Enhancing Waste Management with Automated Garbage Bag Classification with ResNet50V2

The aim of this study is to create a robust model to classify synthetic image datasets into three classes: Plastic, Paper and Garbage. However, it is essential to accurately classify waste to improve recycling systems and waste management processes. The dataset contains 15,000 images and is synthetic, ensuring that all classes are represented equally. Pixel rescaling and data augmentation techniques such as random rotations and horizontal flips were preprocessed to strengthen the model’s generalization potential. The dataset was split into 12,000 training images and 3,000 validation images and there was no bias of the evaluation. For feature extraction, a transfer learning-based approach with ResNet50V2, Global Average Pooling and fully connected layers was used for classification. Training the model took place for 50 epochs using Adam optimizer, Sparse Categorical Crossentropy loss function and EarlyStopping and ModelCheckpoint to not overfit. The classification model had an accuracy of 99%, with very high precision, recall and F1 scores for all classes. Confusion matrix analysis showed few misclassifications, with Paper and Plastic Bags being slightly confused because they appear similar. Results of this research show that transfer learning combined with data augmentation is an effective approach to accurate waste classification and has great potential to open the door for automated waste segregation systems. This finding highlights the crucial role of good preprocessing, balanced datasets, and optimized model architectures to solve real world classification problems in the development of smart waste management technologies.