ABSTRAK
Pemilahan sampah secara manual memiliki keterbatasan dalam hal efisiensi dan konsistensi sehingga diperlukan sistem deteksi objek otomatis berbasis computer vision. Penelitian ini bertujuan untuk mengimplementasikan dan mengevaluasi algoritma YOLOv8 Nano dalam mendeteksi dan mengklasifikasikan objek sampah secara real-time. Metode penelitian yang digunakan adalah Cross Industry Standard Process for Data Mining (CRISP-DM). Dataset diperoleh dari Kaggle yang terdiri atas 1.281 citra dengan 9 kelas objek sampah. Data melalui tahap pre-processing berupa resize citra menjadi 640 × 640 piksel, normalisasi, dan augmentasi menggunakan Roboflow. Proses pelatihan dilakukan menggunakan Google Colab dengan GPU NVIDIA Tesla T4. Evaluasi performa model dilakukan menggunakan metrik Precision, Recall, dan Mean Average Precision (mAP). Hasil pengujian menggunakan data testing menunjukkan nilai Precision sebesar 49,87%, Recall sebesar 25,38%, [email protected] sebesar 27,48%, dan [email protected]:0.95 sebesar 13,80%. Pengujian real-time menggunakan webcam menunjukkan bahwa model mampu mendeteksi objek berukuran large, sedangkan objek berukuran small dan medium belum berhasil terdeteksi. Hasil penelitian menunjukkan bahwa YOLOv8 Nano dapat diimplementasikan untuk deteksi objek sampah secara real-time, namun masih terdapat keterbatasan dalam mendeteksi objek berdasarkan variasi ukuran.

ABSTRACT
Manual waste sorting has limitations in terms of efficiency and consistency, requiring an automated object detection system based on computer vision. This study aims to implement and evaluate the YOLOv8 Nano algorithm for detecting and classifying waste objects in real time. The research method used is the Cross Industry Standard Process for Data Mining (CRISP-DM). The dataset was obtained from Kaggle and consists of 1,281 images with 9 waste object classes. The data underwent preprocessing, including image resizing to 640 × 640 pixels, normalization, and augmentation using Roboflow. The training process was conducted using Google Colab with an NVIDIA Tesla T4 GPU. Model performance was evaluated using Precision, Recall, and Mean Average Precision (mAP) metrics. The testing results showed a Precision of 49.87%, Recall of 25.38%, [email protected] of 27.48%, and [email protected]:0.95 of 13.80%. Real-time testing using a webcam showed that the model was able to detect large-sized objects, while small- and medium-sized objects were not successfully detected. The results indicate that YOLOv8 Nano can be implemented for real-time waste object detection; however, the model still has limitations in detecting objects with different sizes.