Scribd
Upload
40 views

Region-Based Deep Learning Method (Mask R-CNN)

The document discusses the use of Mask R-CNN, a deep learning technique, for detecting and segmenting structural damages. Mask R-CNN is a two-stage detector that uses region proposals, ROIAlign, and a mask prediction head for pixel-level segmentation. It provides advantages over other methods like FCN and achieves state-of-the-art performance. Studies demonstrate Mask R-CNN can accurately detect and measure concrete cracks, roof tile spalling, and refine crack masks using edge filtering for improved identification of building cracks. Future work may involve detecting multiple damage types simultaneously and reducing training time.

Uploaded by

Sal Saad Al Deen Taher
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
40 views

Region-Based Deep Learning Method (Mask R-CNN)

The document discusses the use of Mask R-CNN, a deep learning technique, for detecting and segmenting structural damages. Mask R-CNN is a two-stage detector that uses region proposals, ROIAlign, and a mask prediction head for pixel-level segmentation. It provides advantages over other methods like FCN and achieves state-of-the-art performance. Studies demonstrate Mask R-CNN can accurately detect and measure concrete cracks, roof tile spalling, and refine crack masks using edge filtering for improved identification of building cracks. Future work may involve detecting multiple damage types simultaneously and reducing training time.

Uploaded by

Sal Saad Al Deen Taher
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 18

Region-Based Deep Learning Method

(Mask R-CNN) for detection &


segmentation of Damages in Structures
Presented by
Sal Saad Al Deen Taher
ID: 20ME128
Introduction
Deep Machine Learning Techniques (Convolutional Neural Network):

• Classification
• Object Detection
• Segmentation

Image Courtesy: Machine Learning for Crack Detection: Review and Model Performance Comparison by
Yung-An Hsieh et al. (2020). https://ascelibrary.org/doi/10.1061/%28ASCE%29CP.1943-5487.0000918
Trend Over the Years:

Distribution of ML-based crack detection research. Larger dots represent a higher number of papers
published that year.
Image Courtesy: Machine Learning for Crack Detection: Review and Model Performance Comparison by
Yung-An Hsieh et al. (2020). https://ascelibrary.org/doi/10.1061/%28ASCE%29CP.1943-5487.0000918
Instance Segmentation

Object Detection Semantic Segmentation Instance Segmentation


Person 4

Person 5
Person 4 Person 5
Person Pixels
Person 2 Person 3
Person 1 Person 1 Person 2 Person 3

Image from PASCAL VOC


Mask R-CNN: Model Overview
R-CNN-style detection system
1. Backbone architecture
2. Feature Pyramid Network (FPN) Kaiming Georgia Piotr Ross
He Gkioxari Dollar Girshick
3. Region Proposal Network (RPN)
4. Region of interest feature alignment (RoIAlign)
5. Multi-task network head
Box classifier
Box regressor
Mask predictor
Keypoint predictor
Architecture of Mask R-CNN

Image Courtesy: Review On Autonomous Damage Segmentation And Measurement Of


Glazed Tiles In Historic Buildings Via Deep Learning by Niannian Wang et al. (2019).
https://onlinelibrary.wiley.com/doi/abs/10.1111/mice.12488
Advantages & Disadvantages of Mask R-
CNN
• Pixel Level Image Segmentation
• Two-Stage Proposal Driven Detectors

Disadvantages
• Uses region proposals network to
Advantages

find objects in images. • Slow to train


• More accurate than single stage
detector YOLOv3
• Outperforms all existing, single-model
entries on every task, including the
COCO 2016 challenge winners
Use of Mask R-CNN in
Detecting Structural Damages
Detecting Spalling Damages In Roof Tiles

Detection of Spalling Damages in Complex Background & Lighting Condition

Image Courtesy: Review On Autonomous Damage Segmentation And Measurement Of Glazed Tiles In Historic Buildings
Via Deep Learning by Niannian Wang et al. (2019). https://onlinelibrary.wiley.com/doi/abs/10.1111/mice.12488
Detecting Spalling Damages In Roof Tiles

Measurement result of the four kinds of typical damage


Image Courtesy: Review On Autonomous Damage Segmentation And Measurement Of Glazed Tiles In Historic Buildings
Via Deep Learning by Niannian Wang et al. (2019). https://onlinelibrary.wiley.com/doi/abs/10.1111/mice.12488
Detecting Spalling Damages In Roof Tiles

Comparison results of Mask region-based convolutional neural networks (R-CNN) and full convolution
network (FCN)

Image Courtesy: Review On Autonomous Damage Segmentation And Measurement Of Glazed Tiles In Historic Buildings
Via Deep Learning by Niannian Wang et al. (2019). https://onlinelibrary.wiley.com/doi/abs/10.1111/mice.12488
Concrete Crack Assessment

Test Structure

True Positive False Negative


False Positive

Left Section
Concrete Crack Assessment

Right Section Middle Section


Image Courtesy: Image‐based concrete crack assessment using mask and region‐based convolutional neural network by
Kim et al. (2019). https://onlinelibrary.wiley.com/doi/abs/10.1002/stc.2381
Mask Refinement Using Filter

Framework of the enhanced Mask R-CNN with Sobel filter for


building crack detection

Image Courtesy: Automated Building Crack Identification using Enhanced Mask R-CNN by Lee et al. (2020).
https://www.akademisains.gov.my/asmsj/article/automated-building-crack-identification-using-enhanced-mask-r-cnn/
Mask Refinement Using Filter

YOLOV3 Detection Results Mask R-CNN Detection Results Mask R-CNN after Edge Agreement
Head Applied Detection Results
(a) Crocodile cracks, (b) Longitudinal cracks, (c) Transverse cracks
Future Expected Use for Mask R-CNN
• Detecting other damage types i.e. Rebar Expose, Steel Corrosion, Pit
Corrosion etc.
• Detecting multiple damage types at once and their level of damage.
• Decreasing the training time
• Use in real time damage detection system/application
References
• Yung-An Hsieh and Yichang James Tsai, M.ASCE. Machine Learning for Crack
Detection: Review and Model Performance Comparison. ASCE Journal of Computing
in Civil Engineering-Volume 34 Issue 5. (
https://doi.org/10.1061/(ASCE)CP.1943-5487.0000918)
• Niannian Wang, Xuefeng Zhao, Zheng Zou, Peng Zhao , Fei Qi Autonomous damage
segmentation and measurement of glazed tiles in historic buildings via deep learning.
Computer‐Aided Civil and Infrastructure Engineering Journal Volume35, Issue3 March
2020, Pages 277-291 https://onlinelibrary.wiley.com/doi/abs/10.1111/mice.12488
• Byunghyun Kim & Soojin Cho. Image‐based concrete crack assessment using mask
and region‐based convolutional neural network. Wiley Structural Control Health
Monitoring Journal-Volume 26 Issue 8. (  https://doi.org/10.1002/stc.2381)
• Wei Qin Lee, King Hann Lim, Chin Hong Lim, Wen Loong Lim Automated Building
Crack Identification using Enhanced Mask R-CNN
https://www.akademisains.gov.my/asmsj/article/automated-building-crack-identifi
cation-using-enhanced-mask-r-cnn/

You might also like