Comparison of YOLOv5 for Classifying Mangrove Leaf Species using CNN-Based

Abstract

Indonesia has many species of mangrove plants scattered throughout the coast to the river's edge. Species of mangrove plants can be distinguished based on root type, stem size, leaf shape, flower color, and fruit. Although each type of mangrove plant has different characteristics, several types look similar, especially on the leaves. Therefore, a model was needed to classify mangrove plant species by applying current technology to make it easier to recognize the type of mangrove plant. This research aims to implement the Convolutional Neural Network (CNN) method in classifying mangrove plant species. The algorithm used is the 5th version of You Only Look Once (YOLO) with 3 different variants (YOLOv5s, YOLOv5m, and YOLOv5l). The three variants have various processing times and numbers of layers. This study uses mangrove leaf images with a total image dataset of 400 images consisting of 4 types of mangrove plants: Avicennia alba, Bruguiera gymnorhiza, Rhizopora apiculata, and Sonneratia alba. The model performance achieved 82.50%, 88.75%, and 93.75% accuracy using YOLOv5s, YOLOv5m, and YOLOv5l, respectively.