Hitra izvedba rešitev strojnega vida na mobilnih napravah

Ban, Jakob (author), Meža, Marko (mentor)

V diplomski nalogi sem obravnaval področje strojnega vida, ki omogoča računalnikom prepoznavo objektov iz slik in video posnetkov. Glede na to, da strojni vid spada pod področje umetne inteligence, sem opisal tudi delovanje strojnega učenja in nevronskih mrež. Glavni cilj naloge je bil izdelati aplikacijo, ki bi s strojnim vidom prepoznavala vrste sadja. Izdelavo sem pričel tako, da sem iz interneta prenesel slikovni podatkovni set, kateremu sem mu za večjo natančnost dodal še slike sadja, ki sem jih posnel sam. Nato sem uporabil Googlov Teachable Machine, ki na podlagi prenesenega učenja omogoča lažje učenje modela slikovnega vida. Za izdelavo aplikacije sem uporabljal Android Studio, v pomoč pa so mi bili internetni viri, ki so podobno aplikacijo že sestavili. Izdelano aplikacijo sem testiral s sadjem, ki bi ga aplikacija morala prepoznati. Ugotovil sem, da deluje zadovoljivo, saj je le nekajkrat napačno identificirala kivi.

• Language: Slovenian
• Keywords: Računalniški vid, Strojno učenje, Prenosno učenje
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FE - Faculty of Electrical Engineering
• Year: 2024
• PID: 20.500.12556/RUL-162169
• COBISS.SI-ID: 208226819
• Publication date in RUL: 19.09.2024

Secondary language

• Language: English
• Title: Fast Implementation of Machine Vision Solutions on Mobile Devices

Abstract

In my diploma thesis, I discussed the field of machine vision, which enables computers to recognize objects from images and videos. Considering that machine vision belongs to the field of artificial intelligence, I also described the operation of machine learning and neural networks. The main goal of the task was to create an application that would recognize types of fruit using machine vision. I started the project by downloading an image data set from the Internet, to which I added fruit images that I took myself for greater accuracy. Then I used Google's Teachable Machine, which makes it easier to learn the image vision model based on transferable learning. I used Android Studio to create the app, and I was helped by internet resources that had already built a similar app. I tested the built application with the fruit that the application should recognize. I found it to work satisfactorily as it only misidentified the kiwi a couple of times.

• Keywords: Computer vision, Machine learning, Transfer learning