IMPLEMENTASI KOMBINASI ARSITEKTUR VGG DAN SQUEEZENET UNTUK KLASIFIKASI PENYAKIT PADA MATA

SAPUTRA, M ALDI and Suprihatin, Bambang and Amran, Ali (2025) IMPLEMENTASI KOMBINASI ARSITEKTUR VGG DAN SQUEEZENET UNTUK KLASIFIKASI PENYAKIT PADA MATA. Undergraduate thesis, Sriwijaya University.

	Text RAMA_44201_08011382126096.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (2MB) | Request a copy
	Text RAMA_44201_08011382126096_TURNITIN.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (7MB) | Request a copy
	Text RAMA_44201_08011382126096_0026017102_0013126603_01_front_ref.pdf - Accepted Version Available under License Creative Commons Public Domain Dedication. Download (3MB)
	Text RAMA_44201_08011382126096_0026017102_0013126603_02.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (4MB) | Request a copy
	Text RAMA_44201_08011382126096_0026017102_0013126603_03.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (832kB) | Request a copy
	Text RAMA_44201_08011382126096_0026017102_0013126603_04.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (993kB) | Request a copy
	Text RAMA_44201_08011382126096_0026017102_0013126603_05.pdf - Accepted Version Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (264kB) | Request a copy
	Text RAMA_44201_08011382126096_0026017102_0013126603_06_ref.pdf - Bibliography Restricted to Repository staff only Available under License Creative Commons Public Domain Dedication. Download (2MB) | Request a copy

Abstract

Cataract, diabetic retinopathy (DR), and glaucoma are eye diseases that are at risk of blindness. Early detection of these diseases can be done by utilising deep learning methods. This research proposes a combined model of Visual Geometry Group (VGG) and SqueezeNet architecture with the addition of batch normalisation for eye disease classification. The convolution and max pooling layers of the VGG architecture are added in the first block to capture important features in the image in detail. SqueezeNet is added in the middle of the model to reduce the number of parameters and prevent overfitting. In the last block, the model uses fully connected layers and softmax activation function to provide prediction results. Batch normalisation is added to each convolution layer, before the SqueezeNet architecture and before the fully connected layer to help the model converge. The proposed model obtained an average accuracy of 99.5%, sensitivity of 99.2%, specificity of 99.46%, F1-Score of 99.62%, and cohen’s kappa of 78%. In this study, the proposed model works very well in recognising the DR class compared to cataract, glaucoma, and normal classes with a sensitivity value of 100%. Although the values obtained in cataract, glaucoma, and normal classes are still below DR, the results obtained are very good because the sensitivity values for cataract, glaucoma, and normal classes are above 90%. The cohen’s kappa value is still below 80% due to the low cohen’s kappa value in the cataract class, which is only 65%. For further research, architecture development can be carried out to overcome the cohen’s kappa value which is still below 80%.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	Diabetic Retinopathy, Glaukoma, Katarak, Klasifikasi, SqueezeNet, VGG
Subjects:	Q Science > QA Mathematics > QA75-76.95 Calculating machines > QA76.9.B45 Big data. Machine learning. Quantitative research. Metaheuristics.
Divisions:	08-Faculty of Mathematics and Natural Science > 44201-Mathematics (S1)
Depositing User:	M Aldi Saputra
Date Deposited:	30 Jan 2025 08:21
Last Modified:	30 Jan 2025 08:21
URI:	http://repository.unsri.ac.id/id/eprint/167332

Actions (login required)

View Item