http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
SAMT: Shape-aware Multi-task Deep Convolutional Model for Crack Identification
Youssef Oumate(우마트 유셉),Seongmin Kim(김성민),JeanPierre Lomaliza(로말리자 장피에르),Joonhyeok Shin(신준혁),Jiyoung Lee(이지영),Seowoo Han(한서우) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
In this paper, we present a novel pipeline for crack classification using a multi-task model. Our model consists of two branches: a classification branch that distinguishes between crack and no crack images, and a segmentation branch that locates the crack region in the image. By combining these two tasks, our model can focus more on the crack itself and avoid being distracted by the surrounding regions. We also use the segmentation output to extract shape features from the crack region and combine them with the classification output to obtain a final crack score. This way, we can reduce the false positive rate significantly and achieve stateof-the-art results on various public datasets. We demonstrate that our multi-task model for crack identification is simple yet effective, achieving 87.29% and 90.14% Average Precision on SDNet2018 and Concrete Crack Images for Classification datasets, recursively.
균열 검출 알고리즘 고도화 방법: 객체 검출 모델을 활용한 거짓 양성 유발 요소 제거
신준혁(Joonhyeok Shin),우마트 유셉(Youssef Oumate),김성민(Seongmin Kim),로말리자 장피에르(JeanPierre Lomaliza),이지영(Jiyoung Lee),한서우(Seowoo Han) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
This paper introduces a structure removal algorithm to prevent the occurrence of false positive during the exterior wall safety diagnosis by using computer vision. This algorithm proceeds by removing the area where the FP factors are installed based on the object detection model. This algorithm shows more than 85.0 average precision and shows relatively accurate object detection performance. This suggests that more accurate crack detection results can be provided by reducing the false positive occurrence rate in the crack detection algorithm through the corresponding algorithm.
Quantifying Building Crack Widths from Drone Images: A Classification Approach
JeanPierre Lomaliza(로말리자 장피에르),Youssef Oumate(우마트 유셉),Seongmin Kim(김성민),Joonhyeok Shin(신준혁),Jiyoung Lee(이지영),Seowoo Han(한서우) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
Monitoring and quantifying the structural health of buildings is crucial for maintaining their safety and extending their service life. Crack detection and measurement are essential for assessing the state of buildings, but traditional methods are often timeconsuming and labor-intensive. This paper presents a comprehensive pipeline for crack detection and width classification using drone images. The proposed approach involves crack detection with a TransUnet segmentation model, segmentation refinement, and crack width classification into three classes (<0.1 mm, 0.2-0.3 mm, and >0.3 mm). The algorithm achieved 74% classification accuracy, which can be further improved in future studies through refining the mask refinement algorithm and upgrading the quality of hardware used for capturing imagery.