Wardani, Candra (2026) Deteksi Hama Dan Penyakit Pada Daun Bawang Merah Menggunakan CNN-Transformer Hybrid. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5002221045-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5002221045-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy |
Abstract
Bawang merah (Allium cepa) merupakan komoditas hortikultura bernilai ekonomi tinggi yang produksinya sering terhambat oleh Organisme Pengganggu Tanaman (OPT) seperti ulat grayak, penyakit moler, dan slabung. Pengendalian dan identifikasi OPT yang masih dilakukan secara konvensional oleh petani umumnya kurang efisien karena memakan banyak waktu dan tenaga, sehingga berpotensi menyebabkan kerugian ekonomi yang lebih besar. Oleh karena itu, diperlukan sebuah sistem deteksi hama dan penyakit menggunakan bantuan Computer Vision untuk mendukung penerapan ekosistem Smart Farming. Penelitian ini dilakukan melalui beberapa tahapan, diawali dengan pengumpulan data primer berupa 2.447 citra daun bawang merah (kelas ulat grayak, slabung, dan moler), dilanjutkan tahap pra-pemrosesan yang meliputi pelabelan bounding box, resizing, normalisasi, dan penajaman fitur. Data kemudian dibagi secara proporsional menjadi 70% data latih, 20% data validasi, dan 10% data uji untuk melatih model. Pada arsitektur CNN-Transformer Hybrid yang digunakan, proses ekstraksi fitur diawali oleh modul Convolution Token Embedding (CTE). Selanjutnya, pemrosesan diteruskan pada blok Convolutional Parameter-Sharing Self-Attention (CPSA) yang secara bersamaan memanfaatkan CNN untuk mengekstraksi fitur lokal melalui operasi konvolusi, dan Transformer untuk menangkap konteks global melalui mekanisme self-attention. Tahapan ini diakhiri oleh Local Feed Forward Network (LFFN) guna mempertahankan detail fitur lokal. Melalui mekanisme ini, model mampu mengenali pola dan ciri-ciri visual hama serta penyakit pada daun bawang merah secara lebih tepat. Setelah proses pelatihan, model dievaluasi menggunakan confusion matrix dan Mean Average Precision (mAP) untuk mengukur kinerja prediksi pada kumpulan data uji, baik dari segi klasifikasi kelas maupun ketepatan penentuan posisi objek dengan bounding box. Hasil pengujian menunjukkan bahwa model CNN-Transformer Hybrid berhasil mengungguli model baseline dengan perolehan 0,9007 akurasi dan 0,4060 mAP@0.50.
==================================================================================================================================
Shallot (Allium cepa) is a highly valuable horticultural commodity whose production is frequently hindered by plant-disturbing organisms (OPT) such as beet armyworm, moler disease, and slabung disease. Conventional pest identification and control methods practiced by farmers are generally inefficient, time-consuming, and labor-intensive, potentially leading to greater economic losses. Therefore, a pest and disease detection system utilizing Computer Vision is required to support the implementation of a Smart Farming ecosystem. This research was conducted through several stages, beginning with the collection of primary data consisting of 2,447 shallot leaf images (beet armyworm, slabung, and moler classes), followed by pre-processing steps including bounding box labeling, resizing, normalization, and feature sharpening. The data was then proportionally divided into 70% training data, 20% validation data, and 10% testing data to train the model. In the utilized CNN-Transformer Hybrid architecture, the feature extraction process is initiated by the Convolution Token Embedding (CTE) module. Furthermore, the processing continues to the Convolutional Parameter-Sharing Self-Attention (CPSA) block, which simultaneously utilizes CNN to extract local features through convolution operations, and Transformer to capture global context through the self-attention mechanism. This stage is concluded by the Local Feed Forward Network (LFFN) to preserve local feature details. Through this mechanism, the model is able to recognize visual patterns and characteristics of pests and diseases on shallot leaves more accurately. Following the training process, the model was evaluated using a confusion matrix and Mean Average Precision (mAP) to measure predictive performance on the testing dataset, both in terms of class classification and the accuracy of object positioning with bounding boxes. The evaluation results demonstrate that the CNN-Transformer Hybrid model successfully outperformed the baseline model, achieving an accuracy of 0.9007 and an mAP@0.50 of 0.4060.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Detection, Shallot Leaf Images, CNN-Transformer Hybrid, Plant Pests and Diseases, Smart Farming, Deteksi, Citra Daun Bawang Merah, CNN-Transformer Hybrid, Organisme Pengganggu Tanaman (OPT), Smart Farming. |
| Subjects: | Q Science Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence |
| Divisions: | Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis |
| Depositing User: | Candra Wardani |
| Date Deposited: | 04 Aug 2026 02:24 |
| Last Modified: | 04 Aug 2026 02:24 |
| URI: | http://repository.its.ac.id/id/eprint/142817 |
Available Versions of this Item
- Deteksi Hama Dan Penyakit Pada Daun Bawang Merah Menggunakan CNN-Transformer Hybrid. (deposited 04 Aug 2026 02:24) [Currently Displayed]
Actions (login required)
 |
View Item |