Investigation of PlanetScope and UAV imagery for automated plastic waste detection in Vientiane

Abstract

Effective monitoring of urban plastic pollution is crucial for sustainable waste management, especially in rapidly urbanising regions. This study, conducted in Vientiane, Laos, focuses on four key components: (1) characterising urban plastic waste using spectral and textural features, (2) evaluating the effectiveness of PlanetScope and UAV imagery (RGB) for plastic waste detection, (3) assessing the accuracy of Random Forest (RF) and Support Vector Machine (SVM) algorithms in classifying plastic, and (4) examining the operability and scalability of these methodologies for urban waste management.

The study identified mean saturation and standard deviation of RGB values as critical spectral features, while contrast and dissimilarity were key textural features for accurately classifying plastic. PlanetScope’s 3-meter resolution was found to be inadequate for detecting urban plastic waste accumulations, whereas UAV imagery (~5 cm resolution) provided more precise data for identifying plastic across the city. The RF and SVM models achieved 86% and 96% accuracy, respectively, during training but experienced significant accuracy drops in real-world applications, with 38% for RF and 28% for SVM, due to misclassification errors, particularly between buildings and plastics. Consequently, this study recommends several improvements for future research.

To enhance waste management practices, this study proposes integrating these methodologies with GIS to optimise plastic waste monitoring and collection strategies. This approach shows promise for scaling across Vientiane, provided that model improvements and increased computational efficiency are achieved. This study offers a foundational framework for advancing urban environmental monitoring and addressing plastic pollution in rapidly developing cities.