Iliyas, Kazbek2024-09-252024-09-252024-08-06Iliyas, K. (2024). The Effectiveness of Machine Learning in Construction and Demolition Waste Recognition From Satellite Images in Astana. Nazarbayev University School of Engineering and Digital Scienceshttps://nur.nu.edu.kz/handle/123456789/8268In rapidly urbanizing areas like Astana city, identifying and managing construction and demolition waste (CDW) is becoming more and more difficult. The growing volume and complexity of CDW are excessive for traditional waste management techniques to handle, which results in operational and environmental inefficiencies. To solve the issue, this thesis assesses how well machine learning techniques work in identifying CDW in these cities from satellite images. Accurate detection can greatly help with the management of waste, which is essential for maintaining the health of urban environments. The model was trained with state-of-the-art object recognition and segmentation techniques, resulting in a mean intersection over union (IoU) of 0.380. Although this performance is below the benchmark norms (0.457 to 0.56), as reported in similar research, it still shows great potential. 200 photos were gathered and annotated as part of the process, which was then used to train and validate the model. Key findings include the effect of image quality on detection accuracy and notable differences in performance across various waste types. The model showed an accuracy of 0.80 for both training and validation; however, recall (2.12%) and precision (3.22%) still need to be improved. Some misclassifications were observed during visual inspection since CDW and non-waste materials had similar appearances. To improve detection accuracy, we suggest that future studies look into more sophisticated data augmentation methods and effective model architectures. The use of Google Earth imagery and a simplified two-class classification scheme are two of the study's limitations. These drawbacks imply that to fully reflect the complexity of CDW detection, future research should take multi-class classification into account and include a wider range of data sources. Our findings contribute to the field of environmental monitoring by demonstrating both the potential and challenges of applying machine learning to urban waste management.enAttribution-NonCommercial-ShareAlike 3.0 United StatesType of access: EmbargoConstruction and demolition waste (CDW)Machine learningSatellite imageryObject recognitionSemantic segmentationMaster`s thesis