[人工智能]Halcon木板纹理识别

* 功能是：通过木板纹理识别树种
* 过程是：特征提取：纹理区+图片（灰度共生矩阵的标量4、边缘灰度直方图1）组成的特征向量；用于训练多层感知机；最后用于识别木板

* Image Acquisition 01: Code generated by Image Acquisition 01
list_files ('E:/03 CV（古）/Halcon/代码/木板纹理识别', ['files','follow_links'], ImageFiles)
tuple_regexp_select (ImageFiles, ['\\.(tif|tiff|gif|bmp|jpg|jpeg|jp2|png|pcx|pgm|ppm|pbm|xwd|ima|hobj)$','ignore_case'], ImageFiles)
FeaturesExtended:=[]
FeaturesExtended1:=[]
for Index := 0 to |ImageFiles| - 1 by 1
    read_image (Image, ImageFiles[Index])
    dev_clear_window ()
    get_image_size (Image, Width, Height)
    dev_open_window (0, 0, Width/5, Height/5, 'black', WindowHandle)
    dev_display (Image)
    
    rgb1_to_gray (Image, GrayImage)
    threshold (GrayImage, Regions, 31, 254)
    connection (Regions, ConnectedRegions)
    select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 40825.7, 500000)
    Classes:=['apple','beech','cherry','maple','oak']
    *整个图片的特征向量
    cooc_feature_image (Image, Image, 6, 90, Energy, Correlation, Homogeneity, Contrast)#计算共生矩阵并导出其灰度值特征
    sobel_amp (Image, EdgeAmplitude, 'sum_abs', 3)#轮廓特征：使用Sobel算子检测边缘（幅度）特征
    gray_histo_abs (EdgeAmplitude, EdgeAmplitude, 8, AbsoluteHistoEdgeAmplitude)  #计算灰度值分布。Quantization=8表示输入灰度值的量子化。
    FeaturesExtended:=[Energy, Correlation, Homogeneity, Contrast]
    FeaturesExtended:=[FeaturesExtended,AbsoluteHistoEdgeAmplitude]
    *纹理区域的特征向量   
    cooc_feature_image (SelectedRegions, Image, 6, 90, Energy, Correlation, Homogeneity, Contrast)
    sobel_amp (Image, EdgeAmplitude, 'sum_abs', 3)
    gray_histo_abs (EdgeAmplitude, EdgeAmplitude, 8, AbsoluteHistoEdgeAmplitude)
    FeaturesExtended1:=[FeaturesExtended,Energy, Correlation, Homogeneity, Contrast]
    FeaturesExtended1:=[FeaturesExtended1,AbsoluteHistoEdgeAmplitude]
    NumFeatures:=|FeaturesExtended1|
    NumClasses:=|Classes|
    NumHidden:=15
    
    *2创建分类器
    create_class_mlp (NumFeatures, 15, 5, 'softmax', 'normalization', 10, 42, MLPHandle)
    
    for i := 0 to |ImageFiles| - 1 by 1
        read_image (Image, ImageFiles[i])  
        rgb1_to_gray (Image, GrayImage)
        threshold (GrayImage, Regions, 31, 254)
        connection (Regions, ConnectedRegions)
        select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 40825.7, 500000)

        cooc_feature_image (SelectedRegions, Image, 6, 90, Energy, Correlation, Homogeneity, Contrast)
        sobel_amp (Image, EdgeAmplitude, 'sum_abs', 3)
        gray_histo_abs (EdgeAmplitude, EdgeAmplitude, 8, AbsoluteHistoEdgeAmplitude)
        
        FeaturesExtended:=[Energy, Correlation, Homogeneity, Contrast]
        FeaturesExtended:=[FeaturesExtended,AbsoluteHistoEdgeAmplitude]
    
        cooc_feature_image (Image, Image, 6, 90, Energy, Correlation, Homogeneity, Contrast)
    
        sobel_amp (Image, EdgeAmplitude, 'sum_abs', 3)
        gray_histo_abs (EdgeAmplitude, EdgeAmplitude, 8, AbsoluteHistoEdgeAmplitude)
        FeaturesExtended1:=[FeaturesExtended,Energy, Correlation, Homogeneity, Contrast]
        FeaturesExtended1:=[FeaturesExtended1,AbsoluteHistoEdgeAmplitude]
        FeatureVector:=real(FeaturesExtended1)#实数化
    
        *添加样本
        add_sample_class_mlp (MLPHandle, FeatureVector, i)
    endfor
    *训练分类器
    train_class_mlp (MLPHandle, 200, 1, 0.01, Error, ErrorLog)
    stop ()
    write_class_mlp (MLPHandle, 'D://1.gmc')

    *测试
    for i := 0 to |ImageFiles| - 1 by 1
        read_image (Image, ImageFiles[i])   
        rgb1_to_gray(Image, GrayImage)
    
        threshold (GrayImage, Regions, 31, 254)
        connection (Regions, ConnectedRegions)
        select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 40825.7, 500000)
        cooc_feature_image (SelectedRegions, Image, 6, 90, Energy, Correlation, Homogeneity, Contrast)
    
        sobel_amp (Image, EdgeAmplitude, 'sum_abs', 3)
        gray_histo_abs (EdgeAmplitude, EdgeAmplitude, 8, AbsoluteHistoEdgeAmplitude)
        FeaturesExtended := [Energy,Correlation,Homogeneity,Contrast]
        FeaturesExtended := [FeaturesExtended,AbsoluteHistoEdgeAmplitude]
        cooc_feature_image (Image, Image, 6, 90, Energy, Correlation, Homogeneity, Contrast)
        
        sobel_amp (Image, EdgeAmplitude, 'sum_abs', 3)
        gray_histo_abs (EdgeAmplitude, EdgeAmplitude, 8, AbsoluteHistoEdgeAmplitude)
        FeaturesExtended1 := [FeaturesExtended,Energy,Correlation,Homogeneity,Contrast]
        FeaturesExtended1 := [FeaturesExtended1,AbsoluteHistoEdgeAmplitude]
        FeatureVector := real(FeaturesExtended1)
        ******3 识别
        classify_class_mlp (MLPHandle, FeatureVector, 1, FoundClassIDs, k)
        dev_display(Image)
        ImageFiles1:= 'found class:   ' + Classes[FoundClassIDs[0]]
        disp_message(WindowHandle, ImageFiles1, 'image', 12, 12, 'black', 'true')
        stop()
    endfor
endfor

总结：

1.使用多层感知机的步骤和颜色识别是一模一样的，都是分三步，如代码里注释所写。
2.不过它们还是有些差别的。颜色识别输入是图片或者图片区域，而纹理识别输入的是72维特征向量，所以在添加样本时有细微差别，前者是add_samples_image_class_mlp，后者是add_sample_class_mlp。