20.500.12556/RUL-188222 Avtomatizacija procesa sortiranja svilenih kokonov s pomočjo strojnega učenja Automation of the silk cocoon sorting process using machine learning Trenutno se v Evropi iz več razlogov spet vzpostavlja pridelava svile. Ker je proces proizvodnje svile časovno potraten, ga je potrebno kar se da avtomatizirati. Eden od procesov je sortiranje svilenih kokonov po kvaliteti in napakah, na katerega smo se osredotočili. V nalogi smo razvili stroj in program, za avtomatizirano sortiranje kokonov. Pri tem smo uporabili strojno učenje. Za učenje modelov smo uporabili slike svilenih kokonov in maso le teh. Strojno učenje smo implementirali z več algoritmi, kot so: naključni gozdovi, linearna metoda podpornih vektorjev, metoda podpornih vektorjev z jedrom, stohastični gradientni spust z jedrom, algoritem k-tih najbližjih sosedov in model YOLO26n v kombinaciji z logistično regresijo. Iz eksperimentalnih rezultatov smo razbrali, da se je najbolje odnesla metoda podpornih vektorjev z jedrom, s katero dosežemo natančnost 96 \%. Enaka natančnost je dosežena tudi z našim prilagojenim modelom YOLO26n, ko ga spremenimo tako, da kot vhod upošteva tudi maso. Najvišja storilnost stroja je znašala 15 kokonov na minuto. Silk production is currently being reestablished in Europe for several reasons. Since the process of silk production is time-consuming, it needs to be automated as much as possible. One of the processes is the sorting of silk cocoons by quality and defects, which we focused on. In the assignment, we developed a machine and a program for automated sorting of cocoons. For this we used machine learning. Image data of silk cocoons and their mass were used to train the models. Machine learning was implemented using several algorithms, including random forests, linear support vector machines, kernel support vector machines, kernel stochastic gradient descent, the k-nearest neighbors algorithm, and the YOLO26n model combined with logistic regression. Experimental results indicate that the kernel support vector machine model performed the best, achieving an accuracy of 96 %. The same accuracy was achieved with our customized YOLO26n model, modified to incorporate mass as an input feature. The machine's maximum productivity was 15 cocoons per minute. strojno učenje avtomatizacija kamera strojni vid klasifikacija slik svileni kokoni machine learning automatization camera machine vision image classification silkworm cocoons true false false [P. Spačal] Slovenski jezik Angleški jezik Magistrsko delo/naloga 2026-09-19 08:30:49 2026-09-19 08:30:51 2026-09-21 10:53:53 0000-00-00 00:00:00 2026 0 Ljubljana 0 XX, 63 str. 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.85:677.37(043.2) 290708 291769347 49708.pdf 49708.pdf 1 C8EB5D8B5237E0988A3D10F5226B0D71 daeb8736381fda4592233a44138696ce70c276049645d9343d5bffe50b7f6f11 9c734db6-b3f3-11f1-8bc5-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=248251 Fakulteta za strojništvo 0 0 0