YOLOv8을 이용한 실시간 함정 추진기 VCIS 탐지 모델 개발 연구

Abstract

This elementary study was conducted to develop a quantitative real-time vibration-captured infrared spectroscopy (VCIS) detection model. In this s tudy, the deep learning model YOLOv8 was used to t rain cavitations and propellers. Among several models with different neural network sizes, an appropriate model was selected based on accuracy and inference time. Subsequently, cavitation detection was performed using a model trained using model tests and full-ship measurement results. The trained model could accurately detect cavitations and propellers in learned cases and cavitations in unlearned cases. Additionally, it could detect cavitations in full-ship measurements. Furthermore, it could detect the cavitation inception speed. These results demonstrate the feasibility of creating a quantitative VCIS detection model using a deep learning-based computer vision model.