Halcon 进行基于深度学习的异常检测

Halcon 在 2022年5月对外宣布更新了异常检测深度学习算法，本文记录使用方法。

简介

Halcon 深度学习异常检测可以用若干没有瑕疵的数据训练模型，用于检测出现问题 (异常) 的数据。

官网简介

核心流程

1. 准备数据

   需要准备一个包含数据的字典列表，每个字典表示一张数据图像，字典中 image 字段下存放图像，其他字段记录该图像配套的信息

   过程中对图像做需要的预处理

   数据集建议使用 MVTec 自己的异常检测数据集

   list_image_files (train_dir, 'default', [], ImageFiles)
   
   train_samples := []
   
   image_width := image_size[0]
   image_height := image_size[1]
   
   
   for Index := 0 to |ImageFiles|-1 by 1
       
       image_path := ImageFiles[Index]
       create_dict (sample_info)
       set_dict_tuple (sample_info, 'image_id', Index)
       set_dict_tuple (sample_info, 'anomaly_label', 'ok')
       set_dict_tuple (sample_info, 'anomaly_label_id', 0)
       read_image (Image, image_path)
       convert_image_type (Image, Image, 'real')
       
       if (Index < 1)
           image_sum := Image
       else
           add_image (Image, image_sum, image_sum, 1, 0)
       endif
   
   endfor
   
   sub_image (image_sum, image_sum, Imagezero, 1, |ImageFiles|)
   
   div_image (image_sum, Imagezero, ImageResult, 1, 0)
   
   count_channels (ImageResult, Channels)
   
   * 初始化均值和方差列表
   MeanValues := []
   DeviationValues := []
   
   if (Channels == 3)
   * 处理彩色图像
   decompose3(ImageResult, Image1, Image2, Image3)
   
   else
       Image1 := ImageResult
       Image2 := ImageResult
       Image3 := ImageResult
   endif
   
   gen_empty_obj (EmptyObject)
   
   concat_obj (EmptyObject, Image1, EmptyObject)
   concat_obj (EmptyObject, Image2, EmptyObject)
   concat_obj (EmptyObject, Image3, EmptyObject)
   
   * 计算每个通道的均值和标准差
   for ChannelIndex := 1 to 3 by 1
       select_obj (EmptyObject, ObjectSelected, ChannelIndex)
       intensity(ObjectSelected, ObjectSelected, Mean, Deviation)
       MeanValues := [MeanValues, Mean]
       DeviationValues := [DeviationValues, Deviation]
   endfor
   
   create_dict (data_param)
   set_dict_tuple (data_param, 'DeviationValues', DeviationValues)
   set_dict_tuple (data_param, 'MeanValues', MeanValues)
   
   
   for Index := 0 to |ImageFiles|-1 by 1
       
       image_path := ImageFiles[Index]
       create_dict (sample_info)
       set_dict_tuple (sample_info, 'image_id', Index)
       set_dict_tuple (sample_info, 'anomaly_label', 'ok')
       set_dict_tuple (sample_info, 'anomaly_label_id', 0)
       
       read_image (Image, image_path)
       convert_image_type (Image, Image, 'real')
       zoom_image_size (Image, Image, image_width, image_height, 'bilinear')
       
       count_channels (Image, Channels)
       
       if (Channels == 3)
           * 处理彩色图像
           decompose3(Image, Image1, Image2, Image3)
       else
           Image1 := Image
           Image2 := Image
           Image3 := Image
       endif
       
       gen_empty_obj (ImageScaled)
       
       Scale1 := 1.0 / DeviationValues[0]
       Shift1 := - Scale1 * MeanValues[0]
       scale_image (Image1, ImageScaled1, Scale1, Shift1)
       append_channel (ImageScaled, ImageScaled1, ImageScaled)
       
       Scale2 := 1.0 / DeviationValues[1]
       Shift2 := - Scale1 * MeanValues[1]
       scale_image (Image2, ImageScaled2, Scale2, Shift2)
       append_channel (ImageScaled, ImageScaled2, ImageScaled)
       
       Scale3 := 1.0 / DeviationValues[2]
       Shift3 := - Scale3 * MeanValues[2]
       scale_image (Image3, ImageScaled3, Scale3, Shift3)
       append_channel (ImageScaled, ImageScaled3, ImageScaled)
       
       gen_empty_obj (ImagesScaled)
       concat_obj (ImagesScaled, ImageScaled, ImagesScaled)
       
       set_dict_object (ImagesScaled, sample_info, 'image')
       
       train_samples :=[train_samples, sample_info]
   endfor
   
   
   return ()

2. 训练参数

   需要配置训练时的部分参数，主要包括提前停止阈值、数据使用率、训练轮数、正则项水平等

   TrainParamAnomaly := dict{error_threshold: 0.001}
   TrainParamAnomaly.MaxNumEpochs := 30
   TrainParamAnomaly.domain_ratio := 0.25
   TrainParamAnomaly.regularization_noise := 0.01

3. 初始化模型

   模型初始化，使用 Halcon 的模型加载算子加载预训练的异常检测模型

   ImageWidth := 256
   ImageHeight := 256
   
   Complexity := 15
   query_available_dl_devices (['runtime', 'id'], ['gpu', 0], DLDevice)
   
   read_dl_model ('initial_dl_anomaly_medium.hdl', DLModelHandle)
   set_dl_model_param (DLModelHandle, 'image_width', ImageWidth)
   set_dl_model_param (DLModelHandle, 'image_height', ImageHeight)
   set_dl_model_param (DLModelHandle, 'complexity', Complexity)
   set_dl_model_param (DLModelHandle, 'device', DLDevice)

4. 训练模型

   将上述数据用训练算子一起训练

   train_dl_model_anomaly_dataset (DLModelHandle, train_samples, TrainParamAnomaly, DLTrainResult)

5. 模型保存

   保存训练完成的模型

   model_file_name := result_home_dir + '/model.hdl'
   write_dl_model (DLModelHandle, model_file_name)

6. 模型推断

   加载测试数据，与训练好的模型一起送入推断算子

   apply_dl_model (DLModelHandle, test_dataset, [], DLResultBatch)

参考资料

• https://www.mvtec.com/cn/services-support/videos-tutorials/single-video/deep-learning-based-anomaly-detection-with-mvtec-halcon

文章链接：
https://www.zywvvd.com/notes/coding/halcon/halcon-anomaly-det/halcon-anomaly-det/