Sistem prepoznavanja in ločevanja odpadkov na osnovi globokega učenja z uporabo mehke robotike

Čampelj, Matej (author), Vrabič, Rok (mentor), Brojan, Miha (comentor)

Največji problem pri sortiranju smeti je prepoznavanje in klasifikacija smeti, ki je lahko direktna glede na fizikalne lastnosti ali pa indirektna s kamero. Pri slednji se po navadi uporabljajo nevronske mreže, naučene na pripravljeni učni množici za prepoznavanje objektov na sliki. V nalogi smo natrenirali algoritem YOLO na javno dostopni učni množici z imenom TACO. Za namen testiranja smo naredili testno robotsko celico, tako da smo izdelali mehko odrivalo in ga namestili na sodelovalnega robota Fanuc CR-7iA/L ter razvili spletno aplikacijo za sledenje in upravljanje sortiranja.

• Language: Slovenian
• Keywords: globoko učenje, prepoznavanje objektov, YOLO, TACO, mehka robotika, sodelovalni roboti
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FS - Faculty of Mechanical Engineering
• Year: 2024
• Number of pages: XX, 59 str.
• PID: 20.500.12556/RUL-161421
• UDC: 007.52:681.52:628.4(043.2)
• COBISS.SI-ID: 218803715
• Publication date in RUL: 11.09.2024

ČAMPELJ, Matej, 2024, Sistem prepoznavanja in ločevanja odpadkov na osnovi globokega učenja z uporabo mehke robotike [online]. Master’s thesis. [Accessed 11 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=161421

Secondary language

• Language: English
• Title: Deep learning-based trash recognition and sorting system using soft robotics

Abstract

The biggest challenge in waste sorting is the recognition and classification of waste, which can be done either directly, based on physical properties, or indirectly, using a camera. The latter usually involves the use of neural networks trained on a prepared training set to recognize objects in the image. We trained the YOLO algorithm on a publicly available dataset called TACO. For testing purposes, we created a test robotic cell by developing a soft pusher, mounting it on the collaborative robot Fanuc CR7iA/L and developed a web application for tracking and managing the sorting process.

• Keywords: deep learning, object recognition, YOLO, TACO, soft robotics, collaborative robots