Spectral discrimination and separability analysis of beach macroplatisc litter from high-resolution RPAS images

The extensive presence of macroplastic pollutants along coastlines poses a significant environmental challenge, threatening both human health and coastal ecosystems. Accurate identification of these pollutants is essential for planning effective cleanup activities and improving overall quality of life. Traditionally, in-situ surveys have been the go-to method for locating macroplastics, but they are costly and time-consuming. To overcome these limitations, this study explores the use of Remotely Piloted Aircraft Systems (RPAS) equipped with high-resolution cameras as a cost-effective alternative for mapping macroplastic litter on beaches and distinguishing between different types of contaminants. The Brindisi shoreline was selected as a pilot site, where it was surveyed using a DJI MAVIC MINI drone with an RGB camera. The collected images were processed with Metashape software and ground control points from a GARMIN Forerunner 245 were used for georeferencing. The resulting RGB orthophoto was analyzed using Transformed Divergence and Bhattacharyya Distance criteria to assess the inter-class spectral separability and classification accuracy. The analysis identified 1,154 waste elements, demonstrating that RPAS imagery is effective for detecting macroplastic items. However, the study found that while the RPAS RGB orthophoto was suitable for manual classification, its performance in automatic classification was limited. The separability algorithms used affected the accuracy of the final classification maps, indicating that while RPAS technology is promising, improvements in spectral analysis and classification algorithms are needed for better-automated results.