Optimasi Deteksi Objek Dengan Segmentasi dan Data Augmentasi Pada Hewan Siput Beracun Menggunakan Algoritma You Only Look Once (YOLO)

Abstract

Significant progress has been achieved in visual detection, and the abundance of remarkable models has been proposed. Object detection is an important task in various popular fields such as medical diagnosis, robot navigation, autonomous driving, augmented reality, and more. This research aims to develop an optimized object detection model with segmentation and augmentation using the YOLO (You Only Look Once) algorithm for recognizing 10 types of toxic snails in images and videos. The dataset consists of 5,720 images that have been augmented using Roboflow, divided into 5,000 images for training, 480 images for validation, and 240 images for testing. With a model training of 50 epochs, YOLOv8 Box_Curve F1-Confidence achieved "0.98 at 0.625", Precision Confidence "1.00 at 0.997", Precision Recall "0.987 mAP@0.5", and Recall Confidence "1.00 at 0.000". Mask_Curve, YOLOv8 achieved "0.98 at 0.625", Precision Confidence "1.00 at 0.997", Precision Recall "0.986 mAP@0.5", and Recall Confidence "1.00 at 0.000".