top of page
Illuminated Abstract Shapes_edited.jpg

AI 이미지인식

AI model performance and data validation 

Korea Expert is participating in a project to build learning data

for AI image recognition promoted

by NIA Korea Intelligence and Information Society Agency.

스크린샷 2024-07-11 105255.png

2023'

Building crack detection image [Advanced]
SOC facility crack pattern image data

2022' - 2023'

Using multi-channel complex collection equipment
Ship operation data labeling project underway

2021' - 2022'

Using navigation equipment (ECDIS/ECS)
Maritime environmental conditions and optimal safe port
Labeling business in progress

Training data construction process

ai_11.png

Collection/acquisition
video data

Raw data

그림3.png

Imaging the object appearance section based on raw data

Source data

그림2.png

Label objects based

on raw data

Labeling data

그림1.png

AI model development

Select a model (algorithm) suitable for the task (subject) being performed and verify the constructed data.

(example)

Building crack detection image [Advanced]

Precisely determine the condition of concrete and identify concrete defect areas. Image classification and segmentation model to perform accurate classification.

AlexNet+LeNet(Li+2020)

One of the latest studies on concrete defect image classification

Developed an image classification model using AlexNet and LeNet

A model that obtained scale invariance by applying the concept of convolution to the vision transformer

Image classification model

스크린샷 2024-06-11 104911.png

CvT(Wu+2022)

SegFomer(Xie+2021)

Introducing an encoder with a hierarchical structure to overcome the shortcomings (uni-scale) of the NFL Transformer

​Development of a model to precisely detect concrete cracks based on U-Net, one of the latest artificial neural networks

​Image segmentation model

스크린샷 2024-06-11 104911.png

U-Net(Fan+2020)

bottom of page