Multi-class Waste Classification Using Convolutional Neural Network

• DOI: 10.35762/aer.2024021
• Publication Date: 2024-06-25T08:41:47Z
• Authors: Yanfahmi Rayhan; Achmad Pratama Rifai
• Keywords: Environmental sciences; Deep learning; Convolutional neural network; GE1-350; Full learning; Classification; Environmental technology. Sanitary engineering; TD1-1066; Solid waste; Transfer learning

This study explores a method of waste classification using deep learning, specifically employing the Convolutional Neural Network (CNN). research involves creation unique dataset, hybrid publicly accessible data and newly compiled collection images across 13 classes: paper, glass, wood, metal, clothes, PCB e-waste, non-PCB PET, HDPE, LDPE, PP, PVC, PS. The development CNN model was approached in two ways: transfer learning full learning. In approach, pre-trained models, MobileNetV2 DenseNet121, were utilized. While architecture is constructed sequential method. experimental results indicated that DenseNet121 outperformed others, achieving an impressive accuracy 95.2% average F-1 score 0.95 on test data. closely followed by model, which attained 92% 0.92. comparison, reached 65% 0.65. Generally, models yielded more optimal than those model. efficiency can be attributed to pre-existing knowledge eliminates need learn input patterns from ground up. However, it's important note dataset size 4586 classes may not sufficient for developing robust machine scratch.