Halcon mlp分类器实现颜色检测

HeliosXxzh

于 2024-07-25 15:22:51 发布

阅读量343

点赞数 7

分类专栏： Halcon 文章标签：机器学习算法计算机视觉

本文链接：https://blog.csdn.net/weixin_44170250/article/details/140690432

版权

Halcon 专栏收录该内容

8 篇文章 1 订阅

订阅专栏

一、项目背景

锂电池卷绕后的红标需检测是否合格，红标的位置不固定，通过mlp分类器将背景色与红色进行训练，用分类器进行识别红标是否OK

二、halcon实现过程

1.图像增强

scale_image (Image, ImageScaled, 4, 0)

2.提取红色，背景色

gen_rectangle1 (ROI_red, 1159.09, 2716.76, 1169.44, 2742.93)
gen_rectangle1 (ROI_back, 1425.73, 2642.53, 1446.39, 2709.4)
concat_obj (ROI_red, ROI_back, ObjectsConcat)

3.训练

create_class_mlp (3, 12, 2, 'softmax', 'normalization', 2, 40, MLPHandle)
add_samples_image_class_mlp (ImageScaled, ObjectsConcat, MLPHandle)
train_class_mlp (MLPHandle, 500, 0.5, 0.01, Error, ErrorLog)
write_class_mlp (MLPHandle, 'mlp.gmc')

4.分类识别

for Index := 0 to |ImageFiles|-1 by 1
    read_image (Image1, ImageFiles[Index])
    dev_display (Image1)
    scale_image (Image1, ImageScaled1, 3, 0)
    classify_image_class_mlp (ImageScaled1, ClassRegions, MLPHandle, 0.5)  
    for Color := 1 to 1 by 1
        copy_obj (ClassRegions, ObjectsSelected, Color, 1)
        connection (ObjectsSelected, ConnectedRegions)
        select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 15000, 99999)
        count_obj (SelectedRegions, Number)
        area_center (SelectedRegions, Area, Row, Column)
        if (Number==1)
            get_image_size (Image1, Width, Height)                     
            dev_set_draw ('margin')
            dev_set_line_width (3)
            dev_display (Image1)
            dev_display (ObjectsSelected) 
            dev_disp_text ('红标OK', 'image', Row-200, Column, 'green', [], [])
        else
            dev_display (Image1)
            dev_disp_text ('红标NG', 'image', 500, 1500, 'red', [], [])
        endif
                
    endfor 
    stop ()
endfor

三、完整代码及图像

list_image_files ('image', 'default', [], ImageFiles)

read_image (Image, ImageFiles[0])
scale_image (Image, ImageScaled, 4, 0)
gen_rectangle1 (ROI_red, 1159.09, 2716.76, 1169.44, 2742.93)
gen_rectangle1 (ROI_back, 1425.73, 2642.53, 1446.39, 2709.4)
concat_obj (ROI_red, ROI_back, ObjectsConcat)
create_class_mlp (3, 12, 2, 'softmax', 'normalization', 2, 40, MLPHandle)
add_samples_image_class_mlp (ImageScaled, ObjectsConcat, MLPHandle)
train_class_mlp (MLPHandle, 500, 0.5, 0.01, Error, ErrorLog)
write_class_mlp (MLPHandle, 'mlp.gmc')
for Index := 0 to |ImageFiles|-1 by 1
    read_image (Image1, ImageFiles[Index])
    dev_display (Image1)
    scale_image (Image1, ImageScaled1, 3, 0)
    classify_image_class_mlp (ImageScaled1, ClassRegions, MLPHandle, 0.5)  
    for Color := 1 to 1 by 1
        copy_obj (ClassRegions, ObjectsSelected, Color, 1)
        connection (ObjectsSelected, ConnectedRegions)
        select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 15000, 99999)
        count_obj (SelectedRegions, Number)
        area_center (SelectedRegions, Area, Row, Column)
        if (Number==1)
            get_image_size (Image1, Width, Height)                     
            dev_set_draw ('margin')
            dev_set_line_width (3)
            dev_display (Image1)
            dev_display (ObjectsSelected) 
            dev_disp_text ('红标OK', 'image', Row-200, Column, 'green', [], [])
        else
            dev_display (Image1)
            dev_disp_text ('红标NG', 'image', 500, 1500, 'red', [], [])
        endif               
    endfor 
    stop ()
endfor

https://download.csdn.net/download/weixin_44170250/89578964https://download.csdn.net/download/weixin_44170250/89578964