Deteksi Real-Time Citra Sampah Menggunakan YOLOv11 untuk Mendukung Sistem Pemilahan Sampah Otomatis

Abstract

The large volume of unmanaged waste in Indonesia poses a significant challenge that requires innovative solutions, particularly in the area of automated waste classification and detection. This study aims to develop an image-based waste detection model using You Only Look Once version 11 (YOLOv11), which is integrated into an automatic waste sorting system to support efficient waste management. The model performs real-time detection of waste into five main categories: paper, cardboard, plastic, bottles, and cans. The dataset consists of 3,088 images collected from Kaggle, direct field collection in Indonesia, and data augmentation techniques to enhance dataset diversity. Experimental results demonstrate high performance, achieving an accuracy of 94.7%, precision of 96.6%, recall of 97.3%, and an F1-score of 96.9%. The optimal configuration was obtained by training the model for 100 epochs with a batch size of 8. The model is implemented within a web-based interface capable of real-time detection using a Fantech Webcam 2K 4MP Luminous C30 QHD 1440P, operating at a resolution of 2560 × 1440 pixels and 30 fps at 1080p. This research contributes to improving the efficiency and accuracy of the waste sorting process while also providing valuable data on the dominant types of waste, which can support informed decision-making in the pursuit of sustainable waste management in Indonesia.