<p>Waste classification and recycling are essential strategies for transforming waste into useful and functional waste, that helps protect land, reduce pollution, and improve resource use. However, in the real world, sorting and identifying recyclable wastes faces significant challenges due to the complexity and unpredictability of waste, and the lack of comprehensive waste datasets. These constraints limit the effectiveness of current research efforts in waste management. In this paper, we propose a ‘RecycleTransformerNet’ method for classifying recyclable waste. It uses two different datasets, the TrashNet and Garbage classification datasets, to achieve high performance metrics across a wide range of waste categories. The pre-processing phase includes data normalization and image augmentation steps to improve model performance. The feature extraction is performed using a modified AlexNet framework, known as Trans-AlexNet. The proposed RecycleTransformerNet framework facilitates waste classification at four levels and integrates the external transformer, adaptive patch merging, and pixel transformer. These modules extract hierarchical features, store local information, and allow accurate waste classification. This study uses the cross-entropy loss function for classification tasks. The proposed model is validated on two datasets for recyclable classification, and its performance is compared with other algorithms. Experimental results demonstrate that this model achieves a waste classification accuracy of 99.15% and 99.07% on the TrashNet and Garbage classification datasets, respectively.</p> Graphic abstract <p></p>