Analysis of floating objects based on non-intrusive measuring methods and machine learning

Škerjanec, Mateja (avtor), Kregar, Klemen (avtor), Štebe, Gašper (avtor), Rak, Gašper (avtor)

Floating objects in rivers and streams present a growing problem, not only as they may cause clogging of bridges and other hydraulic structures, and consequently floods, but also because they can have a diverse impact on river (and marine) ecosystems, either positive (in case of in-channel wood) or negative (in case of anthropogenic floating objects). To automatically identify different types of floating objects (i.e., wood pieces, EPS and XPS boards, and plastic and metal containers) and their volumes in an open channel, we propose a novel methodology based on non-intrusive measuring methods and machine learning. To this end, we tested the combination of an industrial 2D laser scanner, a high-speed camera, and an ultrasonic sensor. In the laboratory experiment, 36 samples were scanned separately, two to three times in a row, resulting in 77 raw LIDAR clouds and image sequences. Raw data were post-processed with custom-developed algorithms to determine the volumes of samples above the water surface and their intensity histograms. The latter were analyzed with the machine learning algorithm to distinguish between different material types of floating objects. For each of them, the material density was assigned. Based on the identified floating object's material type, pre-assigned density, and measured volume above the water surface, the sample volumes were calculated and compared with the actual ones determined before setting up the experiment. The results show that the proposed approach enables material recognition with accuracy higher than 90%. The average volume calculation error based on detected material type, assigned densities, and measured floating object's volume above the water surface is approx. 2%. The proposed methodology proved promising for automatic differentiation between different types of floating objects and remote measurement of their volume. To use the method in real-world applications (e.g., on bridges) for forecasting downstream quantities of floating objects, and consequently adjusting their management accordingly, additional measurements are needed, focusing on simultaneous scanning of multiple floating objects, under different flow conditions.

• Jezik: Angleški jezik
• Ključne besede: floating objects, laser scanning, machine learning, volume estimation
• Vrsta gradiva: Članek v reviji
• Tipologija: 1.01 - Izvirni znanstveni članek
• Organizacija: FGG - Fakulteta za gradbeništvo in geodezijo
• Status publikacije: Objavljeno
• Različica publikacije: Objavljena publikacija
• Leto izida: 2022
• Št. strani: 12 str.
• Številčenje: Vol. 408, art. 108254
• PID: 20.500.12556/RUL-137051
• UDK: 626/627
• ISSN pri članku: 0169-555X
• DOI: 10.1016/j.geomorph.2022.108254
• COBISS.SI-ID: 105599235
• Datum objave v RUL: 31.05.2022

Gradivo je del revije

• Naslov: Geomorphology
• Skrajšan naslov: Geomorphology
• Založnik: Elsevier
• ISSN: 0169-555X
• COBISS.SI-ID: 9839621

Licence

• Licenca: CC BY 4.0, Creative Commons Priznanje avtorstva 4.0 Mednarodna
• Povezava: http://creativecommons.org/licenses/by/4.0/deed.sl
• Opis: To je standardna licenca Creative Commons, ki daje uporabnikom največ možnosti za nadaljnjo uporabo dela, pri čemer morajo navesti avtorja.

Sekundarni jezik

• Jezik: Slovenski jezik
• Ključne besede: plavje, lasersko skeniranje, strojno učenje, meritve volumna

Projekti

• Financer: ARRS - Agencija za raziskovalno dejavnost Republike Slovenije
• Številka projekta: P2-0180
• Naslov: Vodarstvo in geotehnika: orodja in metode za analize in simulacije procesov ter razvoj tehnologij

• Financer: ARRS - Agencija za raziskovalno dejavnost Republike Slovenije
• Številka projekta: P2-0227
• Naslov: Geoinformacijska infrastruktura in trajnostni prostorski razvoj Slovenije