Enhancing Trash Classification Accuracy through Feature Extraction Techniques

Abstract

This project investigates the performance of MobileNetV2 and a custom CNN model in trash classification, comparing feature extraction techniques such as edge detection, Local Binary Patterns (LBP), and no explicit extraction. Results show that MobileNetV2 without explicit feature extraction achieved the highest accuracy, providing insights into optimizing CNN-based trash classification for real-world waste management applications.

Introduction

Accurate trash classification is vital for promoting recycling and effective waste management. This study evaluates how different feature extraction methods impact the performance of MobileNetV2 and a custom CNN model on a curated trash image dataset.

Methodology

This collaborative research focused on building and training CNN models, with feature extraction techniques like edge detection and LBP, and developing a web application for real-time trash classification.

Authors

• Research Direction: Kaye Akira H. Regulacion
• Technical Development: Heroshi Joe Abejuela