RAHARDJO, RAFLY WIDYAVIANTO (2026) KLASIFIKASI VARIETAS PISANG MENGGUNAKAN MOBILEVIT-XS DAN PEMBELAJARAN TRANSFER DI BAWAH KONDISI CITRA ALAMI. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
Cover.pdf Download (558kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
Bab 1.pdf Restricted to Registered users only Download (52kB) |
|
|
Text (BAB II)
Bab 2.pdf Restricted to Registered users only Download (335kB) |
|
|
Text (BAB III)
Bab 3.pdf Restricted to Registered users only Download (399kB) |
|
|
Text (BAB IV)
Bab 4.pdf Restricted to Registered users only Download (410kB) |
|
|
Text (BAB V)
Bab 5.pdf Restricted to Registered users only Download (29kB) |
|
|
Text (DAFTAR PUSTAKA)
Daftar Pustaka.pdf Restricted to Registered users only Download (157kB) |
|
|
Text (LAMPIRAN)
Lampiran.pdf Restricted to Registered users only Download (493kB) |
Abstract
The manual identification process of banana varieties is prone to subjectivity and heavily influenced by variations in lighting and image backgrounds. To address this issue, this study develops an automated classification system using the MobileViT-XS architecture and a transfer learning approach. MobileViT-XS was selected due to its computational efficiency for resource-constrained devices, as well as its ability to combine the advantages of local feature extraction from Convolutional Neural Networks (CNNs) with the global contextual understanding of Vision Transformers (ViTs). The dataset used consists of 3,525 images, encompassing four banana varieties (Tanduk, Kepok, Cavendish, and Ambon) compiled from various online sources (e-commerce, Roboflow, web searches, and YouTube), along with the addition of a Background class as negative samples to suppress false positives. Based on the evaluation results, the MobileViT-XS model successfully achieved an accuracy rate of 91%. This achievement proves that the combination of MobileViT-XS and transfer learning is highly effective in recognizing banana varieties under diverse natural image (in-the-wild) conditions, demonstrating significant potential for practical implementation in real-world environments. Keywords: MobileViT-XS, Transfer Learning, Banana Variety Classification, Vision Transformer, Deep Learning, Natural Images Proses identifikasi varietas pisang secara manual rentan terhadap subjektivitas dan sangat dipengaruhi oleh variasi pencahayaan maupun latar belakang citra. Mengatasi hal tersebut, penelitian ini mengembangkan sistem klasifikasi otomatis menggunakan arsitektur MobileViT-XS dan pendekatan transfer learning. MobileViT-XS dipilih karena efisien secara komputasi untuk perangkat bersumber daya terbatas, sekaligus mampu menggabungkan keunggulan ekstraksi fitur lokal dari Convolutional Neural Network (CNN) dengan pemahaman konteks global dari Vision Transformer (ViT). Dataset yang digunakan berjumlah 3.525 citra, meliputi empat varietas pisang (Tanduk, Kepok, Cavendish, dan Ambon) yang dikompilasi dari berbagai sumber daring (e-commerce, Roboflow, web, dan YouTube), serta penambahan kelas Background sebagai negative sample untuk menekan false positive. Berdasarkan hasil pengujian, model MobileViT-XS berhasil mencapai tingkat akurasi sebesar 91%. Pencapaian ini membuktikan bahwa kombinasi MobileViT-XS dan transfer learning sangat efektif dalam mengenali varietas pisang pada kondisi citra alami yang bervariasi, serta berpotensi besar untuk diimplementasikan pada lingkungan nyata. Kata kunci: MobileViT-XS, Transfer learning, Klasifikasi Varietas Pisang, Vision Transformer, Deep learning, Citra Alami
Actions (login required)
 |
View Item |
