2 = tf.nn.relu(conv2)
 # 第1层池化层
 pool1 = tf.nn.max_pool2d(conv2, ksize=[1,2,2,1], strides=[1,2,2,1], padding='VALID')

# 第3层卷积层
 conv3 = tf.nn.conv2d(pool1, W_conv['conv3'], strides=[1,1,1,1], padding='VALID')
 conv3 = tf.nn.bias_add(conv3, b_conv['conv3'])
 conv3 = tf.nn.relu(conv3)

# 第4层卷积层
 conv4 = tf.nn.conv2d(conv3, W_conv['conv4'], strides=[1,1,1,1], padding='VALID')
 conv4 = tf.nn.bias_add(conv4, b_conv['conv4'])
 conv4 = tf.nn.relu(conv4)
 # 第2层池化层
 pool2 = tf.nn.max_pool2d(conv4, ksize=[1,2,2,1], strides=[1,2,2,1], padding='VALID')

# 第5层卷积层
 conv5 = tf.nn.conv2d(pool2, W_conv['conv5'], strides=[1,1,1,1], padding='VALID')
 conv5 = tf.nn.bias_add(conv5, b_conv['conv5'])
 conv5 = tf.nn.relu(conv5)

# 第4层卷积层
 conv6 = tf.nn.conv2d(conv5, W_conv['conv6'], strides=[1,1,1,1], padding='VALID')
 conv6 = tf.nn.bias_add(conv6, b_conv['conv6'])
 conv6 = tf.nn.relu(conv6)
 # 第3层池化层
 pool3 = tf.nn.max_pool2d(conv6, ksize=[1,2,2,1], strides=[1,2,2,1], padding='VALID')

#第1_1层全连接层
 # print(pool3.shape)
 reshape = tf.reshape(pool3, [-1, 5 * 30 * 128])
 fc1 = tf.nn.dropout(reshape, keep_prob)
 fc1_1 = tf.add(tf.matmul(fc1, W_conv['fc1_1']), b_conv['fc1_1'])

#第1_2层全连接层
 fc1_2 = tf.add(tf.matmul(fc1, W_conv['fc1_2']), b_conv['fc1_2'])

#第1_3层全连接层
 fc1_3 = tf.add(tf.matmul(fc1, W_conv['fc1_3']), b_conv['fc1_3'])

#第1_4层全连接层
 fc1_4 = tf.add(tf.matmul(fc1, W_conv['fc1_4']), b_conv['fc1_4'])

#第1_5层全连接层
 fc1_5 = tf.add(tf.matmul(fc1, W_conv['fc1_5']), b_conv['fc1_5'])

#第1_6层全连接层
 fc1_6 = tf.add(tf.matmul(fc1, W_conv['fc1_6']), b_conv['fc1_6'])

#第1_7层全连接层
 fc1_7 = tf.add(tf.matmul(fc1, W_conv['fc1_7']), b_conv['fc1_7'])

return fc1_1, fc1_2, fc1_3, fc1_4, fc1_5, fc1_6, fc1_7

def calc_loss(logit1, logit2, logit3, logit4, logit5, logit6, logit7, labels):
 labels = tf.convert_to_tensor(labels, tf.int32)

loss1 = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
   logits=logit1, labels=labels[:, 0]))
 tf.compat.v1.summary.scalar('loss1', loss1)

loss2 = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
   logits=logit2, labels=labels[:, 1]))
 tf.compat.v1.summary.scalar('loss2', loss2)

loss3 = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
   logits=logit3, labels=labels[:, 2]))
 tf.compat.v1.summary.scalar('loss3', loss3)

loss4 = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
   logits=logit4, labels=labels[:, 3]))
 tf.compat.v1.summary.scalar('loss4', loss4)

loss5 = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
   logits=logit5, labels=labels[:, 4]))
 tf.compat.v1.summary.scalar('loss5', loss5)

loss6 = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
   logits=logit6, labels=labels[:, 5]))
 tf.compat.v1.summary.scalar('loss6', loss6)

loss7 = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
   logits=logit7, labels=labels[:, 6]))
 tf.compat.v1.summary.scalar('loss7', loss7)

return loss1, loss2, loss3, loss4, loss5, loss6, loss7

def train_step(loss1, loss2, loss3, loss4, loss5, loss6, loss7, learning_rate):
 optimizer1 = tf.compat.v1.train.AdamOptimizer(learning_rate=learning_rate)
 train_op1 = optimizer1.minimize(loss1)

optimizer2 = tf.compat.v1.train.AdamOptimizer(learning_rate=learning_rate)
 train_op2 = optimizer2.minimize(loss2)

optimizer3 = tf.compat.v1.train.AdamOptimizer(learning_rate=learning_rate)
 train_op3 = optimizer3.minimize(loss3)

optimizer4 = tf.compat.v1.train.AdamOptimizer(learning_rate=learning_rate)
 train_op4 = optimizer4.minimize(loss4)

optimizer5 = tf.compat.v1.train.AdamOptimizer(learning_rate=learning_rate)
 train_op5 = optimizer5.minimize(loss5)

optimizer6 = tf.compat.v1.train.AdamOptimizer(learning_rate=learning_rate)
 train_op6 = optimizer6.minimize(loss6)

optimizer7 = tf.compat.v1.train.AdamOptimizer(learning_rate=learning_rate)
 train_op7 = optimizer7.minimize(loss7)

return train_op1, train_op2, train_op3, train_op4, train_op5, train_op6, train_op7

def pred_model(logit1, logit2, logit3, logit4, logit5, logit6, logit7, labels):
 labels = tf.convert_to_tensor(labels, tf.int32)
 labels = tf.reshape(tf.transpose(labels), [-1])
 logits = tf.concat([logit1, logit2, logit3, logit4, logit5, logit6, logit7], 0)
 prediction = tf.nn.in_top_k(logits, labels, 1)
 accuracy = tf.reduce_mean(tf.cast(prediction, tf.float32))
 tf.compat.v1.summary.scalar('accuracy', accuracy)
 return accuracy

五、模型训练


import os
import time
import datetime
import numpy as np
import tensorflow as tf
from input_data import OCRIter
import model

os.environ["TF_CPP_MIN_LOG_LEVEL"] = '3'

img_h = 72
img_w = 272
num_label = 7
batch_size = 32
epoch = 10000
learning_rate = 0.0001

logs_path = 'logs\\1005'
model_path = 'saved_model\\1005'

image_holder = tf.compat.v1.placeholder(tf.float32, [batch_size, img_h, img_w, 3])
label_holder = tf.compat.v1.placeholder(tf.int32, [batch_size, 7])
keep_prob = tf.compat.v1.placeholder(tf.float32)

def get_batch():
 data_batch = OCRIter(batch_size, img_h, img_w)
 image_batch, label_batch = data_batch.iter()
 return np.array(image_batch), np.array(label_batch)

logit1, logit2, logit3, logit4, logit5, logit6, logit7 = model.cnn_inference(
 image_holder, keep_prob)

loss1, loss2, loss3, loss4, loss5, loss6, loss7 = model.calc_loss(
 logit1, logit2, logit3, logit4, logit5, logit6, logit7, label_holder)

train_op1, train_op2, train_op3, train_op4, train_op5, train_op6, train_op7 = model.train_step(
 loss1, loss2, loss3, loss4, loss5, loss6, loss7, learning_rate)

accuracy = model.pred_model(logit1, logit2, logit3, logit4, logit5, logit6, logit7, label_holder)

input_image=tf.compat.v1.summary.image('input', image_holder)

summary_op = tf.compat.v1.summary.merge(tf.compat.v1.get_collection(tf.compat.v1.GraphKeys.SUMMARIES))

init_op = tf.compat.v1.global_variables_initializer()

with tf.compat.v1.Session() as sess:
 sess.run(init_op)

train_writer = tf.compat.v1.summary.FileWriter(logs_path, sess.graph)
 saver = tf.compat.v1.train.Saver()

start_time1 = time.time()
 for step in range(epoch):
   # 生成车牌图像以及标签数据
   img_batch, lbl_batch = get_batch()

start_time2 = time.time()
   time_str = datetime.datetime.now().isoformat()

feed_dict = {image_holder:img_batch, label_holder:lbl_batch, keep_prob:0.6}
   _1, _2, _3, _4, _5, _6, _7, ls1, ls2, ls3, ls4, ls5, ls6, ls7, acc = sess.run(
     [train_op1, train_op2, train_op3, train_op4, train_op5, train_op6, train_op7,
      loss1, loss2, loss3, loss4, loss5, loss6, loss7, accuracy], feed_dict)
   summary_str = sess.run(summary_op, feed_dict)
   train_writer.add_summary(summary_str,step)
   duration = time.time() - start_time2
   loss_total = ls1 + ls2 + ls3 + ls4 + ls5 + ls6 + ls7
   if step % 10 == 0:
     sec_per_batch = float(duration)
     print('%s: Step %d, loss_total = %.2f, acc = %.2f%%, sec/batch = %.2f' %
       (time_str, step, loss_total, acc * 100, sec_per_batch))
   if step % 5000 == 0 or (step + 1) == epoch:
     checkpoint_path = os.path.join(model_path,'model.ckpt')
     saver.save(sess, checkpoint_path, global_step=step)
 end_time = time.time()
 print("Training over. It costs {:.2f} minutes".format((end_time - start_time1) / 60))

六、训练结果展示

训练参数:
batch_size = 32
epoch = 10000
learning_rate = 0.0001
在tensorboard中查看训练过程
accuracy :

Python+Tensorflow+CNN实现车牌识别的示例代码accuracy

曲线在epoch = 10000左右时达到收敛,最终精确度在94%左右

loss :
Python+Tensorflow+CNN实现车牌识别的示例代码

Python+Tensorflow+CNN实现车牌识别的示例代码

Python+Tensorflow+CNN实现车牌识别的示例代码

以上三张分别是loss1,loss2, loss7的曲线图像,一号位字符是省份简称,识别相对字母数字较难,loss1=0.08左右,二号位字符是字母,loss2稳定在0.001左右,但是随着字符往后,loss值也将越来越大,7号位字符loss7稳定在0.6左右。

七、预测单张车牌


import os
import cv2 as cv
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
from PIL import Image
import model

os.environ["TF_CPP_MIN_LOG_LEVEL"] = '3' # 只显示 Error

index = {"京": 0, "沪": 1, "津": 2, "渝": 3, "冀": 4, "晋": 5, "蒙": 6, "辽": 7, "吉": 8, "黑": 9,
      "苏": 10, "浙": 11, "皖": 12, "闽": 13, "赣": 14, "鲁": 15, "豫": 16, "鄂": 17, "湘": 18, "粤": 19,
      "桂": 20, "琼": 21, "川": 22, "贵": 23, "云": 24, "藏": 25, "陕": 26, "甘": 27, "青": 28, "宁": 29,
      "新": 30, "0": 31, "1": 32, "2": 33, "3": 34, "4": 35, "5": 36, "6": 37, "7": 38, "8": 39,
      "9": 40, "A": 41, "B": 42, "C": 43, "D": 44, "E": 45, "F": 46, "G": 47, "H": 48, "J": 49,
      "K": 50, "L": 51, "M": 52, "N": 53, "P": 54, "Q": 55, "R": 56, "S": 57, "T": 58, "U": 59,
      "V": 60, "W": 61, "X": 62, "Y": 63, "Z": 64}

chars = ["京", "沪", "津", "渝", "冀", "晋", "蒙", "辽", "吉", "黑",
      "苏", "浙", "皖", "闽", "赣", "鲁", "豫", "鄂", "湘", "粤",
      "桂", "琼", "川", "贵", "云", "藏", "陕", "甘", "青", "宁",
      "新", "0", "1", "2", "3", "4", "5", "6", "7", "8",
      "9", "A", "B", "C", "D", "E", "F", "G", "H", "J",
      "K", "L", "M", "N", "P", "Q", "R", "S", "T", "U",
      "V", "W", "X", "Y", "Z"]

def get_one_image(test):
 """ 随机获取单张车牌图像 """
 n = len(test)
 rand_num =np.random.randint(0,n)
 img_dir = test[rand_num]
 image_show = Image.open(img_dir)
 plt.imshow(image_show)  # 显示车牌图片
 image = cv.imread(img_dir)
 image = image.reshape(72, 272, 3)
 image = np.multiply(image, 1 / 255.0)
 return image

batch_size = 1
x = tf.compat.v1.placeholder(tf.float32, [batch_size, 72, 272, 3])
keep_prob = tf.compat.v1.placeholder(tf.float32)

test_dir = 'data\\plate\\'
test_image = []
for file in os.listdir(test_dir):
 test_image.append(test_dir + file)
test_image = list(test_image)

image_array = get_one_image(test_image)

logit1, logit2, logit3, logit4, logit5, logit6, logit7 = model.cnn_inference(x, keep_prob)

model_path = 'saved_model\\1005'

saver = tf.compat.v1.train.Saver()

with tf.compat.v1.Session() as sess:
 print ("Reading checkpoint...")
 ckpt = tf.train.get_checkpoint_state(model_path)
 if ckpt and ckpt.model_checkpoint_path:
   global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
   saver.restore(sess, ckpt.model_checkpoint_path)
   print('Loading success, global_step is %s' % global_step)
 else:
   print('No checkpoint file found')

pre1, pre2, pre3, pre4, pre5, pre6, pre7 = sess.run(
   [logit1, logit2, logit3, logit4, logit5, logit6, logit7],
   feed_dict={x:image_array, keep_prob:1.0})
 prediction = np.reshape(np.array([pre1, pre2, pre3, pre4, pre5, pre6, pre7]), [-1, 65])

max_index = np.argmax(prediction, axis=1)
 print(max_index)
 line = ''
 result = np.array([])
 for i in range(prediction.shape[0]):
   if i == 0:
     result = np.argmax(prediction[i][0:31])
   if i == 1:
     result = np.argmax(prediction[i][41:65]) + 41
   if i > 1:
     result = np.argmax(prediction[i][31:65]) + 31
   line += chars[result]+" "
 print ('predicted: ' + line)
plt.show()

随机测试20张车牌,18张预测正确,2张预测错误,从最后两幅预测错误的图片可以看出,模型对相似字符以及遮挡字符识别成功率仍有待提高。测试结果部分展示如下:

Python+Tensorflow+CNN实现车牌识别的示例代码

Python+Tensorflow+CNN实现车牌识别的示例代码

Python+Tensorflow+CNN实现车牌识别的示例代码

Python+Tensorflow+CNN实现车牌识别的示例代码

Python+Tensorflow+CNN实现车牌识别的示例代码

Python+Tensorflow+CNN实现车牌识别的示例代码

八、总结

本次构建的CNN模型较为简单,只有6卷积层+3池化层+1全连接层,可以通过增加模型深度以及每层之间的神经元数量来优化模型,提高识别的准确率。此次训练数据集来源于自动生成的车牌,由于真实的车牌图像与生成的车牌图像在噪声干扰上有所区分,所以识别率上会有所出入。如果使用真实的车牌数据集,需要对车牌进行滤波、均衡化、腐蚀、矢量量化等预处理方法。

来源:https://blog.csdn.net/weixin_43397593/article/details/102372100

0
投稿

猜你喜欢

手机版 网络编程 asp之家 www.aspxhome.com