research & Publications

Proposes a few-shot learning approach for classifying Resin Identification Codes (RIC) on plastic packaging using a Prototypical Network. The model is trained on a limited dataset of seven plastic categories (PET, HDPE, PVC, LDPE, PP, PS, OTHER) using episodic 7-way 5-shot learning and evaluated across different backbone architectures. Results show EfficientNet-B2 achieves the best performance, demonstrating strong effectiveness for plastic waste classification under low-data conditions and supporting automated recycling systems.

Explores digital forensic and digital evidence research within the legal domain through a curated bibliometric dataset from Lens.org. Surfaces key insights on research trends, leading authors, dominant disciplines, publishing patterns, and international contributions using VOSviewer analysis.