Research on Garbage Detection based on Image Feature Differences using Siamese Network Model

The growing volume of garbage disposal creates a rising burden on manual sorting. Existing solutions struggle to identify new types of waste, limiting their effectiveness. This paper proposes a novel approach for classifying garbage discharge using a Siamese Neural Network (SNN) model. The SNN is trained on datasets generated by extracting and annotating regions of interest from collected data. To address data imbalance, we employ Copy Paste augmentation method and data augmentation techniques using torchvision. Our experiments demonstrate that the proposed method using both augmentation methods achieves the highest accuracy, reaching an F1-Score of 94.90%.