1. The document discusses using deep learning techniques for surface defect detection, focusing on strategies for dealing with imbalanced training data.
2. It proposes using generative adversarial networks (GANs) to generate synthetic defect samples in order to address the class imbalance problem. Convolutional neural networks (CNNs) are then used for classification.
3. Autoencoding models like convolutional autoencoders (CAE) and variational autoencoders (VAE) can also be used for unsupervised defect detection based on image reconstruction.
Slides presented in the All Japan Computer Vision Study Group on May 15, 2022. Methods for disentangling the relationship between multimodal data are discussed.
PR-207: YOLOv3: An Incremental ImprovementJinwon Lee
TensorFlow Korea 논문읽기모임 PR12 207번째 논문 review입니다
이번 논문은 YOLO v3입니다.
매우 유명한 논문이라서 크게 부연설명이 필요없을 것 같은데요, Object Detection algorithm들 중에 YOLO는 굉장히 특색있는 one-stage algorithm입니다. 이 논문에서는 YOLO v2(YOLO9000) 이후에 성능 향상을 위하여 어떤 것들을 적용하였는지 하나씩 설명해주고 있습니다. 또한 MS COCO의 metric인 average mAP에 대해서 비판하면서 mAP를 평가하는 방법에 대해서도 얘기를 하고 있는데요, 자세한 내용은 영상을 참고해주세요~
논문링크: https://arxiv.org/abs/1804.02767
영상링크: https://youtu.be/HMgcvgRrDcA
Slides presented in the All Japan Computer Vision Study Group on May 15, 2022. Methods for disentangling the relationship between multimodal data are discussed.
PR-207: YOLOv3: An Incremental ImprovementJinwon Lee
TensorFlow Korea 논문읽기모임 PR12 207번째 논문 review입니다
이번 논문은 YOLO v3입니다.
매우 유명한 논문이라서 크게 부연설명이 필요없을 것 같은데요, Object Detection algorithm들 중에 YOLO는 굉장히 특색있는 one-stage algorithm입니다. 이 논문에서는 YOLO v2(YOLO9000) 이후에 성능 향상을 위하여 어떤 것들을 적용하였는지 하나씩 설명해주고 있습니다. 또한 MS COCO의 metric인 average mAP에 대해서 비판하면서 mAP를 평가하는 방법에 대해서도 얘기를 하고 있는데요, 자세한 내용은 영상을 참고해주세요~
논문링크: https://arxiv.org/abs/1804.02767
영상링크: https://youtu.be/HMgcvgRrDcA
Towards Total Recall in Industrial Anomaly Detectionharmonylab
公開URL:https://openaccess.thecvf.com/content/CVPR2022/papers/Roth_Towards_Total_Recall_in_Industrial_Anomaly_Detection_CVPR_2022_paper.pdf
出典:Karsten Roth, Latha Pemula, Joaquin Zepeda, Bernhard Schölkopf, Thomas Brox, Peter Gehler: Towards Total Recall in Industrial Anomaly Detection, Conference on Computer Vision and Pattern Recognition (CVPR), pp. 14318-14328 (2022)
概要:本論文では位置情報を考慮した特徴量の集合和であるメモリバンクとCoresetによる画像パッチ特徴量の削減を行うPatchCoreアルゴリズムを提案する.結果として、異常検出のベンチマークであるMVTecにおいてAUROC99%以上の精度を出力し,2022年時点でのSoTAを記録した.また,PatchCoreによる特徴量削減により,学習のサンプル数を20%に減らした場合でも以前のSoTAに匹敵する精度となった.
This slide is for the keynote speech in JaSST Hokkaido 2020. It analysis problems of Softhouses, Japanese software companies, and proposes how to transform softhouses to good companies.
Towards Total Recall in Industrial Anomaly Detectionharmonylab
公開URL:https://openaccess.thecvf.com/content/CVPR2022/papers/Roth_Towards_Total_Recall_in_Industrial_Anomaly_Detection_CVPR_2022_paper.pdf
出典:Karsten Roth, Latha Pemula, Joaquin Zepeda, Bernhard Schölkopf, Thomas Brox, Peter Gehler: Towards Total Recall in Industrial Anomaly Detection, Conference on Computer Vision and Pattern Recognition (CVPR), pp. 14318-14328 (2022)
概要:本論文では位置情報を考慮した特徴量の集合和であるメモリバンクとCoresetによる画像パッチ特徴量の削減を行うPatchCoreアルゴリズムを提案する.結果として、異常検出のベンチマークであるMVTecにおいてAUROC99%以上の精度を出力し,2022年時点でのSoTAを記録した.また,PatchCoreによる特徴量削減により,学習のサンプル数を20%に減らした場合でも以前のSoTAに匹敵する精度となった.
This slide is for the keynote speech in JaSST Hokkaido 2020. It analysis problems of Softhouses, Japanese software companies, and proposes how to transform softhouses to good companies.
The slides for the techniques used in the Temporal Segment Network (TSN), including the basic ideas, recall of BN-Inception, optical flow and tricks in application. Used in group paper reading in University of Sydney.
Build Your Own 3D Scanner: 3D Scanning with Structured LightingDouglas Lanman
Build Your Own 3D Scanner:
3D Scanning with Structured Lighting
http://mesh.brown.edu/byo3d/
SIGGRAPH 2009 Courses
Douglas Lanman and Gabriel Taubin
This course provides a beginner with the necessary mathematics, software, and practical details to leverage projector-camera systems in their own 3D scanning projects. An example-driven approach is used throughout; each new concept is illustrated using a practical scanner implemented with off-the-shelf parts. The course concludes by detailing how these new approaches are used in rapid prototyping, entertainment, cultural heritage, and web-based applications.
DEEP NEURAL NETWORKS APPLIED TO LOW POWER ONBOARD IMAGE COMPRESSION
Over the past decade, rapid developments in digital technologies and access to space have enabled unprecedented capabilities of monitoring our planet and, more generally, our Universe.
This new space race is pushing for a paradigm shift in order to respond to the ever-increasing challenge of delivering the useful information to the end users. With huge number of satellites, greater spatial and spectral resolutions, higher temporal cadence and shrinking spectrum resources, on-board data reduction becomes not only a cost saving solution but, in many cases also, a key enabling technology to achieve viable missions.
https://atpi.eventsair.com/obpdc2022/
Epistemic Interaction - tuning interfaces to provide information for AI supportAlan Dix
Paper presented at SYNERGY workshop at AVI 2024, Genoa, Italy. 3rd June 2024
https://alandix.com/academic/papers/synergy2024-epistemic/
As machine learning integrates deeper into human-computer interactions, the concept of epistemic interaction emerges, aiming to refine these interactions to enhance system adaptability. This approach encourages minor, intentional adjustments in user behaviour to enrich the data available for system learning. This paper introduces epistemic interaction within the context of human-system communication, illustrating how deliberate interaction design can improve system understanding and adaptation. Through concrete examples, we demonstrate the potential of epistemic interaction to significantly advance human-computer interaction by leveraging intuitive human communication strategies to inform system design and functionality, offering a novel pathway for enriching user-system engagements.
State of ICS and IoT Cyber Threat Landscape Report 2024 previewPrayukth K V
The IoT and OT threat landscape report has been prepared by the Threat Research Team at Sectrio using data from Sectrio, cyber threat intelligence farming facilities spread across over 85 cities around the world. In addition, Sectrio also runs AI-based advanced threat and payload engagement facilities that serve as sinks to attract and engage sophisticated threat actors, and newer malware including new variants and latent threats that are at an earlier stage of development.
The latest edition of the OT/ICS and IoT security Threat Landscape Report 2024 also covers:
State of global ICS asset and network exposure
Sectoral targets and attacks as well as the cost of ransom
Global APT activity, AI usage, actor and tactic profiles, and implications
Rise in volumes of AI-powered cyberattacks
Major cyber events in 2024
Malware and malicious payload trends
Cyberattack types and targets
Vulnerability exploit attempts on CVEs
Attacks on counties – USA
Expansion of bot farms – how, where, and why
In-depth analysis of the cyber threat landscape across North America, South America, Europe, APAC, and the Middle East
Why are attacks on smart factories rising?
Cyber risk predictions
Axis of attacks – Europe
Systemic attacks in the Middle East
Download the full report from here:
https://sectrio.com/resources/ot-threat-landscape-reports/sectrio-releases-ot-ics-and-iot-security-threat-landscape-report-2024/
Transcript: Selling digital books in 2024: Insights from industry leaders - T...BookNet Canada
The publishing industry has been selling digital audiobooks and ebooks for over a decade and has found its groove. What’s changed? What has stayed the same? Where do we go from here? Join a group of leading sales peers from across the industry for a conversation about the lessons learned since the popularization of digital books, best practices, digital book supply chain management, and more.
Link to video recording: https://bnctechforum.ca/sessions/selling-digital-books-in-2024-insights-from-industry-leaders/
Presented by BookNet Canada on May 28, 2024, with support from the Department of Canadian Heritage.
Generating a custom Ruby SDK for your web service or Rails API using Smithyg2nightmarescribd
Have you ever wanted a Ruby client API to communicate with your web service? Smithy is a protocol-agnostic language for defining services and SDKs. Smithy Ruby is an implementation of Smithy that generates a Ruby SDK using a Smithy model. In this talk, we will explore Smithy and Smithy Ruby to learn how to generate custom feature-rich SDKs that can communicate with any web service, such as a Rails JSON API.
LF Energy Webinar: Electrical Grid Modelling and Simulation Through PowSyBl -...DanBrown980551
Do you want to learn how to model and simulate an electrical network from scratch in under an hour?
Then welcome to this PowSyBl workshop, hosted by Rte, the French Transmission System Operator (TSO)!
During the webinar, you will discover the PowSyBl ecosystem as well as handle and study an electrical network through an interactive Python notebook.
PowSyBl is an open source project hosted by LF Energy, which offers a comprehensive set of features for electrical grid modelling and simulation. Among other advanced features, PowSyBl provides:
- A fully editable and extendable library for grid component modelling;
- Visualization tools to display your network;
- Grid simulation tools, such as power flows, security analyses (with or without remedial actions) and sensitivity analyses;
The framework is mostly written in Java, with a Python binding so that Python developers can access PowSyBl functionalities as well.
What you will learn during the webinar:
- For beginners: discover PowSyBl's functionalities through a quick general presentation and the notebook, without needing any expert coding skills;
- For advanced developers: master the skills to efficiently apply PowSyBl functionalities to your real-world scenarios.
Key Trends Shaping the Future of Infrastructure.pdfCheryl Hung
Keynote at DIGIT West Expo, Glasgow on 29 May 2024.
Cheryl Hung, ochery.com
Sr Director, Infrastructure Ecosystem, Arm.
The key trends across hardware, cloud and open-source; exploring how these areas are likely to mature and develop over the short and long-term, and then considering how organisations can position themselves to adapt and thrive.
Essentials of Automations: Optimizing FME Workflows with ParametersSafe Software
Are you looking to streamline your workflows and boost your projects’ efficiency? Do you find yourself searching for ways to add flexibility and control over your FME workflows? If so, you’re in the right place.
Join us for an insightful dive into the world of FME parameters, a critical element in optimizing workflow efficiency. This webinar marks the beginning of our three-part “Essentials of Automation” series. This first webinar is designed to equip you with the knowledge and skills to utilize parameters effectively: enhancing the flexibility, maintainability, and user control of your FME projects.
Here’s what you’ll gain:
- Essentials of FME Parameters: Understand the pivotal role of parameters, including Reader/Writer, Transformer, User, and FME Flow categories. Discover how they are the key to unlocking automation and optimization within your workflows.
- Practical Applications in FME Form: Delve into key user parameter types including choice, connections, and file URLs. Allow users to control how a workflow runs, making your workflows more reusable. Learn to import values and deliver the best user experience for your workflows while enhancing accuracy.
- Optimization Strategies in FME Flow: Explore the creation and strategic deployment of parameters in FME Flow, including the use of deployment and geometry parameters, to maximize workflow efficiency.
- Pro Tips for Success: Gain insights on parameterizing connections and leveraging new features like Conditional Visibility for clarity and simplicity.
We’ll wrap up with a glimpse into future webinars, followed by a Q&A session to address your specific questions surrounding this topic.
Don’t miss this opportunity to elevate your FME expertise and drive your projects to new heights of efficiency.
The Art of the Pitch: WordPress Relationships and SalesLaura Byrne
Clients don’t know what they don’t know. What web solutions are right for them? How does WordPress come into the picture? How do you make sure you understand scope and timeline? What do you do if sometime changes?
All these questions and more will be explored as we talk about matching clients’ needs with what your agency offers without pulling teeth or pulling your hair out. Practical tips, and strategies for successful relationship building that leads to closing the deal.
Dev Dives: Train smarter, not harder – active learning and UiPath LLMs for do...UiPathCommunity
💥 Speed, accuracy, and scaling – discover the superpowers of GenAI in action with UiPath Document Understanding and Communications Mining™:
See how to accelerate model training and optimize model performance with active learning
Learn about the latest enhancements to out-of-the-box document processing – with little to no training required
Get an exclusive demo of the new family of UiPath LLMs – GenAI models specialized for processing different types of documents and messages
This is a hands-on session specifically designed for automation developers and AI enthusiasts seeking to enhance their knowledge in leveraging the latest intelligent document processing capabilities offered by UiPath.
Speakers:
👨🏫 Andras Palfi, Senior Product Manager, UiPath
👩🏫 Lenka Dulovicova, Product Program Manager, UiPath
3. Inspection in manufacturing
Product & process properties in manufacturing:
• Small , subtle local defect in
− Size
− Shape
− Deformation
− Gray scale/color unevenness
• Imbalanced data
– Many positive (defect-free) samples
– Only a few or No negative (defect) samples
Requirements:
• High precision/accuracy
– Location
– Detection rate
• Computationally fast
(Real-time computation to meet the cycle time)
3
4. Implement the deep learning models for
• CNN regression for image positioning
• GAN-based techniques to generate synthesized defects for
the data-imbalance problem
• Autoencoding models
– Convolutional Autoencoders (CAE) with one-class SVM classification
for unsupervised anomaly detection
– Variational Autoencoders (VAE) image reconstruction for unsupervised
defect detection
4
6. 6PCB assembly PCB inspection
Deep learning for PCB positioning
A precise positioning system can be used for
Automated assembly
Automated visual defect detection
Bare PCB Assembled PCB
7. Problem
Predict Angle (θ)、Horizontal displacement (x) and Vertical displacement (y)
of a PCB w.r.t. the template.
7
(θ, x, y)=(-20°, -20, -20) (θ, x, y)=(4°, 0, 12) (θ, x, y)=(20°, 20, 20)
(θ, x, y)=(0°, 0, 0)PCB board
Deep learning for PCB positioning
Template
8. 8
Traditional machine learning for regression
Deep learning for regression
Deep learning regression v.s. traditional ML
regression
9. 9
CNN for regression:
CNN as feature extractor
for SVR regression:
DNN for regression:
CNN + SVR for regression:
Using Convolutional Neural Network (CNN)
and Support Vector Regression (SVR) models
to predict θ, x and y.
10. 10
Linear SVR
Nonlinear SVR (with Kernel transformation)
Polynomial:
Radial Basis Function(RBF):
Support Vector Regression (SVR)
x = input feature(s)
y = output response(s)
11. 11
• Input PCB image into DNN model to predict (θ, x, y)
• Number of hidden layers: 3
Deep Neural Network (DNN) model
12. 12
• Input PCB image into CNN model to predict (θ, x, y)
Convolutional Neural Network (CNN) model
13. 13
Note:
1. Kernel function “Radial Basis Function” is used
2. CNN model is the same as the previous one,
the number of features to SVR is 3.
3. Training time: 2.2 hours (for 9,261 samples)
CNN+SVR model for regression
14. 14
CNN as feature extractor for regression
Note:
1. Kernel function “Radial Basis Function” is used
2. CNN model is the same as the previous one; the last layer of CNN
contains 128 nodes (i.e., number of features to SVR is 128)
3. Training time: 2.2 hours (for 9,261 samples)
15. 15
The good thing about Deep learning for Regression:
The user needs only provide ONE SINGLE template image
16. 16
It creates the input image (x) and the corresponding output (θ, x, y) for DL training
Reference PCB image
provided by users:
Augmented image:
Corresponding output: (θ, x, y)=(-20°, -20, -20) (θ, x, y)=(0°, 0, 0) (θ, x, y)=(20°, 20, 20)
180 pixel
180 pixel
The training samples are automatically generated
by Image Augmentation from the template
17. 17
Range of
Training samples
• Image size: 180 × 180 (pixels)
• Angles: 0°、±2°、±4°、 …、± 20° (every two angles for θ)
• Horizontal displacement: 0、±2、±4、…、 ± 20 (pixel) (every two pixels for x)
• Vertical displacement : 0、±2、±4、…、 ± 20 (pixel) (every two pixels for y)
• Number of training samples: 9,261 (21 × 21 × 21)
Testing samples
• Image size :180 × 180 (pixels)
• Angles : 0°、±1°、±2°、…、± 20°
• Horizontal displacement : 0、±1、±2、…、±20 (pixel)
• Vertical displacement : 0、±1、±2、…、±20 (pixel)
• Number of testing samples :68,921 (41 × 41 × 41)
Note: The images with odd θ, x and y are unseen to the training model
18. 18
Evaluate positioning accuracy by mean error, variance and maximum error.
Angle error (degree) Horizontal error (pixel) Vertical error (pixel)
Mean Variance Max Mean Variance Max Mean Variance Max
DNN 0.111 0.008 0.689 0.119 0.009 1.009 0.120 0.008 0.876
CNN 0.124 0.008 0.634 0.162 0.008 0.657 0.133 0.007 0.563
CNN+SVR 0.049 0.001 0.464 0.049 0.001 0.344 0.055 0.002 0.506
CNN as
feature
extractor
0.068 0.003 0.499 0.066 0.003 0.484 0.069 0.003 0.515
Positioning accuracy
19. 19
Model Time (seconds)
DNN 0.00123
CNN 0.00166
CNN+SVR 0.00230
CNN as feature extractor 0.00541
Note: Equipment
1. CPU: Intel® Core™ i7-6700K CPU @ 4.00GHz × 8
2. GPU: GeForce GTX 1080 Ti
Computation time of each model
It achieves 2-milliseconds efficiency.
20. 20
Defect inspection by image subtraction
Template fT ,
(a) Normal (b) Scratch (c) Extrusion
Test image
,
Image
subtraction
from ,
Result
∆ ,
(d) Intrusion
∆ , fT , ,
Template Aligned
21. Saw-mark defect detection
in heterogeneous solar wafer images using
- GAN-based training samples generation
- CNN classification
21
22. Multicrystalline solar wafer inspection
• Multicrystalline silicon wafers
A multicrystalline solar wafer presents random shapes, sizes and directions of
crystal grains in the surface and results in a heterogeneous texture.
A saw-mark defect is a severe flaw of wafers when cutting the silicon ingot into
thin wafers using the multi-wire saws.
Defect-free solar wafer image
White saw-mark defect Saw-mark defect caused
by impurity
Solar wafer image
with a black saw-mark defect
22
23. The proposed deep learning scheme is
composed of two phases:
Defect samples generation using the CycleGAN (Cycle-consistent Generative Adversarial Networks), and then
Defect detection using the CNN (Convolutional Neural Networks) based on
the true defect-free samples and the synthesized defective samples.
• The CycleGAN model combines both the adversarial loss (i.e. GAN) and the cycle consistency loss .
• GAN measures the adversarial lose between the generated images and the target image.
• The consistency lose prevents the learned forward and backward mappings from contradiction.
• It uses unpaired datasets (not specific paired samples in GAN) for the training, and is suited for our
application.
CycleGAN model used for defect patches generation
23
24. Real solar wafer surfaces
For training the CycleGAN:
• Use a small set of true defect patches (60 for black, 90 for white defects) as the target dataset,
and then randomly collect a small set of defect-free patches (60 & 90) as the input dataset to
the CycleGANs.
Real defect-free samples
Real black saw-mark samples
Real white saw-mark samples
24
25. Using the CycleGAN model to generate
the defective samples
Synthesized defects:
• Whenever we change the input set with different defect-free patches to the trained
CycleGAN, a new defective set is created.
Real defect-free samples input to the trained CycleGAN
Generated black saw-mark patches
Generated white saw-mark patches
25
26. The CNN model for classification
• A simple CNN with 3 convolutional layers is used for the training.
• A lean CNN model gives better computational efficiency in the inspection process.
• Training information:
– For the CycleGAN models, 150 (60 & 90 Black and White sawmarks) real defective
patches and 150 (60&90) real defect-free patches are used as the training samples.
– For the CNN model, a total of 4000 real defect-free patches and 4000 synthesized
sawmark patches are used as the training samples.
– Patch size 50 50 pixels
– Training time : 3 hours for CycleGAN , and 1 hour for CNN
CNN model used for defect detection
26
27. Postprocessing with conventional machine
vision techniques
• The saw-mark in a small windowed patch contains only subtle changes and, thus,
the entire saw-mark region may not be completely detected in the full-sized solar
wafer image.
• Apply the horizontal projection line by line in the resulting binary image B , to
intensify the horizontal saw-mark in the image . That is
P ∑ , , ∀
• The maximum projection value is used as the discriminant measure for saw-mark
detection, i.e. P ∗
P , ∀ . If the horizontal projection P ∗
is large
enough, a saw-mark at line ∗
is declared.
Note: The mean computation time is 0.004 seconds for an image patch of size 50×50
pixels on a PC with an Intel Core 2, 3.6GHz CPU and an NVIDIA GTX 1070 GPU .
27
28. Detection results on sawmarks
• Detection results of defect-free solar wafer images
Test images Detection result projection 28
29. Detection results on sawmarks
)( yP
y
)( yP
y
)( yP
y
• Detection results of defective solar wafer images
Test images Detection result projection 29
30. Detection results on stains / foreign particles
30
Real stains defect samples
Synthesized stain defect samples
31. Detection results on stains / foreign particles
• Detection results of defective solar wafer images
(a) Test images
(b) Detection result
31
32. Additional test: Using the CycleGAN model
to generate bump defects
• Bumps defect
32
Real defect samples
Real defect-free samples
Generated defect samples
33. Autoencoders for defect detection
-Autoencoders for image reconstruction
- Autoencoders for feature extraction
33
35. Defect detection in TFT-LCD
• Thin Film Transistor-Liquid Crystal Display (TFT-LCD) comprises vertical
data lines and horizontal gate lines .
• The main types of defects are pinholes, particles and scratches defect.
defect-free
LCD image
LCD image with
Particle defects
LCD image with
Pinhole defects
LCD image with
Scratch defects 35
36. VAE (Variational AutoEncoder) model for
image reconstraction
- The Model
• Structure of the VAE model
Zp Z p Z
Defect-free Defect-free
Latent variables
36
37. VAE model for image reconstruction
- Detection results by image subtraction
• Defect-free image
• Defect image (true defects)
Original image Restored image Image subtraction Binarization
, , ∆ ,
Original image Restored image Image subtraction Binarization
, , ∆ ,
37
38. Use AE for feature extraction for
anomaly detection (with one-class SVM)
• Testing image is reconstructed from the trained AE model.
• The features are extracted from the last layer of the Encoder , and used as
the input data of the one-class SVM to identify the anomalies.
ZEncoder DecoderEncoder Decoder
Test
image
Anomaly detection
Feature maps
One-class
SVM
Trained AE model
38
39. Use AE for feature extraction for
anomaly detection (one-class SVM)
• One-class SVM (Support Vector Machine)
Ni
NiR
CR
i
ii
N
i
i
,,2,1,0
,,2,1,
s.t.
Min
22
1
2
ax
Ra
Positive samples
Support vector
Outlier
Center point (a)
Radius (R)
a ,
39
40. Use AE for feature extraction for anomaly
detection (one-class SVM) : The Model
• Structure of the AE model:
ZEncoder DecoderEncoder Decoder
Feature maps
41. Use AE for feature extraction for anomaly
detection (one-class SVM) : Training samples (LCD)
Train only positive (defect-free) samples:
− Original 256 256 LCD images are rotated between 0°
and 35°
with 5°
increment.
− 256 random image patches of size 28×28 are used for training.
− Using the AE model to extract the features for the one-class SVM model.
Testing data:
− 160 positive samples
− 33 negative (true defect) samples
Computation time : training 15 minutes , testing 0.00008 seconds
41
42. Use AE for feature extraction for anomaly
detection (one-class SVM) : Detection results (LCD)
• Feature size : 490 (10 feature maps of size 7*7)
• Testing data : 160 positive samples 、 33 negative (true defect) samples
Prediction
Actual Normal Outlier
Normal 89% 11%
Outlier 0% 100%
Type I error 11%
Type II error 0%
Overall recognition rate 90%
42
43. Use AE for feature extraction for anomaly
detection (one-class SVM) : Detection results (LCD)
• Feature size : 49 (use only the best feature map with size 7*7)
Testing data : 160 positive samples , 33 negative (true defect) samples
Prediction
Actual Normal Outlier
Normal 92% 8%
Outlier 0% 100%
Type I error 8%
Type II error 0%
Overall recognition rate 92%
43
44. A thought on Deep Learning:
• Can Deep Learning replace Machine Vision?
• MV as preprocessing , and DL as postprocessing
– MV for defect detection, and DL for defect classification
– Or, vice versa
• MV in DL models?
– “Convolution” and “pooling” are parts of image processing operations
– Human knowledge in DL models
(e.g. embed the known defect features to the DL model)
44