An Object Detection Algorithm for Anomaly Detection
- 주제(키워드) Anomaly detection , Oral image , Object detection
- 주제(DDC) 510
- 발행기관 아주대학교
- 지도교수 신동욱
- 발행년도 2023
- 학위수여년월 2023. 8
- 학위명 석사
- 학과 및 전공 일반대학원 수학과
- 실제URI http://www.dcollection.net/handler/ajou/000000032831
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
In this thesis, an anomaly detection algorithm is applied to a mobile oral health care application. Particularly, one class YOLOv3 has been investigated as an anomaly detection model to classify pictures of mouths, which will serve as inputs for subsequent machine learning models. Outstanding performance has been achieved by proposing appropriate annotation strategies for the datasets and modifying the loss function. Notably, the model can classify not only oral and non-oral pictures but also output preprocessed pictures that only contain the area around the lips by using the predicted bounding box. Thus, the model performs both prediction and preprocessing simultaneously.
more목차
1 Introduction 1
2 Related Works 2
3 Methods 5
3.1 Data 5
3.2 Loss function 8
3.3 Training 10
3.4 Prediction 10
4 Results 11
4.1 Experiment 1 : Using SSE loss function 11
4.2 Experiment 2 : Using GIoU and CE loss function 15
4.3 Experiment 3 : Using GIoU, CE and focal loss function 18
4.4 Experiment 4 : Using SSE loss function without classification loss 22
4.5 Experiment 5 : Using GIoU, CE loss function without classifiaction loss 25
4.6 Experiment 6 : Using GIoU,CE and focal loss function without classifiaction loss 28
5 Conclusion 31
6 Appendix 35
국문초록 37
