Skripsi
DETEKSI DAN LOKALISASI CITRA PRA-KANKER SERVIKS MENGGUNAKAN YOLOV7 DENGAN PENDEKATAN INSTANCE SEGMENTATION.
This research discusses the implementation of the object detection method You Only Look Once (YOLO), capable of performing detection and localization in the form of object segmentation on an image. The input data consists of cervical pre-cancer images, which are processed into YOLO version 7 and YOLO version 8 models for detection and localization. There are four objects or classes to be detected in this study: columnar area (CA), transformation zone (TZ), columnar, and lesions, which are the main focus of this research. The final results of this study include the evaluation of training and testing using unseen data, resulting in output metrics such as mean average precision (mAP), F1 score, precision, recall, confusion matrix, and image prediction results for each model. The YOLOv8x-seg model demonstrates the best performance in object detection and segmentation, achieving mAP(box) accuracy of 85.3% and mAP(mask) accuracy of 64.1%, while achieving mAP(box) accuracy of 63.1% and mAP(mask) accuracy of 59.5% for the lesions class. During the testing phase with unseen data, YOLOv8x-seg achieves mAP(box) accuracy of 78.9% and mAP(mask) accuracy of 70.5%, and for the lesions class, it attains mAP(box) accuracy of 53.4% and mAP(mask) accuracy of 53.4%. The results of this research are expected to aid in lesion detection in cervical pre-cancer images.
Inventory Code | Barcode | Call Number | Location | Status |
---|---|---|---|---|
2307002920 | T123330 | T1233302023 | Central Library (Referens) | Available |
No other version available