Institutional Repository of Vilnius University
Vaistų atpažinimas naudojant gilųjį mokymąsi /
Abstract
dc:descriptionThe most common causes of poisoning are pharmaceutical drugs, which are often misused, overdosed or confused because of their visual similarity. This problem could be addressed by making it easy for people to identify medicines before use. This study explores the application of deep learning in the pharmaceutical field by using transfer learning with pre-trained models to develop a system capable of correctly identifying medications from photos. The study attempts to outperform the recognition accuracy of a convolutional neural network based on the ResNet50 model used in one of the previous studies by using more modern models from the EfficientNet family with a different architecture. In addition, a system that adjusts the size of a picture of a drug using a coin as a reference before attempting to recognise it is tested. The models tested showed that any EfficientNet family model outperformed ResNet50 by a wide margin, with the best performing EfficientNet-B7 recognising drugs with an accuracy of 98.11%. When tested using the dataset generated by the size adjustment system, it was found that neural networks with fewer parameters, which were not overfitted to the dataset, gave a significant advantage when the size of the drug was adjusted. EfficientNet B0, B1 and B2 were 1.72%, 1.96% and 1.97% better when the images were adjusted.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Macijauskas, Vilius,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192846893&prefLang=en_US