Identifying materials in spatial unauthorized waste sites for local circular economy enhancement

Aerial imagery has emerged as a powerful tool for environmental analysis and decision-making, offering valuable insights for addressing pressing challenges. This study presents a comprehensive approach for performing semantic segmentation on aerial images of illegal disposal of construction waste, with a focus on waste detection and analysis to support circular economy (CE) initiatives. Leveraging the Segment Anything Model (SAM) developed by Meta, we produced highly accurate segmentation masks from aerial drone images. We created a dataset of over 46,000 manually labeled masks, which serve as the foundation for training and evaluation. Then, we fine-tuned the ResNet-50 classification model to classify the masks. By integrating the prediction of the classification model with the detailed segmentation masks, our methodology produced the final waste stream map. The map offers a comprehensive understanding of the open area, allowing for further potential stock analysis and economic evaluation. Overall, the model achieved classification accuracy of 86% for the area and 67% for the masks. This waste identification methodology can be used for economic and environmental decision-making necessary for cleanup operations. The results also allow improved planning for the recovery of potential unused stocks and the treatment of different waste streams, aiding local CE initiatives and waste management strategies. Our model has practical applications for the waste management and recycling sectors as well as municipal and national policymakers.