Waste classification using deep learning architectures

In recent years, increasing concerns regarding environmental sustainability have made waste management a significant issue. How waste should be properly classified and how the quality of human life can be improved constitute a shared global concern. In particular, the high-accuracy and high-performance classification of organic and recyclable (inorganic) waste is crucial for reducing environmental impacts and economic losses. Traditionally employed manual classification methods are generally prone to errors and inefficiency. The adoption of innovative technologies provides substantial benefits to the recycling sector. Computer vision offers an environmentally friendly classification approach by identifying and categorizing waste materials. The primary aim of this study is to distinguish organic waste from inorganic waste. To achieve this objective, the first dataset used in the study was constructed from the open-source Kaggle dataset "Waste Images," which originally contains 9 classes and 8346 images. The most meaningful and representative images were selected, resulting in a balanced dataset of 580 organic and 575 inorganic images, totaling 1155 samples. The second dataset was created from another open-source Kaggle dataset, "Recyclable Waste Images," which includes 9 classes and 3180 images. From this dataset, 325 images belonging to the "organic" category and 325 images selected from the remaining 8 non-organic classes were used, yielding a total of 650 samples. Using these datasets, the performance of ResNet-50 (Residual Network), YOLOv11 (You Only Look Once, version 11) and Vision Transformer (ViT) models was evaluated. The obtained results demonstrated accuracy scores of 97.84% - 96.20% - 97.66% for the first dataset and 97.85% - 94.21% - 98.00% for the second dataset, respectively. Among these, ViT achieved the most stable and highest performance. Nevertheless, all three models exhibited strong learning capability and high classification accuracy. Unlike multi-class waste classification approaches, this study focuses exclusively on distinguishing between organic and inorganic waste. The clear structural differences between these two types of waste enable more effective separation and contribute directly to various environmental management processes, particularly energy recovery, compost production, and broader sustainability efforts.