機器視覺與機器學(xué)習(xí)應(yīng)用ApplicationsofMachineVisionandMachineLearning漏印Missing

prints模糊

Blurring污點

Stains字符缺陷檢測Character

Defect

Detection課

程

導(dǎo)

入Course

Introduction主講教師

Teacher陳

雨Chen

Yu課

程

導(dǎo)

入CourseIntroduction如何高效準確地檢測這些缺陷呢?Howto

detectthese

defects字符缺陷檢測Character

Defect

DetectionefFicientlyandaccurately?字符缺陷檢測CharacterDefect

Detection課程導(dǎo)入

CourseIntroduction字符缺陷檢測Characterdefectdetection*技術(shù)原理Technicalprinciples*實際操作Practicaloperations案例導(dǎo)入Case

introduction生產(chǎn)日期Production

date*批次號Batch

number案例導(dǎo)入Case

introduction*

模糊或漏印Blurred

or

missing

prints*無法識別產(chǎn)品信息

引發(fā)質(zhì)量問題Unable

to

identifyproduct

informationCause

quality

issues發(fā)現(xiàn)并剔除有問題的產(chǎn)品Detect

and

eliminate

problematic

products*

確保出廠產(chǎn)品的質(zhì)量Ensure

the

quality

of

products

leaving

the

factory使用視覺套件完成字符缺陷檢測的全過程Completetheentireprocessofcharacterdefectdetectionusingthevision

kit任務(wù)描述Task

Description無缺陷的字符Characterswithoutdefects帶有缺陷的字符Characterswithdefects字符缺陷檢測CharacterdefectdetectionDetection字符缺陷檢測CharacterDefectDetection*將目標圖像與標準圖像進行驗證對比，檢測印刷字符，圖案是否存在缺失、

冗余等非一致性外觀缺陷。To

verify

and

compare

the

target

image

with

the

standard

image

to

detect

whether

the

printedcharactersand

patterns

have

inconsistentappearancedefects

such

as

missing

or

redundant

ones.字箍缺陷檢

Detection*

工業(yè)生產(chǎn)中，字符有機器印刷，由于產(chǎn)品漏放，偏放或機器誤操

作，都會導(dǎo)致產(chǎn)品字符缺陷。In

industrial

production,characters

areprinted

by

machines.Omission,misplacement

or

machine

misoperation

of

the

product

may

cause

character

defects

on

the

product.產(chǎn)品的字符檢測尤為重要Product

character

detection

is

particularlyimportant.字體缺失Missing

font字體缺失Missing

font字體缺失Missing

Font字符缺陷類型CharacterDefect

Types字符缺陷示例Characterdefect

examplesHW1HO385GEMINIP.N:1AOQ58[OO5]200046659271H0385污點

Stain污點

Stain污點Stain字符缺陷：字符模糊、漏印、污點、

位置偏移Character

defects:blurred

characters,missingcharacters,stains,andpositionshifts.字符缺陷類型Character

Defect

Types1HO385污點

StainHW1HO385GEMINIP.N:1AOQ58[OO5]20004665927字符缺陷示例Characterdefect

examples1H0385字體缺失Missing

font字體缺失Missing

font字體缺失Missing

Font污點

Stain污點Stain開始Start獲取圖像Imageacquisition缺陷檢測結(jié)束Defectdetectioncompleted比對樣本Samplescomparison訓(xùn)練Or檢測Training

or

Detection字符缺陷檢測分為兩個過程：Character

defect

detection

is

divided

into

two

processes:字符訓(xùn)練字符缺陷檢測Charactertraining

Character

defect

detection字符訓(xùn)練：主要是完成標準字符

的訓(xùn)練Character

training:mainly

to

complete

the

training

of

standard

characters字符缺陷檢測：就是與標準字符

進行對比，分析出字符的缺陷Characterdefectdetection:comparewith

standard

characters

to

analyze

characterdefects字符缺陷檢測流程Characterdefectdetectionprocess字箍缺降檢測過程ctionProcess圖像預(yù)處理ImagepreprocessingROI分割ROIsegmentation缺陷檢測Defectdetection訓(xùn)練結(jié)束End

of

training樣本數(shù)據(jù)

Sampledata訓(xùn)練

Training*

字符模板參數(shù)Charactertemplateparameters*

訓(xùn)練模板參數(shù)Trainingtemplateparameters*

統(tǒng)計訓(xùn)練參數(shù)Statisticaltrainingparameters*

運行參數(shù)OperatingparametersDobotVisionStudio字符缺陷檢測介紹Do

botVisionStudio

Character

Defect

Detection

Introduction字符模板——字符模板參數(shù)Charactertemplates-charactertemplateparameters字符分割閾值：字符分割的臨界值，太小了容易提取到多余的特征圖形，太大了容易丟失目標圖形，建議一般設(shè)置成128左右。Character

segmentation

threshold:The

critical

value

for

character

segmentation.If

itis

too

small,it

is

easy

to

extract

redundant

feature

graphics,and

if

it

is

too

large,it

is

easy

to

lose

the

target

graphics.It

is

generally

recommended

to

set

it

to

around

128.*

顯示二值圖：開啟以后會進行圖像二值化，背景圖像和字符灰度差更明顯。Display

binary

image:After

turning

it

on,the

image

will

be

binarized,and

the

grayscaledifFerence

between

the

background

image

and

the

characters

will

be

more

obvious.*

字符極性：有黑底白字，白底黑字兩種。Character

polarity:There

are

two

types:white

characters

on

a

black

background

and

blackcharacters

on

a

white

background.*

字符高度/寬度范圍：可以設(shè)置目標字符的高度和寬度值范圍。Character

height/width

range:You

can

set

the

height

and

width

value

range

of

the

targetcharacters.*

高度/寬度增加值：目標字符區(qū)域高度和寬度的單次增加值。Height/width

increase

value:The

single

increase

value

of

the

height

and

width

of

thetarget

character

area.DobotVisionStudio字符缺陷檢測介紹DobotVisionStudioCharacterDefect

DetectionIntroduction分割提取字符分割

字符提取

字符分割閾值|128顯示二值圖字符極性

白底黑字字符寬度范圍

512字符高度范圍10

512字符模板——字符模板參數(shù)Charactertemplates-charactertemplateparameters字符分割Charactersegmentation縮放模式：可選取手動和自動兩種模式，推薦使用自動模式Scalingmode:Manualandautomaticmodesare

available,automatic

mode

isrecommended

精細顆粒度：精細顆粒度參數(shù)，表示提取特征顆粒的精細程度，當(dāng)精細尺度取值為1時精

細程度最大，取邊緣點數(shù)量最多，精度最高。Finegranularity:Thefinegranularityparameterindicatesthedegree

of

Fineness

oftheextractedfeatureparticles.Whenthefinescalevalueis1,thedegree

offineness

isthe

largest,thenumberofedgepointsisthe

largest,andthe

accuracy

is

the

highest.粗糙顆粒度：粗糙顆粒度參數(shù)，數(shù)值越大抽取邊緣點就越稀疏，但會加快匹配速度，默認

范圍是1~20。Coarsegranularity:Coarsegranularityparameter.Thelargerthevalue,thesparserthe

edge

pointswillbe,butitwillspeedupthe

matching

process.The

default

range

is

1~20.閾值模式：可選自動以及手動兩種模式Thresholdmode:Automaticandmanualmodes

are

available自動，根據(jù)目標字符自動決定閾值參數(shù)，自動適應(yīng)。Automatic:Automaticallydeterminesthethresholdparametersbasedonthetarget

character

andautomaticallyadapts.手動，以用戶設(shè)定的閾值作為查找的閾值參數(shù)。Manual:Usingtheuser-definedthresholdasthesearchthresholdparameter.模型低閾值：閾值模式選取手動時生效，查找的最低閾值參數(shù)。Modellowthreshold:EFfectivewhenmanualthresholdmode

isselected,the

lowest

thresholdparametertobesearched.DobotVisionStudio字符缺陷檢測介紹DobotVisionStudio

Character

Defect

Detection

Introduction字符模板——字符模板參數(shù)Charactertemplates-charactertemplateparameters訓(xùn)練模板自

動1.0003.000自

動訓(xùn)練參數(shù)Trainingparameters訓(xùn)練參數(shù)縮放模式精細顆粒度粗糙顆粒度閾值模式模型低閾值****均值標準差法：通過多張圖片得到均值圖像與標準差圖像，

再與高低比例偏移系數(shù)計算得到正常高低閾值圖像，檢測

輸入圖像通過與高低閾值圖像進行像素灰度比對來判斷圖

像中是否存在缺陷。Meanstandarddeviation

method:The

mean

imageandstandarddeviationimageareobtainedfrom

multiple

images,andthenthe

highandlow

proportionaloffsetcoefficientsarecalculatedtoobtainnormal

high

and

low

thresholdimages.The

input

image

isdetected

bycomparingthe

pixelgrayscalewiththehighandlowthreshold

imagestodeterminewhetherthereare

defects

intheimage.相關(guān)法：通過計算均值圖像與輸入檢測圖像的灰度相關(guān)性

來進行缺陷判斷。Correlation

method:Defectjudgment

is

performedbycalculatingthegrayscale

correlation

betweenthe

mean

imageandthe

inputdetection

image.DobotVisionStudio字符缺陷檢測介紹DobotVisionStudioCharacterDefectDetection

Introduction字符模板——統(tǒng)計訓(xùn)練參數(shù)Charactertemplates-statisticaltrainingparameters字符統(tǒng)計Characterstatistics*

歸一化類型：有不處理、直方圖均衡化、直方圖歸一

化、均值標準差歸一化4種歸一化模式。Normalizationmodes:Noprocessing,histogramequalization,

histogram

normalization,and

mean

standard

deviationnormalization.相關(guān)分數(shù)閾值：低于評分閾值的目標認定為含缺陷。

Correlation

score

threshold:Targets

below

the

score

threshold

are

considered

to

contain

defects.*

寬度/高度方向分塊：對字符進行分塊相關(guān)評分，都設(shè)

為1為不分塊。Width/height

direction

blocking:The

characters

are

scored

forblocking,and

all

are

set

to

1for

no

blocking.3字符缺陷檢測基本參數(shù)字符模板

運行參數(shù)結(jié)果顯示歸一化處理歸一化類型

均值標準差歸一化字符檢測相關(guān)分數(shù)閾值高度方向區(qū)塊寬度方向區(qū)塊3.50050.000執(zhí)行

確定DobotVisionStudio字符缺陷檢測介紹DobotVisionStudio

Character

Defect

Detection

Introduction運行參數(shù)Operatingparameters運行參數(shù)Qperating

parameters高閾值比例高閾值容忍任務(wù)實施Task

Implementation流程0本地圖像相機圖像存儲圖像0相機圖像相機圖像Camera

image步驟1Step1打開DobotVisionStudio1.4.2,選擇通用方案。

Open

DobotVisionStudio1.4.2and

select

the

GeneralOption.*在工具箱的“采集”模塊選擇“相機圖像”,拖拽到流程編輯區(qū)。Select"Camera

Image"in

the

"Acquisition"module

ofthe

toolbox

and

drag

it

to

the

process

editing

area.任務(wù)實施Task

Implementation步

驟

2SER2Detection相機采集到的圖形Graphics

captured

by

the

camera雙擊“相機圖像”模塊，進行參數(shù)設(shè)置。R8ukle-clickthe"Camera

Image"moduletosetthe

parameters,任務(wù)實施Task

Implementation0

相

機

圖

像觸發(fā)設(shè)置相機連接選擇相機

U3VMV-CE050-30UC(C圖像參數(shù)圖像寬度圖像高度像素格式幀率實際幀率25921944MONO831.00012.156觸發(fā)設(shè)置SOFTWARE觸發(fā)延遲0.000觸發(fā)字符串字符觸發(fā)過濾相機圖像基本參數(shù)設(shè)置Camera

imagebasicparametersettings相機圖像觸發(fā)參數(shù)設(shè)置Camera

imagetriggerparametersettingsIO控制IO選擇項IO

模

式LINEOInput*在工具箱的定位模塊，將“快速特征匹配”拖拽

到流程編輯區(qū)，并與“O相機圖像”相連。In

the

positioning

module

of

the

toolbox,drag

"QuickFeatureMatching"totheprocesseditingarea,and

connect

itwith

"O

Camera

Image".1快速特征..方案流程

Program

process任務(wù)實施Task

Implementation步驟3Step3任務(wù)實施TaskImplementation*

雙擊“1快速特征匹配”進行參數(shù)設(shè)置。

Double-click

"1Fast

Feature

Matching"to

setparameters.創(chuàng)建特征模板Creatingafeaturetemplate特征模板Featuretemplate運行參數(shù)結(jié)果顯示Detectlon編輯模板執(zhí)行確定1快速特征匹配基本參數(shù)

特征模板+

創(chuàng)

建新建模板1任務(wù)實施Task

Implementation0相機圖像1快速特征

….*

將“字符缺陷檢測”拖拽到流程編輯區(qū)，與“1快

速特征匹配”連接。Drag"CharacterDefect

Detection"tothe

process

editingarea,andconnectitwith

"1

Fast

Feature

Matching".3字符缺陷…方案流程界面Solutionprocess

interface步驟

4Step4任務(wù)實施Double

click

"Character

Defect

Detection",andTask

Implementation*雙擊“字符缺陷檢測”,進行參數(shù)設(shè)置。-

perform

parameter

settings.運行參數(shù)

結(jié)果顯示0相機圖像.圖像數(shù)據(jù)繼承按參數(shù)1快速特征匹配.匹配框口執(zhí)行

確定基本參數(shù)界面parameter

interface輸入源ROI

區(qū)域RO

創(chuàng)

建繼承方式區(qū)域Basic任務(wù)實施Task

Implementation過程1:標準字符的訓(xùn)練Process1:Standard

character

training字符模板Charactertemplate空白字符模板Blankcharactertemplate*

創(chuàng)建字符模板。Create

character

templates.Detection2字特缺陷檢測執(zhí)行字符模板基本參數(shù)*

創(chuàng)建字符模板。Createcharactertemplates.均值標準差法任務(wù)實施Task

Implementation過程1:標準字符的訓(xùn)練Process

1:Standardcharactertraining字符模板Charactertemplate字符模板結(jié)果CharactertemplateresultsDetection

統(tǒng)計訓(xùn)練結(jié)果2字符缺陷檢測字符模板執(zhí)行

確定字符模板*將字符塊換成帶有缺陷的字符塊。點擊“執(zhí)行”,圖像區(qū)會顯示檢測結(jié)果。

Replace

the

character

blockwith

a

character

blockwith

defects.Click

"Execute",theimage

area

will

display

the

detection

results任務(wù)實施Task

Implementation過程2:字符缺陷檢測Process2:Cha