magistrsko delo
Lovro Klenovšek (Author), Jure Žabkar (Mentor)

Abstract

V magistrski nalogi je predstavljena problematika, s katero se soočajo farmacevti, ki izvajajo morfološko analizo farmacevtskih kristalov. Natančna karakterizacija kristalov, kot so velikost in oblika delcev, je ključnega pomena za zagotavljanje kakovosti in učinkovitosti farmacevtskih izdelkov. Tradicionalne metode, kot je ročna analiza slik, posnetih z vrstičnim elektronskim mikroskopom (SEM), so pogosto počasne, subjektivne in nagnjene k napakam. Glavni cilj te naloge je razviti avtomatizirano rešitev, ki bi nadomestila ročne metode in izboljšala natančnost ter hitrost prepoznavanja in merjenja delcev. Za dosego tega cilja so uporabljene metode globokega učenja, pri čemer je implementiranih več naprednih modelov strojnega vida, kot sta LOCA (mreža za štetje objektov z malo primeri z iterativno prilagoditvijo prototipa) in Efficient Segment Anything Model (EfficientSAM). Ta modela omogočata segmentacijo in prepoznavanje objektov na slikah, tudi v primerih, ko imamo malo ali nič učnih podatkov. Rezultati eksperimentov kažejo, da algoritem natančno prepozna in izmeri delce, hkrati pa odpravlja večino napak, ki so pogoste pri ročni analizi. Povprečna napaka pri merjenju dolžine delcev je bila minimalna, najvišja izmerjena odstopanja pa so bila zabeležena pri delcih, kjer je ročna analiza odstopala od dejanskih vrednosti. V teh primerih je algoritem pravilno zaznal delce, kar potrjuje prednosti avtomatizacije nad ročno analizo. Hitrost obdelave je bila znatno izboljšana, saj algoritem deluje skoraj v realnem času, kar omogoča obdelavo večjega števila slik v krajšem času. Tekom naloge smo ugotovili, da lahko sodobni algoritmi globokega učenja bistveno izboljšajo natančnost in učinkovitost pri analizi slik vrstičnega elektronskega mikroskopa. Naša rešitev omogoča natančno prepoznavanje delcev tudi v težkih primerih, kar odpira nove možnosti za avtomatizacijo in izboljšanje procesov v farmacevtski industriji. Rezultati raziskave kažejo, da bi se ta tehnologija lahko uspešno uporabljala v industrijskih okoljih, kjer so hitrost, natančnost in zanesljivost ključni dejavniki.

Keywords

strojni vid;farmacevtski kristali;analiza slik;elektronski mikroskop;magisteriji;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [L. Klenovšek]
UDC: 004.85:615(043.2)
COBISS: 211105795 Link will open in a new window
Views: 38
Downloads: 15
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Scanning electron microscope image analysis using machine learning methods
Secondary abstract: The master's thesis presents the challenges faced by pharmacists conducting morphological analysis of pharmaceutical crystals. Precise characterization of crystals, such as particle size and shape, is crucial for ensuring the quality and effectiveness of pharmaceutical products. Traditional methods, such as manual analysis of images captured by a scanning electron microscope (SEM), are often slow, subjective, and prone to errors. The primary goal of this thesis is to develop an automated solution that would replace manual methods and improve the accuracy and speed of particle recognition and measurement. To achieve this goal, deep learning methods were applied, with several advanced computer vision models implemented, such as LOCA (Low-Shot Object Counting Network with Iterative Prototype Adaptation) and the Segment Anything Model (SAM). These models enable the segmentation and recognition of objects in images, even in cases where little or no training data is available. Experimental results show that the algorithm accurately recognizes and measures particles while eliminating most of the errors commonly associated with manual analysis. The average error in particle length measurement was minimal, with the highest deviations recorded in particles where manual analysis differed from actual values. In these cases, the algorithm correctly identified the particles, further confirming the advantages of automation over manual analysis. The processing speed was significantly improved, as the algorithm operates in near real-time, allowing for the processing of a larger number of images in a shorter time. During the course of the thesis, we found that modern deep learning algorithms can significantly enhance the accuracy and efficiency of image analysis using a scanning electron microscope. Our solution allows for precise particle recognition even in challenging cases, opening up new possibilities for automation and process improvement in the pharmaceutical industry. The results of the research indicate that this technology could be successfully applied in industrial settings, where speed, accuracy, and reliability are key factors.
Secondary keywords: machine vision;computer vision;microscopy;segmentation;deep learning;pharmaceutical crystals;computer science;master's degree;Računalniški vid;Mikroskopija;Farmacija;Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS): Master's thesis/paper
Study programme: 1000471
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 1 spletni vir (1 datoteka PDF (48 str.))
ID: 25194152