In this paper, we present a method for detecting and classifying objects in organic waste using a deep learning algorithm, Faster R‑CNN with ResNet-50 and Feature Pyramid Network (FPN). This approach leverages the robustness of convolutional neural networks (CNNs) for feature extraction and the efficiency of region proposal networks (RPNs) for identifying object boundaries, providing a comprehensive solution for waste management applications.
Efficient organic waste management is critical for sustainable environmental practices, yet the presence of non-biodegradable impurities poses significant challenges to composting and recycling processes. This study introduces a novel approach to detecting and classifying impurities in organic waste using a deep learning framework, leveraging Faster R‑CNN with ResNet-50 and Feature Pyramid Network (FPN). Our method capitalizes on the robust feature extraction capabilities of ResNet-50 and the multi-scale object detection power of FPN, enabling accurate identification of diverse impurity types within complex waste streams.
A custom dataset was curated, comprising high-resolution images of mixed organic waste annotated with impurity categories, including plastics, metals, and glass. Extensive experiments demonstrate that the proposed model achieves a mean Average Precision (mAP) of 93.00% across impurity classes, significantly outperforming baseline approaches.
This method not only enhances the automation of waste sorting processes but also reduces manual labor and improves recycling efficiency. The findings highlight the potential of integrating deep learning technologies into waste management pipelines, contributing to a circular economy and environmental sustainability. Future work includes expanding the dataset for improved generalization and integrating the system into industrial-scale applications.
