Deep Learning-Based Multi-Class Waste Classification Using the VGG16 Model

Abstract

The manual waste sorting process has faced various challenges, such as low efficiency and a high potential for classification errors. This study aimed to implement and analyze the performance of a deep learning–based VGG16 model for multi-class waste classification using digital images. The dataset used consisted of six waste classes, namely cardboard, glass, metal, paper, plastic, and residual waste, with an imbalanced initial number of images. To address this issue, data augmentation was performed so that each class contained 500 images. The dataset was then divided into 70% training data, 15% validation data, and 15% testing data. The experiments were conducted using a transfer learning approach by varying training parameters, including the RMSProp, Adam, and Stochastic Gradient Descent (SGD) optimizers, as well as batch sizes of 16, 32, and 64. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The results showed that the selection of training parameters significantly affected model performance. The best configuration was achieved using the VGG16 model with the Adam optimizer and a batch size of 16, which produced the highest testing accuracy of 85.87%. This study was expected to serve as a foundation for the development of automated computer vision–based waste sorting systems