Implementasi Algoritma You Only Look Once untuk Deteksi dan Identifikasi Ikan Hias Air Tawar Secara Realtime

Abstract

There are 2,184 freshwater fish species in Indonesia, and some of them are ornamental fish. Where the number is certainly more than in other countries. For people who don't really know freshwater ornamental fish, it is certainly very difficult to recognize freshwater ornamental fish that they see directly. Therefore, making an android-based realtime system with the YOLOv4-Tiny algorithm as a first step to being able to help people recognize freshwater ornamental fish. YOLOv4-Tiny is a compressed version of YOLOv4 where the architecture of YOLOv4-Tiny is compressed so that it can be used optimally on devices that have low computing power such as mobile devices and embedded systems. YOLOv4- Tiny detects images by dividing the image into S x S grids and each grid is responsible for the object to be detected in it. If each grid finds an object to be detected, the IoU (Intersection Over Union) value will be calculated. If the IoU value is above or equal to 50%, the system will display a bounding box. In this study, the ornamental fish that will be detected are the oranda sakura goldfish, gold oranda goldfish, telescope goldfish and black telescope goldfish. The test results in this study were successful in detecting and identifying freshwater ornamental fish with an accuracy of 85% of the total 2417 data. The system also succeeded in detecting fish taken directly from the android camera or from virtual images.