KOMBINASI ARSITEKTUR VGG DAN DENSE CONNECTED (DENSENET) PADA MODEL CONVOLUTIONAL NEURAL NETWORK UNTUK SEGMENTASI PEMBULUH DARAH CITRA RETINA

EFRILIYANTI, FILDA and Desiani, Anita and Yahdin, Sugandi (2021) KOMBINASI ARSITEKTUR VGG DAN DENSE CONNECTED (DENSENET) PADA MODEL CONVOLUTIONAL NEURAL NETWORK UNTUK SEGMENTASI PEMBULUH DARAH CITRA RETINA. Undergraduate thesis, Srwijaya University.

	Text RAMA_44201_08011181722068.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (1MB) | Request a copy
	Text RAMA_44201_08011181722068_TURNITIN.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (13MB) | Request a copy
Preview	Text RAMA_44201_08011181722068_0011127702_0027075803_01_front_ref.pdf - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (778kB) | Preview
	Text RAMA_44201_08011181722068_0011127702_0027075803_02.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (536kB) | Request a copy
	Text RAMA_44201_08011181722068_0011127702_0027075803_03.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (210kB) | Request a copy
	Text RAMA_44201_08011181722068_0011127702_0027075803_04.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (922kB) | Request a copy
	Text RAMA_44201_08011181722068_0011127702_0027075803_05.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (88kB) | Request a copy
	Text RAMA_44201_08011181722068_0011127702_0027075803_06_ref.pdf - Bibliography Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (225kB) | Request a copy

Abstract

The addition of layers to the U-Net architecture will cause more parameters and increase network complexity. The Visual Geometry Group (VGG) architecture has the ability to overcome these problems by using a small convolution filter. To overcome excessive feature learning in VGG, Dense Connected (DenseNet) can be used by directly connecting each layer using input from the previous feature map. In this study, we will combine the advantages of VGG and DenseNet in overcoming the shortcomings of the U-Net architecture for segmenting retinal blood vessels. The results of this study obtained an accuracy value of 95.23%, sensitivity of 79.54%, specificity of 97.52%, F1 Score of 80.95%, and Intersection over Union (IoU) of 68% using the DRIVE dataset. From these results it can be concluded that the proposed architecture has succeeded in segmenting retinal blood vessels and predicting the background very well, indicated by the accuracy and specificity values above 90%. In addition, it can predict retinal blood vessels quite well, seen from the sensitivity value above 70%, and the balance between the sensitivity and specificity values is good, seen from the F1 Score value above 80%, but the similarity between the image segmentation results and ground truth is still not good seen from the IoU value below 70%.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	Segmentasi, Pembuluh Darah Citra Retina, U-Net, VGG, DenseNet
Subjects:	Q Science > QA Mathematics > QA299.6-433 Analysis > Q334.A755 Artificial intelligence. Computational linguistics. Computer science. Q Science > QA Mathematics > QA1-939 Mathematics > QA1.T553 Mathematics--Periodicals. Computer science--Periodicals. Computer science. Q Science > QA Mathematics > QA75-76.95 Calculating machines > QA76.9.D343 Data mining. Database searching. Big data.
Divisions:	08-Faculty of Mathematics and Natural Science > 44201-Mathematics (S1)
Depositing User:	Users 3298 not found.
Date Deposited:	19 Oct 2021 06:58
Last Modified:	19 Oct 2021 06:58
URI:	http://repository.unsri.ac.id/id/eprint/56092

Actions (login required)

View Item