IMPLEMENTASI KLASIFIKASI SAMPAH PLASTIK MENGGUNAKAN YOLO v8 BERBASIS MOBILE

This research focuses on developing a mobile-based application to classify plastic waste using the YOLOv8 model. Given the increasing use of plastic and its environmental impact, innovative solutions are needed to manage plastic waste effectively. The application aims to identify and categorize plastic waste by type using YOLOv8. The research stages include data collection, image annotation, model training, and mobile application development. The dataset comprises 4628 plastic waste images from various sources, augmented through data augmentation processes. The YOLOv8 model is trained using this data on the Google Colab platform, with performance evaluated through metrics like confusion matrix, accuracy, precision, recall, and F1-score. The mobile application, developed using the Flutter framework, utilizes TensorFlow Lite to ensure the model runs efficiently on mobile devices without requiring a server connection. Testing ensures the application functions well on various device types, including those with lower specifications. The author used datasets of 1837 and 4628 images annotated with plastic classes, achieving a training percentage of 92%, validation 4%, and test 4%. The results include a recall value for all classes of 0.94 at a confidence level of 0.0, an F1-score of 0.85%, precision of 1.00 at 0.99%, and precision-recall of 0.87%.

Keywords: Plastic waste classification, YOLOv8, mobile application, deep learning, TensorFlow Lite, Flutter