Identification of Waste Materials in Semi-Controlled Test Site Using UAV Thermal and Multispectral Images

The presence of illegal waste materials is one of the most significant challenges for environmental management and human health. Therefore, their identification reduces environmental hazards significantly. Recently, Unmanned Aerial Vehicles (UAVs) equipped with advanced sensors and specialized machine learning algorithms allow for early identification of illegal waste dumping sites. In this scenario, is pivotal to fine-tuning image-based automatic detection and classification procedures with geo-intelligence data obtained in field acquisition campaign conducted in semi-controlled environment. This study aims to identify illegal waste dumping sites by considering the characteristics of waste materials. The testing activities aimed to evaluate a selected list of UAV payload sensors, and the data derived from them, including thermal and multispectral images, to assess their ability to be utilised for automatic waste detection. For this study, the design and testing of a trial site and relative UAV surveying campaign were conducted to mimic the potential presence of waste materials. Regarding the passive thermal response of the surveyed site, a convergence procedure was implemented through R script to calibrate the raw images. Through photogrammetric reconstruction and GNSS-RTK (Global Navigation Satellite System - Network Real Time Kinematic) control point surveying, multiband georeferenced orthomosaic products have been obtained. The calibrated thermal orthophoto was used for the preliminary identification of the waste materials, in particular distinguishing wet sand from dry sand. While the Semi-Automatic Classification Plugin in QGIS was applied to the multispectral orthophoto by utilising materials' spectral signatures to classify the different types of waste materials