<![CDATA[In image classification, convolutional neural networks (CNNs) focus on local patterns, whereas vision Transformers (ViTs) emphasize global context. Combining the two in a hybrid CNN-ViT model may yield a more comprehensive image representation. Waste image classification with visually heterogeneous characteristics can be used to effectively evaluate the performance of the hybrid CNN-ViT model. In addition, such classification supports the crucial need for accurate waste-type identification to enable effective waste management systems. This study investigates a hybrid CNN-ViT model for classifying 24705 organic and recyclable waste images. The workflow involves resizing, an 80:10:10 split, and data augmentation, with models trained for 50 epochs using BCE loss and the Adam optimizer. Evaluation is conducted at the best epoch, defined as the epoch with the highest validation accuracy. For comparison, CNN and ViT models are also trained and evaluated separately. On the test set, the hybrid CNN-ViT model achieves an accuracy of 91.54%, the CNN achieves 91.78%, and the ViT achieves 87.17%. These findings show that CNNs provide an effective and efficient baseline, while the hybrid CNN-ViT model delivers performance competitive with CNNs and is worth considering as a robust alternative for image-based waste classification tasks.]]>
Copyrights © 2026
