다중 프레임 이미지 분할 기술을 사용한 알루미늄 자동차 도어 결함 감지

Aluminum Car Door Defect Detection by Using Multi-frame Image Segmentation Techniques

  • ;
  • ;
  • ;
  • 김경백 (전남대학교 인공지능융합학과)
  • Ugur Ercelik (Dept. of Artificial Intelligence Convergence, Chonnam National University) ;
  • HaoYu Chen (Dept. of Artificial Intelligence Convergence, Chonnam National University) ;
  • Longfei Li (Dept. of Artificial Intelligence Convergence, Chonnam National University) ;
  • Kyungbaek Kim (Dept. of Artificial Intelligence Convergence, Chonnam National University)
  • 발행 : 2024.10.31

초록

AI-based image detection technology offers a promising solution for identifying defects in car doors, significantly improving efficiency compared to traditional human inspections. This paper introduces an advanced automatic defect detection system utilizing camera-recorded datasets and trained models to identify defects in aluminum car doors. Unlike previous models focused on aluminum castings, this is the first application specifically targeting car doors. Despite progress in defect detection research, challenges such as data imbalance, complex defect characteristics, and limited research on aluminum car doors persist. To address these issues, we propose the LKADenseNet201 model, enhancing the DenseNet201 architecture with a large kernel attention mechanism.While doing this, we focus on 3 important issues: image augmentation, channel attention and model evaluation.Our image processing process mainly include image augmentation. With image augmentation, we aimed to make data diversity suitable for the real world by obtaining data from different angles and to eliminate the imbalance between defect and normal images. This improvement boosts the model's ability to perceive contextual features and increases computational efficiency, essential for detailed spatial understanding and time-critical tasks. Our approach not only enhances operator efficiency but also moves towards automating the inspection process.

키워드

과제정보

This work was supported by the Korea Institute for Advancement of Technology(KIAT) grant funded by the Korea government(MOTIE).(P0024554). This work was supported by Institute of Information & communications Technology Planning & Evaluation (IITP) under the Artificial Intelligence Convergence Innovation Human Resources Development (IITP-2023-RS-2023-00256629) grant funded by the Korea government(MSIT)

참고문헌

  1. Chalapathy, Raghavendra \& Chawla, Sanjay. (2019). Deep Learning for Anomaly Detection: A Survey.
  2. B. Xia, S. Li, N. Li, H. Li and X. Tian, "Surface Defect Recognition of Solar Panel Based on Percolation-Based Image Processing and Serre Standard Model," in IEEE Access, vol. 11, pp. 55126-55138, 2023, doi: 10.1109/ACCESS.2023.3281653.
  3. H. Chen and K. Kim, "Multi-Convolutional Channel Residual Spatial Attention U-Net for Industrial and Medical Image Segmentation," in IEEE Access, vol. 12, pp. 76089-76101,2024, doi: 10.1109/ACCESS.2024.3405957.
  4. HaoYu Chen, Longfei Li, Ugur Ercelik, Kyungbaek Kim. (2024-06-26). LKADenseNet201: A deep learning based defect detection model . Proceedings of the Korea Information Science Society, Jeju.
  5. LongFei Li, HaoYu Chen, Kyungbaek Kim, "SEPtDenseNet201: Car door defect detection model" in JCCI 2024