Author(s): Mateja Skerjanec; Klemen Kregar; Franci Steinman; Gasper Rak

Linked Author(s): Franc Steinman, GAŠPER RAK

Keywords: In-channel wood; Laser scanning; Machine learning; Volume estimation

Abstract: In-channel wood logs can cause clogging of bridges and other hydraulic structures and therefore floods. It also plays an important role in river ecology, morphology, and hydraulics. Thus, knowledge of its quantity, properties, mobilization, and transport is crucial for managing floating wood-related problems. To automatically identify wood pieces among other floating objects and remotely assess their volumes in an open channel, we have developed a novel methodology based on non-intrusive measuring methods and machine learning and applied it in a laboratory environment. To this end, we tested the combination of an industrial 2D laser scanner, a high-speed camera, and an ultrasonic sensor. Raw data were post-processed with custom-developed algorithms to determine the volumes of samples above the water surface and their intensity histograms. Later on, intensity histograms were analyzed by the machine learning algorithm to distinguish wood pieces from other floating object types. Based on the assigned density, the floating wood sample volumes could be calculated and compared with the predetermined (actual) ones. Results of the laboratory experiment show that the proposed approach enables wood identification with accuracy higher than 97 %, while the average error of floating wood volume calculation based on pre-assigned density and measured floating wood volume above the water surface is about 2 %. The research presents a promising step towards automatic identification and remote volume estimation of in-channel wood, which could be used in real-world applications (e.g., on bridges) for forecasting quantities of downstream in-channel wood.

DOI: https://doi.org/10.3850/IAHR-39WC2521711920221406

Year: 2022