Plastic litter collection is a time-consuming and expensive task, and its cost is only marginally balanced by the value of recyclable materials, especially in open sea. Orienting the cleanup effort to areas where a significant amount of litter is located can significantly enhance the efficiency of such activity. When we tackle floating debris detection and collection, of course we must take into account that the material is continuously moving under the action of currents and wind, therefore aerial detection should not be aimed at static mapping, but rather used in direct synergy with collection systems to orient them, possibly in a fully automatic fashion, to the areas where a larger amount of waste is present. For this reason, the detection system must operate in real-time, and provide a flow of information continuously to monitoring stations and waste collection vessels. We assembled a stand-alone push-broom multi-purpose hyperspectral sensor system (fig. 1) that contains in particular a spectrometer device operating in the band 900-1700 nm (SWIR – short-wave infrared), with the purpose of realizing a stand-alone payload for a relatively small drone (a DJI Matrice 600). An Intel-based board PC performs data acquisition and storage, and data processing for detection and communication with a ground station. A compact INS provides Global Navigation Satellite System (GNSS)-based geolocation and attitude sensing, for hyperspectral cube generation and georeferentiation.Our detection methodology is based on linear classifiers, obtained by Linear Discriminant Analysis applied on a subset of spectral bands selected by the minimum-redundancy- maximum-relevance algorithm. The 10 nm bands automatically chosen by this algorithm are mostly located between 1180 and 1270 nm and between 1510 and 1620 nm. The classifier is obtained by optimization on a wide set of manually labeled examples taken from data gathered in various sites and environmental conditions. For waste collection assistance purposes, we need to process the data in real-time on-board the drone, and send detection results to a ground station by radio-link. In this case, we do not need high spatial resolution, and even the localization accuracy is not so strict, so that we may process each hyperspectral image (that has a footprint on the surface of several meters across the flight trajectory) at once, giving a single detection response, and associate a relatively rough geo-localization by reading the GNNS position of the drone simultaneously with each image acquisition. The message is received by radio link at the ground station, a computer running a real-time mapping thread that frequently refreshes a map where yellow or red dots are plotted according to the plastics indicator. Of course, the same data flow can be directly forwarded to a waste-collection system. We tested the system in controlled ground environments where we put some plastic litter on purpose (Fig. 2), and on some river and beach sites (both in water and on the shore), obtaining reliable detection even of single objects, such as bottles or pieces of boxes.

Real-time plastic litter detection and localization from drone for cleanup assistance / Balsi, Marco; Bouchelaghem, Soufyane; Moroni, Monica. - (2023). (Intervento presentato al convegno Remote Sensing for Marine Litter Workshop 2023 tenutosi a ESA Conference Bureau Keplerlaan 1, 2201 AZ Noordwijk, The Netherlands).

Real-time plastic litter detection and localization from drone for cleanup assistance

Marco Balsi
;
Soufyane Bouchelaghem
;
Monica Moroni
2023

Abstract

Plastic litter collection is a time-consuming and expensive task, and its cost is only marginally balanced by the value of recyclable materials, especially in open sea. Orienting the cleanup effort to areas where a significant amount of litter is located can significantly enhance the efficiency of such activity. When we tackle floating debris detection and collection, of course we must take into account that the material is continuously moving under the action of currents and wind, therefore aerial detection should not be aimed at static mapping, but rather used in direct synergy with collection systems to orient them, possibly in a fully automatic fashion, to the areas where a larger amount of waste is present. For this reason, the detection system must operate in real-time, and provide a flow of information continuously to monitoring stations and waste collection vessels. We assembled a stand-alone push-broom multi-purpose hyperspectral sensor system (fig. 1) that contains in particular a spectrometer device operating in the band 900-1700 nm (SWIR – short-wave infrared), with the purpose of realizing a stand-alone payload for a relatively small drone (a DJI Matrice 600). An Intel-based board PC performs data acquisition and storage, and data processing for detection and communication with a ground station. A compact INS provides Global Navigation Satellite System (GNSS)-based geolocation and attitude sensing, for hyperspectral cube generation and georeferentiation.Our detection methodology is based on linear classifiers, obtained by Linear Discriminant Analysis applied on a subset of spectral bands selected by the minimum-redundancy- maximum-relevance algorithm. The 10 nm bands automatically chosen by this algorithm are mostly located between 1180 and 1270 nm and between 1510 and 1620 nm. The classifier is obtained by optimization on a wide set of manually labeled examples taken from data gathered in various sites and environmental conditions. For waste collection assistance purposes, we need to process the data in real-time on-board the drone, and send detection results to a ground station by radio-link. In this case, we do not need high spatial resolution, and even the localization accuracy is not so strict, so that we may process each hyperspectral image (that has a footprint on the surface of several meters across the flight trajectory) at once, giving a single detection response, and associate a relatively rough geo-localization by reading the GNNS position of the drone simultaneously with each image acquisition. The message is received by radio link at the ground station, a computer running a real-time mapping thread that frequently refreshes a map where yellow or red dots are plotted according to the plastics indicator. Of course, the same data flow can be directly forwarded to a waste-collection system. We tested the system in controlled ground environments where we put some plastic litter on purpose (Fig. 2), and on some river and beach sites (both in water and on the shore), obtaining reliable detection even of single objects, such as bottles or pieces of boxes.
2023
Remote Sensing for Marine Litter Workshop 2023
04 Pubblicazione in atti di convegno::04d Abstract in atti di convegno
Real-time plastic litter detection and localization from drone for cleanup assistance / Balsi, Marco; Bouchelaghem, Soufyane; Moroni, Monica. - (2023). (Intervento presentato al convegno Remote Sensing for Marine Litter Workshop 2023 tenutosi a ESA Conference Bureau Keplerlaan 1, 2201 AZ Noordwijk, The Netherlands).
File allegati a questo prodotto
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11573/1698849
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact