Vision Transformer Dengan Blok Squeeze-and-Excitation (SE) Untuk Klasifikasi Penyakit Dan Hama Pada Tanaman Jagung

Maharani, Indira Anindha (2025) Vision Transformer Dengan Blok Squeeze-and-Excitation (SE) Untuk Klasifikasi Penyakit Dan Hama Pada Tanaman Jagung. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211058-Undergraduate _Thesis.pdf Restricted to Repository staff only Download (19MB) | Request a copy

Abstract

Indonesia memiliki potensi sumber daya alam yang mendukung sektor pertanian, dengan jagung sebagai salah satu komoditas strategis yang berkontribusi signifikan terhadap Produk Domestik Bruto. Dalam beberapa tahun terakhir, produktivitas jagung nasional mengalami penurunan, salah satunya disebabkan oleh serangan penyakit dan hama. Tindakan pengendalian yang tepat memerlukan informasi jenis penyakit dan hama yang menyerang. Pendekatan klasifikasi kovensional sering kali bersifat manual yang memakan waktu dan bergantung pada kemampuan individu, sehingga diperlukan sistem klasifikasi secara otomatis yang andal. Oleh karena itu, penelitian Tugas Akhir ini memanfaatkan teknologi computer vision dan deep learning, yang menerapkan model Vision Transformer dengan blok Squeeze-and-Excitation dalam klasifikasi penyakit dan hama pada citra tanaman jagung. Integrasi blok SE terhadap model ViT dilakukan agar model dapat membedakan ciri-ciri visual penyakit dan hama yang sering kali serupa, dengan menekankan pada fitur yang penting. Data citra sekunder yang digunakan terdiri atas delapan kelas, yaitu lima kelas penyakit (Nekrosis Letal, Bercak Daun Abu-abu, Hawar Daun, Karat Daun, dan Virus Garis Daun), dan dua kelas hama (Ulat Grayak dan Penggerek Batang), serta satu kelas tanaman sehat. Pada penelitian ini, dilakukan pelatihan terhadap dua skenario model: baseline ViT-B/16 dan ViT-B/16 dengan blok SE. Berdasarkan uji coba yang telah dilakukan, diperoleh model yang terbaik adalah model ViT-B/16 dengan blok SE. Nilai metrik yang dihasilkan adalah accuracy sebesar 93.96%, precision 93.76%, recall 93.69%, dan F1-score 93.72%. Pada penelitian ini, juga dilakukan uji coba terhadap data yang diberikan gangguan noise blur dan noise exposure.
=====================================================================================================================================
Indonesia possesses abundant natural resources that support the agricultural sector, with maize being one of the strategic commodities contributing significantly to the Gross Domestic Product. In recent years, national maize productivity has declined, partly due to pest and disease infestations. Effective control measures require accurate information on the type of disease and pest present. Conventional classification approaches are often manual, time-consuming, and dependent on individual expertise, highlighting the need for a reliable automated classification system. Therefore, this Final Project utilizes computer vision and deep learning technologies by implementing a Vision Transformer model integrated with a Squeeze-and-Excitation block for the classification of diseases and pests in maize plant images. The integration of the SE block into the ViT model aims to enhance the model’s ability to distinguish between visually similar disease and pest patterns by emphasizing critical features. The secondary image dataset used in this study consists of eight classes, including five disease classes (Lethal Necrosis, Gray Leaf Spot, Leaf Blight, Leaf Rust, and Maize Streak Virus), two pest classes (Armyworm and Corn Borer), and one healthy plant class. The training process was conducted on two model scenarios: baseline ViT-B/16 and ViT-B/16 integrated with the SE block. Based on the experimental results, the best-performing model was ViT-B/16 with the SE block, achieving evaluation metrics of accuracy 93.96%, precision 93.76%, recall 93.69%, and F1-score 93.72%. Additionally, this study conducted robustness testing by evaluating the model’s performance on images affected by noise blur and noise exposure disturbances.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Vision Transformer, Squeeze-and-Excitation, Deep Learning, Klasifikasi Penyakit, Hama, Tanaman Jagung, Vision Transformer, Squeeze-and-Excitation, Deep Learning, Disease Classification, Pests, Corn Plant
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.6 Computer programming. S Agriculture > SB Plant culture > SB191.R5 Rice farming
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Indira Anindha Maharani
Date Deposited:	01 Aug 2025 06:14
Last Modified:	01 Aug 2025 06:14
URI:	http://repository.its.ac.id/id/eprint/123011

Actions (login required)

View Item