前言

上一章讲了检测模型的实现，这一章将一下识别模型的实现，PaddleOCR提供的通用字符集有6623个，但是车牌号就不需要这么多，我们只需要识别这些车牌，蓝牌、黄牌（单行）、绿牌、大型新能源（黄绿）、领使馆车牌、警牌、武警牌（单行）、军牌（单行）、港澳出入境车牌，只用提取极少部分字符即可，下面我们来接着将代码如何实现的。

一、银行卡识别模型实现

1、创建字符集

在ppocr/utils/目录中创建一份只有数字的字符集ppocr_keys_car.txt，71个字符：

川鄂甘赣贵桂黑沪吉冀晋津京辽鲁蒙闽宁青琼陕苏皖湘新渝豫粤云藏浙澳港警学使领
0123456789
ABCDEFGHJKLMNPQRSTUVWXYZ

注意字母字符中不包含“O”，这与数字0在车牌上显示有歧义。

2、导入数据集

数据集可在上一篇文章中下载，目前收集了3万张左右的数据集，为了保证数据集字符平衡，我们就需要合成一部分数据集，这3万张中一半是真实一半是合成的

数据集部分切图，上面是真实的，下面是合成的：

?3、创建配置文件

复制一份yml文件

?文件内容如下：

Global:
  use_gpu: true
  epoch_num: 500
  log_smooth_window: 20
  print_batch_step: 10
  save_model_dir: ./output/rec_chinese_lite_v2.0_car
  save_epoch_step: 50
  # evaluation is run every 5000 iterations after the 4000th iteration
#  eval_batch_step: [0, 2000]
  eval_batch_step: 100
  cal_metric_during_train: True
#  pretrained_model: ./pretrained_models/ch_ppocr_mobile_v2.0_rec_pre/best_accuracy
  pretrained_model:
  checkpoints:
  save_inference_dir: ./inference/rec_chinese_lite_v2.0_car
  use_visualdl: False
  infer_img: doc/imgs_words/ch/word_1.jpg
  # for data or label process
  character_dict_path: ppocr/utils/ppocr_keys_car.txt
  character_type: ch
  max_text_length: 25
  infer_mode: False
  use_space_char: False
  save_res_path: ./output/rec/predicts_chinese_lite_v2.0.txt


Optimizer:
  name: Adam
  beta1: 0.9
  beta2: 0.999
  lr:
    name: Cosine
    learning_rate: 0.001
  regularizer:
    name: 'L2'
    factor: 0.00001

Architecture:
  model_type: rec
  algorithm: CRNN
  Transform:
  Backbone:
    name: MobileNetV3
    scale: 0.5
    model_name: small
    small_stride: [1, 2, 2, 2]
  Neck:
    name: SequenceEncoder
    encoder_type: rnn
    hidden_size: 48
  Head:
    name: CTCHead
    fc_decay: 0.00001

Loss:
  name: CTCLoss

PostProcess:
  name: CTCLabelDecode

Metric:
  name: RecMetric
  main_indicator: acc

Train:
  dataset:
    name: SimpleDataSet
    data_dir: ./train_data/
    label_file_list:
      - "./train_data/car/car1/rec_gt_train.txt"
      - "./train_data/car/car2/rec_gt_train.txt"
      - "./train_data/car/car3/rec_gt_train.txt"
    ratio_list: [1.0,1.0,0.3]
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - RecAug: 
      - CTCLabelEncode: # Class handling label
      - RecResizeImg:
          image_shape: [3, 32, 320]
      - KeepKeys:
          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
  loader:
    shuffle: True
    batch_size_per_card: 32
    drop_last: True
    num_workers: 8

Eval:
  dataset:
    name: SimpleDataSet
    data_dir: ./train_data/
    label_file_list:
      - "./train_data/car/car1/rec_gt_test.txt"
      - "./train_data/car/car2/rec_gt_test.txt"
      - "./train_data/car/car3/rec_gt_test.txt"
    ratio_list: [1.0,1.0,0.3]
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - CTCLabelEncode: # Class handling label
      - RecResizeImg:
          image_shape: [3, 32, 320]
      - KeepKeys:
          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
  loader:
    shuffle: False
    drop_last: False
    batch_size_per_card: 32
    num_workers: 8

4、执行命令进行GPU训练

python -m paddle.distributed.launch --gpus 0 tools/train.py -o Global.use_visualdl=True -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0_car.yml

注意显存问题，需要大于4G的显存，2G显存会溢出

5、显示运行图

visualdl --logdir=output/rec_chinese_lite_v2.0_car/vdl --port 8081

识别模型效果最高能到达：acc：0.99

6、导出模型

python tools/export_model.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0_car.yml
 -o Global.pretrained_model="./output/rec_chinese_lite_v2.0_car/best_accuracy" Global.save
_inference_dir="./output/rec_chinese_lite_v2.0_car/"

检测模型和识别模型串联验证：

python tools/infer/predict_system.py --det_algorithm="DB" --det_model_dir="./inference/ch_db_mv3_car/" --rec_model_dir=./inference/rec_chinese_lite_v2.0_car/ --image_dir="C:\Users\YY007\Desktop\car\1.jpg" --rec_char_dict_path="./ppocr/utils/ppocr_keys_car.txt" --use_gpu=True --use_angle_cls=False --det_db_unclip_ratio=1.8