Garbage Segregation Using Transfer Learning with Ensemble Technique

Climate change is a challenge for humanity, and recycling can make a significant contribution towards its abatement. The purpose of this work is to develop a model that can detect and classify types of garbage through deep learning algorithms. The dataset consisted of 2467 images, grouped into six categories: cardboard, glass, metal, paper, plastic, and trash. In this work, three pre-trained neural networks, namely, DenseNet121, ConvNextTiny, and RestNet101, have been utilized. For classification, an ensemble model was taken, with 70% of the data for training, 10% for validation, and 20% for testing. All three networks have been subjected to preprocessing with Python, with several operations including scaling, resizing, center cropping, and tensor conversion. Best performance was achieved through hyperparameter search. Augmentation of the data was performed exclusively during training. ConvNextTiny performed best during training and validation but showed a high accuracy of 99.80% when utilized in an ensemble model, and therefore, such a model is an effective one.