A Convolutional Neural Network Approach for Multi-Class Waste Image Classification

Abstract

Waste management has become an increasingly important global issue, requiring effective solutions to support proper waste sorting and recycling processes. Manual waste classification is often inefficient and prone to human error, particularly when dealing with large volumes of waste. In recent years, image classification using deep learning techniques has emerged as a promising approach to automatically identify different types of waste based on their visual characteristics. This study aims to develop and implement a Convolutional Neural Network (CNN) model for multi-class waste image classification. The proposed model utilizes a transfer learning approach based on the EfficientNetV2 architecture. The model was trained using a publicly available waste image dataset obtained from Kaggle consisting of 4,752 images across multiple waste categories. The research process includes data preprocessing, image augmentation, model training, and a series of experimental scenarios to analyze the influence of different hyperparameter configurations, including variations in dropout rate, fine-tuning layers, learning rate, dense layer structures, and L2 regularization. Experimental results show that the best-performing configuration, which applies dropout rates of 0.3 and 0.2, achieved an accuracy of 94.03% with a weighted F1-score of 0.9399. To further evaluate the model’s practical performance, the selected model was tested using 30 real-world waste images captured directly from the surrounding environment. The testing results indicate that the proposed CNN model is capable of classifying waste images effectively and demonstrates promising potential for implementation in automated waste sorting systems.