本文采用PaddleOCR开源项目实现渔船牌照识别。因开源数据集少，自行按规则生成1000张渔牌数据，按8:2划分训练集与测试集。经环境安装、预训练模型获取、数据集处理、模型训练等步骤，最终实现识别，虽因训练时长可能效果欠佳，但完成了基本流程。
☞☞☞AI 智能聊天, 问答助手, AI 智能搜索, 免费无限量使用 DeepSeek R1 模型☜☜☜
基于PaddleOCR渔船牌照识别
一、项目介绍
本文采用PaddleOCR开源项目进行渔船牌照识别，流程分为数据集构建、数据集处理、模型搭建与预测、推理等，由于开源渔船牌照数据集较少，本项目自行构建脚本生成1000多张渔船牌照图进行训练，最终实现渔船牌照识别。
二、安装环境
In [ ]
# !git clone https://gitee.com/paddlepaddle/PaddleOCR %cd PaddleOCR !git checkout -b release/2.4 remotes/origin/release/2.4
In [ ]
!pwd !pip install -r requirements.txt !pip install pillow --user !pip uninstall opencv-python -y --user !pip uninstall opencv-contrib-python -y --user !pip install opencv-python==4.2.0.32 --user !pip install --upgrade pip
获取预训练模型
选用PaddleOCR模型地址
In [ ]
# 获取预训练模型!wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_train.tar !tar -xf /home/aistudio/PaddleOCR/pretrain_models/en_number_mobile_v2.0_rec_slim_train.tar -C /home/aistudio/PaddleOCR/pretrain_models
数据集介绍
由于开源渔牌牌照数据集较少，因此本文选择按照渔牌规则自己生成渔船牌照数据，本次生成1000张渔牌数据，按8:2划分训练集与测试集。生成样例如下：
生成数据脚本参考：https://gitee.com/goalaaa/chinese_license_plate_generator
三、数据集处理
- 解压数据集
- 数据集拆分
- 格式转换
In [ ]
# 解压数据集!unzip /home/aistudio/data/data201866/fish_dataset.zip -d /home/aistudio/data/fish_data
In [ ]
# 训练/测试数据清洗path2 = '/home/aistudio/data/fish_data'
# 数据准备# 格式示例: 1016_752_1.jpg I'm Li Hua,chairman of the Student Union from with open(f'/home/aistudio/data/label.txt') as f:
lines = f.readlines() # 9000用于训练, 1000用于测试
with open(f'/home/aistudio/data/train.txt', 'w') as f1: with open(f'/home/aistudio/data/test.txt', 'w') as f2: for index, line in enumerate(lines):
firstSpaceIndex = line.find(' ')
line2 = line[0:firstSpaceIndex] + '\t' + line[firstSpaceIndex+1:]
if index < 800:
f1.write(line2) if index >= 800:
f2.write(line2)print("数据处理完成")
格式转换
生成用于识别的txt格式
云云湘渔65699.jpg 云云湘渔65699云云葫渔36057.jpg 云云葫渔36057云吉桂渔12572.jpg 云吉桂渔12572云宁云渔83850.jpg 云宁云渔83850云川嘉渔30711.jpg 云川嘉渔30711云川闽渔47501.jpg 云川闽渔47501云川黑渔84624.jpg 云川黑渔84624云新津渔90182.jpg 云新津渔90182云新浙渔03236.jpg 云新浙渔03236云晋豫渔69022.jpg 云晋豫渔69022云桂桂渔07075.jpg 云桂桂渔07075云沪渝渔09603.jpg 云沪渝渔09603云沪渝渔31067.jpg 云沪渝渔31067云浙津渔57087.jpg 云浙津渔57087云渝渝渔29063.jpg 云渝渝渔29063云湘鄂渔35418.jpg 云湘鄂渔35418云烟闽渔62305.jpg 云烟闽渔62305云甘云渔45805.jpg 云甘云渔45805
四、模型训练
In [ ]
# 开始训练%cd /home/aistudio/PaddleOCR !python tools/train.py -c /home/aistudio/work/rec_en_number_lite_train.yml# 等待训练是不是很无聊?让它先跑着,看看下一步吧 :)
In [ ]
# 开始训练%cd /home/aistudio/PaddleOCR !python tools/train.py -c /home/aistudio/work/rec_en_number_lite_train_new.yml
继续训练
- 我们在训练过程中经常会遇到各种问题导致训练中断,这个时候如果不想从0开始,就需要继续训练了
- 继续训练的本质是每训练一段时间,就保存一次权重,这样就可以加载最后一次(或者最好)的权重进行训练了
In [ ]
# 这时,上面应该跑了几个epoch了吧,你现在可以把上面的训练停了# 如果上面训练中断了,并且不想再重新开始训练,可以执行本段代码继续上次训练!python tools/train.py -c /home/aistudio/work/rec_en_number_lite_train_new.yml -o Global.checkpoints=/home/aistudio/PaddleOCR/output/rec_en_number_lite_new/latest
In [10]
# 图片显示import matplotlib.pyplot as pltimport cv2def imshow(img_path):
im = cv2.imread(img_path)
plt.imshow(im )# 随便显示一张图片path2 = '/home/aistudio/data/fish_data/fish_dataset/川辽冀渔96794.jpg'imshow(path2)
In [20]
# 预测,这里使用当前的训练结果来预测# PS: 由于训练时长问题,效果可能不理想%cd PaddleOCR
!python tools/infer_rec.py -c /home/aistudio/work/rec_en_number_lite_train_new.yml \
-o Global.infer_img="/home/aistudio/data/fish_data/fish_dataset/川辽冀渔96794.jpg" \
Global.pretrained_model="/home/aistudio/PaddleOCR/output/rec_en_number_lite_new/latest"# 显示该图片path2 = '/home/aistudio/data/fish_data//fish_dataset/川辽冀渔96794.jpg'imshow(path2)
[Errno 2] No such file or directory: 'PaddleOCR' /home/aistudio/PaddleOCR /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:26: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information. Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations def convert_to_list(value, n, name, dtype=np.int): [2023/03/24 15:00:11] root INFO: Architecture : [2023/03/24 15:00:11] root INFO: Backbone : [2023/03/24 15:00:11] root INFO: model_name : small [2023/03/24 15:00:11] root INFO: name : MobileNetV3 [2023/03/24 15:00:11] root INFO: scale : 0.5 [2023/03/24 15:00:11] root INFO: small_stride : [1, 2, 2, 2] [2023/03/24 15:00:11] root INFO: Head : [2023/03/24 15:00:11] root INFO: fc_decay : 1e-05 [2023/03/24 15:00:11] root INFO: name : CTCHead [2023/03/24 15:00:11] root INFO: Neck : [2023/03/24 15:00:11] root INFO: encoder_type : rnn [2023/03/24 15:00:11] root INFO: hidden_size : 48 [2023/03/24 15:00:11] root INFO: name : SequenceEncoder [2023/03/24 15:00:11] root INFO: Transform : None [2023/03/24 15:00:11] root INFO: algorithm : CRNN [2023/03/24 15:00:11] root INFO: model_type : rec [2023/03/24 15:00:11] root INFO: Eval : [2023/03/24 15:00:11] root INFO: dataset : [2023/03/24 15:00:11] root INFO: data_dir : /home/aistudio/data/fish_data [2023/03/24 15:00:11] root INFO: label_file_list : ['/home/aistudio/data/test.txt'] [2023/03/24 15:00:11] root INFO: name : SimpleDataSet [2023/03/24 15:00:11] root INFO: transforms : [2023/03/24 15:00:11] root INFO: DecodeImage : [2023/03/24 15:00:11] root INFO: channel_first : False [2023/03/24 15:00:11] root INFO: img_mode : BGR [2023/03/24 15:00:11] root INFO: CTCLabelEncode : None [2023/03/24 15:00:11] root INFO: RecResizeImg : [2023/03/24 15:00:11] root INFO: image_shape : [3, 32, 320] [2023/03/24 15:00:11] root INFO: KeepKeys : [2023/03/24 15:00:11] root INFO: keep_keys : ['image', 'label', 'length'] [2023/03/24 15:00:11] root INFO: loader : [2023/03/24 15:00:11] root INFO: batch_size_per_card : 8 [2023/03/24 15:00:11] root INFO: drop_last : False [2023/03/24 15:00:11] root INFO: num_workers : 8 [2023/03/24 15:00:11] root INFO: shuffle : False [2023/03/24 15:00:11] root INFO: Global : [2023/03/24 15:00:11] root INFO: cal_metric_during_train : True [2023/03/24 15:00:11] root INFO: character_dict_path : ppocr/utils/EN_symbol_dict.txt [2023/03/24 15:00:11] root INFO: checkpoints : None [2023/03/24 15:00:11] root INFO: debug : False [2023/03/24 15:00:11] root INFO: distributed : False [2023/03/24 15:00:11] root INFO: epoch_num : 200 [2023/03/24 15:00:11] root INFO: eval_batch_step : [0, 100] [2023/03/24 15:00:11] root INFO: infer_img : /home/aistudio/data/fish_data/fish_dataset/川辽冀渔96794.jpg [2023/03/24 15:00:11] root INFO: infer_mode : False [2023/03/24 15:00:11] root INFO: log_smooth_window : 20 [2023/03/24 15:00:11] root INFO: max_text_length : 25 [2023/03/24 15:00:11] root INFO: pretrained_model : /home/aistudio/PaddleOCR/output/rec_en_number_lite_new/latest [2023/03/24 15:00:11] root INFO: print_batch_step : 10 [2023/03/24 15:00:11] root INFO: save_epoch_step : 3 [2023/03/24 15:00:11] root INFO: save_inference_dir : None [2023/03/24 15:00:11] root INFO: save_model_dir : ./output/rec_en_number_lite_new [2023/03/24 15:00:11] root INFO: use_gpu : True [2023/03/24 15:00:11] root INFO: use_space_char : True [2023/03/24 15:00:11] root INFO: use_visualdl : True [2023/03/24 15:00:11] root INFO: Loss : [2023/03/24 15:00:11] root INFO: name : CTCLoss [2023/03/24 15:00:11] root INFO: Metric : [2023/03/24 15:00:11] root INFO: main_indicator : acc [2023/03/24 15:00:11] root INFO: name : RecMetric [2023/03/24 15:00:11] root INFO: Optimizer : [2023/03/24 15:00:11] root INFO: beta1 : 0.9 [2023/03/24 15:00:11] root INFO: beta2 : 0.999 [2023/03/24 15:00:11] root INFO: lr : [2023/03/24 15:00:11] root INFO: learning_rate : 0.005 [2023/03/24 15:00:11] root INFO: name : Cosine [2023/03/24 15:00:11] root INFO: name : Adam [2023/03/24 15:00:11] root INFO: regularizer : [2023/03/24 15:00:11] root INFO: factor : 1e-05 [2023/03/24 15:00:11] root INFO: name : L2 [2023/03/24 15:00:11] root INFO: PostProcess : [2023/03/24 15:00:11] root INFO: name : CTCLabelDecode [2023/03/24 15:00:11] root INFO: Train : [2023/03/24 15:00:11] root INFO: dataset : [2023/03/24 15:00:11] root INFO: data_dir : /home/aistudio/data/fish_data [2023/03/24 15:00:11] root INFO: label_file_list : ['/home/aistudio/data/train.txt'] [2023/03/24 15:00:11] root INFO: name : SimpleDataSet [2023/03/24 15:00:11] root INFO: transforms : [2023/03/24 15:00:11] root INFO: DecodeImage : [2023/03/24 15:00:11] root INFO: channel_first : False [2023/03/24 15:00:11] root INFO: img_mode : BGR [2023/03/24 15:00:11] root INFO: RecAug : None [2023/03/24 15:00:11] root INFO: CTCLabelEncode : None [2023/03/24 15:00:11] root INFO: RecResizeImg : [2023/03/24 15:00:11] root INFO: image_shape : [3, 32, 320] [2023/03/24 15:00:11] root INFO: KeepKeys : [2023/03/24 15:00:11] root INFO: keep_keys : ['image', 'label', 'length'] [2023/03/24 15:00:11] root INFO: loader : [2023/03/24 15:00:11] root INFO: batch_size_per_card : 64 [2023/03/24 15:00:11] root INFO: drop_last : True [2023/03/24 15:00:11] root INFO: num_workers : 4 [2023/03/24 15:00:11] root INFO: shuffle : True [2023/03/24 15:00:11] root INFO: profiler_options : None [2023/03/24 15:00:11] root INFO: train with paddle 2.0.2 and device CUDAPlace(0) W0324 15:00:11.964440 6307 device_context.cc:362] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.2, Runtime API Version: 10.1 W0324 15:00:11.970218 6307 device_context.cc:372] device: 0, cuDNN Version: 7.6. [2023/03/24 15:00:15] root INFO: load pretrain successful from /home/aistudio/PaddleOCR/output/rec_en_number_lite_new/latest [2023/03/24 15:00:15] root INFO: infer_img: /home/aistudio/data/fish_data/fish_dataset/川辽冀渔96794.jpg [2023/03/24 15:00:15] root INFO: result: 川辽冀96794 0.93036735 [2023/03/24 15:00:15] root INFO: success!
