蔬菜水果训练和识别

这段代码是一个用于图像识别的脚本，它使用HALCON软件的命令和函数来实现对图像中的对象进行训练和识别。

dev_update_off ()
*

• Set DoTraining to true to force a new training
  DoTraining := false
• Get the image file names, the object names and IDs
  ImageDirectory := ‘food/vegetables’
  IdentifierFileName := ‘vegetables.sid’
  get_object_images (ImageDirectory, ObjectNames, ObjectIDs, ImageFileNames, TrainingObjectNames, TrainingObjectIDs, TrainingImageFileNames)

read_image (Image, TrainingImageFileNames[0])
dev_close_window ()
dev_open_window_fit_image (Image, 0, 0, -1, -1, WindowHandle)
set_display_font (WindowHandle, 14, ‘mono’, ‘true’, ‘false’)
*
file_exists (IdentifierFileName, FileExists)
if (FileExists and not DoTraining)
disp_message (WindowHandle, ‘Read the trained sample identifier …’, ‘window’, 12, 12, ‘black’, ‘true’)
read_sample_identifier (IdentifierFileName, SampleIdentifier)
else
disp_message (WindowHandle, ‘Create, prepare and train a new sample identifier’, ‘window’, 12, 12, ‘black’, ‘true’)
disp_continue_message (WindowHandle, ‘black’, ‘true’)
stop ()
* If the trained sample identifier is not available yet we create
* a new sample identifier. Because we have color images available
* we want to use the color in addition to the texture information
* for identification.
create_sample_identifier (‘add_color_info’, ‘true’, SampleIdentifier)
* To robustly identify the vegetables it is sufficient to use an
* internal image size of 0.1 megapixels. This speeds up the
* application.
set_sample_identifier_param (SampleIdentifier, ‘image_resize_value’, 0.1)
* Prepare the sample identifier by adding some representive sample
* images that show the different types of objects. For this,
* we simply use the training images.
for Index := 0 to |TrainingImageFileNames| - 1 by 1
read_image (Image, TrainingImageFileNames[Index])
dev_display (Image)
disp_message (WindowHandle, ‘Add a sample of ’ + TrainingObjectNames[Index], ‘window’, 12, 12, ‘black’, ‘true’)
* Add the current image as preparation data. Because we know
* the object ID, we can use it in order to label the preparation
* data. Thus, we can later reuse it for training and do not need to
* add additional training data.
add_sample_identifier_preparation_data (Image, SampleIdentifier, TrainingObjectIDs[Index], [], [], ObjectSampleIdx)
* Optionally add the object name to the sample identifier for
* convinience reasons.
set_sample_identifier_object_info (SampleIdentifier, TrainingObjectIDs[Index], ‘preparation_object_name’, TrainingObjectNames[Index])
endfor
* Now, we can prepare the identifier
dev_clear_window ()
disp_message (WindowHandle, [‘Prepare and train the sample identifier’,’(this might take some seconds)'], ‘window’, 12, 12, ‘black’, ‘true’)
prepare_sample_identifier (SampleIdentifier, ‘true’, [], [])
* Use the preparation data to train the identifier
train_sample_identifier (SampleIdentifier, [], [])
* Write the trained identifier to disk
write_sample_identifier (SampleIdentifier, IdentifierFileName)
endif
disp_continue_message (WindowHandle, ‘black’, ‘true’)
stop ()
*

• Identify new objects
  Errors := 0
  RandomIndices := sort_index(rand(|ImageFileNames|))
• Choose how many images should be tested
  NumTests := 50
• Use the following line to test all images
• NumTests := |RandomIndices|
  for I := 1 to min([NumTests,|RandomIndices|]) by 1
  • Randomly select a run-time image
    Index := RandomIndices[I - 1]
    ImageFileName := ImageFileNames[Index]
    ObjectID := ObjectIDs[Index]
    ObjectName := ObjectNames[Index]
    read_image (Image, ImageFileName)
  • Apply the sample identifier to identify the new objects
    apply_sample_identifier (Image, SampleIdentifier, 1, 0, [], [], ObjectIdx, Rating)
  • Get the name of the identified object for visualization purposes
    get_sample_identifier_object_info (SampleIdentifier, ObjectIdx, ‘training_object_name’, ResultObjectName)
    if (ObjectIdx != ObjectID)
    • Use red color for incorrect identifications
      Color := ‘red’
      Errors := Errors + 1
      else
      Color := ‘forest green’
      endif
      dev_display (Image)
      disp_message (WindowHandle, 'Current object: ’ + ObjectName, ‘window’, 12, 12, ‘black’, ‘true’)
      disp_message (WindowHandle, ‘Identified as: ’ + ResultObjectName, ‘window’, 30, 12, Color, ‘true’)
      disp_message (WindowHandle, ‘Error rate: (’ + Errors + ‘/’ + I + ‘) = ’ + (100.0 * Errors / I)$’.1f’ + ‘%’, ‘window’, 50, 12, ‘black’, ‘true’)
      endfor
• Clear the sample identifier
  clear_sample_identifier (SampleIdentifier)

apple_1.png

arugula_1.png

blueberry_1.png

bonbon_tomato_1.png

carrot_1.png

champignon_1.png

cocktail_tomatoes_1.png

corn_salad_1.png

cress_1.png

mangetout_1.png

potato_1.png

potato_sweet_1.png

raspberry_1.png

strawberry_1.png

程序运行结果如下:

以下是代码的详细解释：

dev_update_off()：关闭设备更新，以避免在处理图像时刷新显示窗口。

DoTraining := false：设置一个变量DoTraining，用于控制是否进行训练。这里设置为false，表示默认不进行训练。

ImageDirectory、IdentifierFileName：定义图像目录和样本标识符文件名。

get_object_images()：获取图像文件名、对象名称和ID。

read_image()、dev_open_window_fit_image()、set_display_font()：读取图像，打开窗口并设置显示字体。

file_exists()：检查样本标识符文件是否存在。

如果样本标识符文件存在且DoTraining为false，则读取训练好的样本标识符；否则，创建新的样本标识符并进行训练。

create_sample_identifier()：创建样本标识符，可以选择是否添加颜色信息。

set_sample_identifier_param()：设置样本标识符参数，如内部图像大小。

add_sample_identifier_preparation_data()：向样本标识符添加准备数据，即训练图像。

set_sample_identifier_object_info()：为样本标识符设置对象信息，如对象名称。

prepare_sample_identifier()、train_sample_identifier()：准备和训练样本标识符。

write_sample_identifier()：将训练好的样本标识符写入文件。

clear_sample_identifier()：清除样本标识符。

识别新对象的过程：

随机选择测试图像。
使用样本标识符对新对象进行识别。
根据识别结果，显示识别正确或错误的信息，并计算错误率。
整个脚本的流程是：首先检查是否存在已训练的样本标识符，如果不存在，则创建并训练新的样本标识符。然后，使用训练好的样本标识符对新图像进行识别，并显示识别结果和错误率。最后，清除样本标识符以释放资源。