IMPLEMENTASI METODE YOLOv8 UNTUK DETEKSI OBJEK LINTASAN AUTONOMOUS UNDERWATER ROBOT (AUR) PADA EVENT SAUVC

Sajiwo, Erutan Dwi and UNSPECIFIED (2025) IMPLEMENTASI METODE YOLOv8 UNTUK DETEKSI OBJEK LINTASAN AUTONOMOUS UNDERWATER ROBOT (AUR) PADA EVENT SAUVC. Diploma thesis, Politeknik Perkapalan Negeri Surabaya.

Text 0921040008 - Erutan Dwi Sajiwo - Implementasi Metode YOLOv8 untuk Deteksi Objek Lintasan _i_Autonomous Underwater Robot__i_ (AUR) pada _i_Event__i_ SAUVC.pdf Download (4MB)

Abstract

This study implements the YOLOv8 method for object detection in the trajectory of an Autonomous Underwater Robot (AUR) during the Singapore Autonomous Underwater Vehicle Challenge (SAUVC) 2025. The AUR is designed to detect colored objects, such as gates, flares (orange, yelLow, red, blue), and blue drums, to execute autonomous missions, including navigation, target acquisition, target reacquisition, and communication and localization. A dataset comprising 5,175 training images and 2,075 validation images, covering eight classes, was used to train the YOLOv8n model over 150 epochs, achieving an average recall of 91.45%, accuracy of 51.6%, precision of 20.8%, and F1-score of 32.5%. Class imbalance in the dataset led to suboptimal performance, necessitating improved data variety. Integration of the HWT905 Attitude Angle Sensor ensured AUR stability, with an average accuracy of 99.42% and an error of 0.78% across ten trials using a protractor. Real-time testing in the Atlas Sport pool (14x5.5x2.5 m) at night, with a light intensity of 102 lux, demonstrated an average detection accuracy of 86.64% and an error rate of 13.36% over ten trials, enabling the AUR to avoid orange flares, pass through gates, sequentially collide with flares, and drop and retrieve a ball at the blue drum in accordance with SAUVC rules. This research enhances the development of reliable AUR technology for complex underwater environments..

Actions (login required)

View Item