CANCIO, JOAO NITOIVAR RIBEIRO (2026) IMPLEMENTATION OF CONVOLUTIONAL NEURAL NETWORK AND GRAY-LEVEL CO-OCCURRENCE MATRIX FOR SKIN DISEASE DETECTION THROUGH IMAGES. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
COVER.pdf Download (845kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
CHAPTER I.pdf Restricted to Registered users only Download (119kB) |
|
|
Text (BAB II)
CHAPTER II.pdf Restricted to Registered users only Download (349kB) |
|
|
Text (BAB III)
CHAPTER III.pdf Restricted to Registered users only Download (236kB) |
|
|
Text (BAB IV)
CHAPTER IV.pdf Restricted to Registered users only Download (293kB) |
|
|
Text (BAB V)
CHAPTER V.pdf Restricted to Registered users only Download (208kB) |
|
|
Text (DAFTAR PUSTAKA)
REFERENCES.pdf Restricted to Registered users only Download (232kB) |
|
|
Text (LAMPIRAN)
ATTACHMENT.pdf Restricted to Registered users only Download (478kB) |
Abstract
Skin diseases pose a significant global health challenge requiring accurate diagnosis. While Convolutional Neural Networks (CNNs) excel in image recognition, they often struggle with lesions sharing similar colours but different textures. This research proposes a hybrid framework integrating CNN spatial feature extraction with Gray-Level Co-occurrence Matrix (GLCM) textural descriptors. The methodology employs a VGG16 backbone for deep features fused with four GLCM statistical properties: Contrast, Homogeneity, Energy, and Correlation. This dual-input architecture analyses both global morphology and surface granularity. Using an 80/20 dataset split, experimental results show the Hybrid CNN-GLCM model significantly outperforms a standalone VGG16 architecture. The proposed model achieved a testing accuracy of 92.43%, a 92.97% improvement over the baseline. The hybrid approach effectively resolved ambiguities between visually similar conditions, proving that fusing handcrafted texture descriptors with deep learning creates a more robust, reliable, and interpretable tool for preliminary skin disease screening. Keywords: Skin Disease Detection, Convolutional Neural Network (CNN), GrayLevel Co-occurrence Matrix (GLCM), Hybrid Model, Feature Fusion, VGG16. Penyakit kulit merupakan tantangan kesehatan global yang signifikan dan membutuhkan diagnosis yang akurat. Meskipun Convolutional Neural Networks (CNN) unggul dalam pengenalan gambar, CNN seringkali kesulitan dalam menangani lesi yang memiliki warna serupa tetapi tekstur yang berbeda. Penelitian ini mengusulkan kerangka kerja hibrida yang mengintegrasikan ekstraksi fitur spasial CNN dengan deskriptor tekstur Gray-Level Co-occurrence Matrix (GLCM). Metodologi ini menggunakan arsitektur VGG16 untuk fitur mendalam yang dipadukan dengan empat properti statistik GLCM: Kontras, Homogenitas, Energi, dan Korelasi. Arsitektur input ganda ini menganalisis morfologi global dan granularitas permukaan. Dengan menggunakan pembagian dataset 80/20, hasil eksperimen menunjukkan bahwa model Hybrid CNN-GLCM secara signifikan mengungguli arsitektur VGG16 mandiri. Model yang diusulkan mencapai akurasi pengujian sebesar 92,43%, peningkatan 92,97% dibandingkan dengan baseline. Pendekatan hibrida secara efektif menyelesaikan ambiguitas antara kondisi yang tampak serupa secara visual, membuktikan bahwa penggabungan deskriptor tekstur buatan tangan dengan pembelajaran mendalam menciptakan alat yang lebih kuat, andal, dan mudah diinterpretasikan untuk skrining awal penyakit kulit. Keywords: Deteksi Penyakit Kulit, Jaringan Saraf Konvolusional (CNN), Matriks Ko-kemunculan Tingkat Abu-abu (GLCM), Model Hibrida, Penggabungan Fitur, VGG16.
Actions (login required)
 |
View Item |
