Super Resolusi Berbasis Residual Channel Attention Network (RCAN) Pada Citra Tanaman Jagung Untuk Peningkatan Kinerja Klasifikasi Hama Fall Armyworm

Deborah, Anasthasya Giovany (2026) Super Resolusi Berbasis Residual Channel Attention Network (RCAN) Pada Citra Tanaman Jagung Untuk Peningkatan Kinerja Klasifikasi Hama Fall Armyworm. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002221043-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (27MB) | Request a copy

Abstract

Tanaman jagung merupakan salah satu komoditas pertanian penting di Indonesia yang produktivitasnya dapat menurun akibat serangan hama Fall Armyworm (FAW). Perkembangan teknologi deep learning memungkinkan proses identifikasi hama dilakukan secara otomatis menggunakan citra digital. Namun, citra yang diambil dari jarak tertentu berpotensi memiliki resolusi rendah sehingga detail visual yang diperlukan untuk proses klasifikasi menjadi berkurang. Oleh karena itu, penelitian ini menerapkan metode super resolution menggunakan arsitektur Residual Channel Attention Network (RCAN) untuk meningkatkan kualitas citra sebelum digunakan pada proses klasifikasi hama Fall Armyworm. Pada tahap pelatihan RCAN digunakan teknik augmentasi CutBlur untuk membantu model mempelajari hubungan antara citra beresolusi rendah dan citra beresolusi tinggi. Dataset yang digunakan terdiri atas empat kelas, yaitu egg, frass, larva, dan healthy. Evaluasi model super resolution dilakukan menggunakan metrik PSNR dan SSIM, sedangkan evaluasi klasifikasi menggunakan accuracy, precision, recall, dan F1-score. Hasil penelitian menunjukkan bahwa RCAN dengan skala perbesaran 2 kali menghasilkan performa yang lebih baik dibandingkan skala perbesaran 4 kali dengan nilai PSNR sebesar 30,00 dB dan SSIM sebesar 0,8792. Selanjutnya, citra hasil super resolution digunakan pada proses klasifikasi menggunakan EfficientNetB1, MobileNetV3Large, dan ResNet50. Hasil pengujian menunjukkan bahwa penggunaan citra hasil super resolution mampu meningkatkan performa klasifikasi dan mengurangi kesalahan klasifikasi pada kelas yang memiliki kemiripan visual tinggi, khususnya egg dan frass. Dengan demikian, penerapan RCAN mampu meningkatkan kualitas citra dan memberikan dampak positif terhadap proses klasifikasi hama Fall Armyworm.
=====================================================================================================================================
Maize is one of the most important agricultural commodities in Indonesia, and its productivity can be significantly reduced by Fall Armyworm (FAW) infestations. Advances in deep learning technology have enabled automatic pest identification using digital images. However, images captured from a certain distance may have low resolution, resulting in the loss of visual details required for accurate classification. Therefore, this study applies a super-resolution approach using the Residual Channel Attention Network (RCAN) architecture to enhance image quality prior to the classification process. During RCAN training, the CutBlur augmentation technique was employed to help the model learn the relationship between low-resolution and high-resolution images more effectively. The dataset consisted of four classes: egg, frass, larva, and healthy. The super-resolution model was evaluated using PSNR and SSIM metrics, while the classification models were evaluated using accuracy, precision, recall, and F1-score. The results showed that RCAN with a 2× upscaling factor outperformed the 4× model, achieving a PSNR of 30.00 dB and an SSIM of 0.8792. The super-resolved images were subsequently used for classification using EfficientNetB1, MobileNetV3Large, and ResNet50. Experimental results demonstrated that the use of super-resolved images improved the classification performance and reduced misclassification errors in visually similar classes, particularly egg and frass. These findings indicate that the proposed RCAN-based super-resolution approach can enhance image quality and provide a positive contribution to the classification of Fall Armyworm pests.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Super Resolusi, Residual Channel Attention Network, Hama Fall Armyworm, Klasifikasi, Augmentasi CutBlur, Super Resolution, Residual Channel Attention Network, Fall Armyworm, Classification, CutBlur Augmentation
Subjects:	Q Science Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.6 Computer programming. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) S Agriculture > S Agriculture (General)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Anasthasya Giovany Deborah
Date Deposited:	31 Jul 2026 01:35
Last Modified:	31 Jul 2026 01:35
URI:	http://repository.its.ac.id/id/eprint/140393

Actions (login required)

View Item