DETEKSI SAMPAH ORGANIK DAN ANORGANIK MENGGUNAKAN MODEL YOLOv8

Bima Prasetio; Nunik Pratiwi;

ABSTRAK

Deteksi Sampah Organik dan Anorganik Menggunakan Model YOLOv8

Bima Prasetio

Penelitian ini bertujuan untuk mengembangkan model deteksi sampah organik dan anorganik
menggunakan algoritma YOLOv8 dengan penambahan noise pada citra seperti normal, gaussian
blur, darkness, low resolution, dan motion blur. Deteksi sampah yang efektif sangat penting untuk
pengelolaan limbah yang berkelanjutan. Dataset yang digunakan terdiri dari 1178 citra yang
dikumpulkan dari Roboflow dan dibagi menjadi set pelatihan, validasi, dan pengujian. Model
YOLOv8 dilatih menggunakan teknik transfer learning selama 30 epoch dengan ukuran gambar
640x640 piksel. Evaluasi dilakukan terhadap metrik performa seperti precision, recall, F1-Score,
mAP, dan IoU dalam berbagai kondisi noise. Hasil penelitian menunjukkan bahwa model mencapai
performa terbaik pada kondisi normal dengan precision sebesar 0,98, recall 1,00, F1-Score 0,99,
dan IoU 0,84. Pada kondisi gaussian blur, model menunjukkan precision 1,00 namun sedikit
penurunan pada recall menjadi 0,93 dan F1-Score 0,96. Kondisi darkness dan low resolution
menunjukkan sedikit penurunan metrik dengan F1-Score masing-masing sebesar 0,98. Kondisi
motion blur adalah yang paling menantang dengan penurunan signifikan pada precision menjadi
0,96, recall 0,96, F1-Score 0,96, dan IoU 0,80.

Kata Kunci: Intersection over Union (IoU), mean Average Precision (mAP), Sampah Organik dan
Anorganik, YOLOv8

ABSTRACT

Detection Organic and Anorganic Waste Using the YOLOv8 Model

Bima Prasetio

This research aims to develop an organic and inorganic waste detection model using the YOLOv8
algorithm with the addition of noise in the image such as nor-mal, gaussian blur, darkness, low
resolution, and motion blur. Effective litter detection is essential for sustainable waste management.
The dataset used consists of 1178 images collected from Ro-boflow and divided into training,
validation, and testing sets. The YOLOv8 model was trained using transfer learning techniques for
30 epochs with an image size of 640x640 pixels. Evaluation was conducted on performance metrics
such as precision, recall, F1-Score, mAP, and IoU under various noise conditions. The results
showed that the model achieved the best performance under normal conditions with a pre-cision of
0,98, recall of 1,00, F1-Score of 0,99, and IoU of 0,84. In the gaussian blur condition, the model
showed a precision of 1.00 but a slight decrease in recall to 0,93 and F1-Score 0,96. Darkness and
low resolution conditions showed a slight decrease in metrics with an F1-Score of 0,98 each. The
motion blur condition was the most challenging with a significant drop in precision to 0,96, recall
0,96, F1-Score 0,96, and IoU 0,80.

Keywords: Intersection over Union (IoU), mean Average Precision (mAP), Organic and Anorganic
Waste, YOLOv8