Avtomatska detekcija lestvic hranilnih vrednosti na živilih s pomočjo računalniškega vida

Drole, Jan (author), Žibert, Janez (mentor), Brodnik, Andrej (comentor)

V diplomski nalogi je predstavljen razvoj sistema za avtomatsko detekcijo prehranskih simbolov, značilnih za sheme Nutri-Score, BIO in V-Label. Glavni cilj je bil vzpostaviti model za prepoznavanje teh simbolov s slik živilskih embalaž na osnovi sistema YOLO. Študija predstavlja uspešno implementacijo modelov YOLOv5 in YOLOv10, pri čemer je podrobno opisan postopek priprave podatkovne zbirke, učenja in integracije modela YOLOv10 v dve aplikaciji: prva služi serijski obdelavi slik znotraj mape, druga pa detekciji simbolov v realnem času prek spletnega vmesnika. Rezultati so pokazali, da sta si modela glede natančnosti zelo podobna, vendar YOLOv10 ponuja za 25% hitrejši čas delovanja, zaradi česar velja kot primernejša izbira za aplikacije v realnem času.

• Language: Slovenian
• Keywords: računalniški, vid, detekcija, yolo
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2024
• PID: 20.500.12556/RUL-162489
• COBISS.SI-ID: 214175235
• Publication date in RUL: 24.09.2024

Secondary language

• Language: English
• Title: Automatic detection of nutrition symbols using computer vision

Abstract

The thesis presents the development of a system for the automatic detection of food labels specific to the Nutri-Score, BIO, and V-Label schemes. The main goal was to establish a model for recognizing these labels on food packaging images based on the YOLO system. The study showcases the successful implementation of YOLOv5 and YOLOv10 models, detailing the process of dataset preparation, training, and integrating the YOLOv10 model into two applications: the first one processes images in bulk within a folder, while the second enables real-time label detection via a web interface. The results show that the two models are very similar in terms of accuracy, but YOLOv10 offers 25% faster performance, making it a more suitable choice for real-time applications.

• Keywords: computer, vision, detection, yolo