Implementasi SSD MobileNetV2 untuk Deteksi Sampah di Industri Daur Ulang Sampah Berbasis Website

Abstract

The high volume of unmanaged waste in Indonesia is a serious environmental problem, where the waste sorting process that is still carried out manually by human labor poses efficiency challenges and high health risks for workers. To address this issue, this research aims to implement a realtime object detection system to help detect waste prior to further waste sorting by actuators. This research focuses on developing a model that is able to detect and classify waste into six main categories, namely cardboard, glass, metal, paper, plastic, and organic waste. The methodology used is to build an object detection model using the SSD (Single Shot MultiBox Detector) algorithm with the MobileNetV2 backbone, which was chosen because of its lightweight and efficient architecture that is suitable for implementation on edge devices and is very fast for realtime detection. The model was built using a transfer learning approach on a diverse image dataset covering all six predefined litter classes. The performance of the model was thoroughly evaluated using standard object detection metrics, with an average precision value across all classes of 96.2%, an average recall of 99.5%, and an average F-1 score of 97.8%. The model achieved an accuracy rate of 95.8% in detecting all six types of waste. The implementation of this system shows great potential to improve the efficiency and accuracy of the waste sorting process, reduce manual workload, and minimize health risks faced by workers in the recycling industry.