Institutional Repository of Vilnius University
Neuroninių tinklų palyginimas rentgeno nuotraukų atpažinimui /
Abstract
dc:descriptionThis bachelor's thesis analyses the calculation models of image similarity metrics of triplet neural network-based neural network models VGG16, DenseNet121 and ResNet50. Graphs depicting their results were formed according to which the calculation models of similarity metrics based on triplet neural networks were compared. The results of the different groups of images they achieved and their totality were discussed. The developed, with data collected for the training and testing of triplet neural networks, which consist of photo triplets, i. e. reduced-dose images, normal-dose images, and normal-dose images that are distant from the respective reduced-dose images in several sections, triplet neural network architectures based on VGG16, DenseNet121 and ResNet50 have been trained and tested. The better the model separates an X-ray image of a normal radiation dose from an X-ray image of a normal radiation dose distant in several sections, the better the model recognizes the images. The experiment showed that the ResNet50 neural network-based similarity metrics best distinguished between similar and dissimilar images, and DenseNet121 did better than VGG16 when evaluated with the data described in section 4.2 of this bachelor's thesis. Similarity metrics based on triplet neural networks can be used to compare X-ray images and evaluate the similarity of high and low dose X-ray images.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Benetis, Algirdas,
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:ELABAETD146240876&prefLang=en_US