自动选择最佳特征进行分类－SVM (HALOCN)

Posted 2020-11-08 冂冋冏囧

HALCON12里的example，classify_pills_auto_select_features.hdev.

执行流程：

1.选取相关特征（本例选取color和region组的所有特征）
（本例用get_feature_names( : : GroupNames : Names)，根据特征组名得到一系列特征
2.创建训练数据，添加样本的相关特征至训练数据结构中。
 （本例用calculate_features(Region, Image : : FeatureNames : Features)），根据特征名计算特征
3.自动选择最适合特征
（本例用select_feature_set_svm( : : ClassTrainDataHandle, SelectionMethod, GenParamNames, GenParamValues : SVMHandle,SelectedFeatureIndices, Score)
4.用分类器对测试图片进行分类

* This example shows how to use the calculate_feature_set
* procedure library together with the automatic feature selection
* to classify different pill types using a SVM classifier.
*  该示例演示如何从calculate_feature得到的特征，进行自动最佳特征选择，
*  并使用SVM分类器对各类型的pill进行分类。
* First, the pills are segmented in some training images.
* Then, a list of color and region features are calculated
* for each pill and stored in a classifier training data structure.
* After that, the best features are automatically selected
* with the operator select_feature_set_svm, and finally
* the resulting classifier is applied on a number of test
* images.
* 首先，对包含有pills的训练图片进行分割。
* 计算color和region组相关的特征。
* 每1个pill的相关特征添加至分类器训练数据结构里。
* 所有的训练数据添加完毕，用select_feature_set_svm
* 也可以用其他的分类器类型选择比如（select_feature_set_knn(mlp,gmm)）
* 自动选择最佳特征，最后应用分类器对测试图像分类。
* Init visualization
dev_close_window ()
dev_update_off ()
read_image (Image, \'color/pills_class_01\')
dev_open_window_fit_image (Image, 0, 0, -1, -1, WindowHandle)
dev_set_draw (\'margin\')
dev_set_line_width (2)
set_display_font (WindowHandle, 16, \'mono\', \'true\', \'false\')
dev_set_colored (12)
*
PillNames := [\'big_round_red\',\'round_green\',\'small_round_red\',\'yellow_trans\',\'brown\',\'brown_green\']
PillNames := [PillNames,\'brown_grain\',\'purple\',\'turquese\',\'pink\']
PillColors := [\'#D08080\',\'#ADC691\',\'#FFB0A1\',\'#D5C398\',\'#B59C87\',\'#BCB3B8\',\'#B7ACA1\',\'#908E99\',\'#97B9BC\',\'#C0ABA9\']
*
* Check, which features and feature groups are available
query_feature_group_names (AvailableGroupNames)
query_feature_names_by_group (AvailableGroupNames, AvailableFeatureNames, AvailableCorrespondingGroups)
*
* Generate list of color and region features using the
* calculate_feature_set library procedures.
* 获取相关特征组的对应的特征名称(1个特征组包含许多特征)，比如region组，有area,width,height,phi,ra,rb等。
* 获取相关特征的特征向量长度，比如rgb_mean特征，特征向量长度为3(每个通道对应1个值)，area特征向量长度为1。
FeatureGroups := [\'region\',\'color\']
get_feature_names (FeatureGroups, FeatureNames)
get_feature_lengths (FeatureNames, FeatureLengths)
*
* Create and prepare classifier training data structure
* 创建和准备分类训练数据结构
create_class_train_data (sum(FeatureLengths), ClassTrainDataHandle)
set_feature_lengths_class_train_data (ClassTrainDataHandle, FeatureLengths, FeatureNames)
*
* Training loop  循环训练
*
for I := 1 to 10 by 1
    * Segment pills in training image
    read_image (Image, \'color/pills_class_\' + I$\'.2d\')
    segment_pills (Image, Pills)
    * Display segmentation result
    dev_display (Image)
    dev_set_color (\'white\')
    dev_display (Pills)
    disp_message (WindowHandle, \'Collecting \' + PillNames[I - 1] + \' samples\', \'window\', 12, 12, \'black\', \'true\')
    *
    * Calculate features for all segmented pills and store
    * them in the training data structure
    count_obj (Pills, Number)
    calculate_features (Pills, Image, FeatureNames, Features)
    add_sample_class_train_data (ClassTrainDataHandle, \'feature_column\', Features, I - 1)
    * 以上注意第二参数\'feature_column\'的选择,此例中由于Number的数大于1，以特征列做为顺序。
    * Visualize processed pills
    dev_set_color (PillColors[I - 1])
    dev_display (Pills)
    GroupList := sum(\'\\\'\' + FeatureGroups + \'\\\', \')
    tuple_str_first_n (GroupList, strlen(GroupList) - 3, GroupList)
    Message := \'Calculate \' + |FeatureNames| + \' features from following feature groups:\'
    disp_message (WindowHandle, [Message,GroupList], \'window\', 40, 12, \'black\', \'true\')
    disp_continue_message (WindowHandle, \'black\', \'true\')
    stop ()
endfor
*
* Automatically select suitable features from the training data
* 自动选择训练数据的最合适的特征
disp_message (WindowHandle, \'Selecting optimal features...\', \'window\', 90, 12, \'black\', \'true\')
select_feature_set_svm (ClassTrainDataHandle, \'greedy\', [], [], SVMHandle, SelectedFeatures, Score)
disp_message (WindowHandle, [\'Selected:\',SelectedFeatures], \'window\', 120, 12, \'black\', \'true\')
disp_continue_message (WindowHandle, \'black\', \'true\')
stop ()
* Free memory, because the training data is no longer used
clear_class_train_data (ClassTrainDataHandle)
*
* Classify pills in test images
* using the automatically trained classifier with the
* automatically selected features
* 用分类器对测试图片进行测试，使用训练分类器和自动选择的最佳特征。
dev_set_line_width (4)
dev_set_colored (12)
for I := 1 to 3 by 1
    * Segment pills in test image
    read_image (Image, \'color/pills_test_\' + I$\'.2d\')
    dev_display (Image)
    segment_pills (Image, Pills)
    * For all pills, calculate selected features
    * using the calculate_features procedure from the
    * calculate_feature_set library and classify them.
    PillsIDs := []
    count_obj (Pills, NPills)
    for P := 1 to NPills by 1
        select_obj (Pills, PillSelected, P)
        calculate_features (PillSelected, Image, SelectedFeatures, Features)
        classify_class_svm (SVMHandle, real(Features), 1, Class)
        * Display results
        PillsIDs := [PillsIDs,Class]
        dev_set_color (PillColors[Class])
        dev_display (PillSelected)
        area_center (PillSelected, Area, Row, Column)
        disp_message (WindowHandle, Class + 1, \'image\', Row, Column - 10, \'black\', \'true\')
    endfor
    disp_message (WindowHandle, \'Classify image \' + I + \' of 3 using following features:\', \'window\', 12, 12, \'black\', \'true\')
    disp_message (WindowHandle, SelectedFeatures, \'window\', 40, 12, \'black\', \'true\')
    if (I < 3)
        disp_continue_message (WindowHandle, \'black\', \'true\')
        stop ()
    endif
endfor
* Clean up memory
* 清除内存
clear_class_svm (SVMHandle)