Segmentasi Area Penyakit Pada Lahan Persawahan Menggunakan Arsitektur U-Net Pada Citra Udara

Laura, Gloria (2025) Segmentasi Area Penyakit Pada Lahan Persawahan Menggunakan Arsitektur U-Net Pada Citra Udara. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211136-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (13MB) | Request a copy

Abstract

Padi merupakan komoditas pangan utama di Indonesia yang memiliki peran krusial dalam ketahanan pangan nasional. Namun, produksi padi kerap menghadapi tantangan serius akibat serangan penyakit tanaman yang dapat menurunkan hasil panen secara signifikan. Deteksi dini terhadap area yang terinfeksi menjadi langkah penting dalam mitigasi kerugian produksi. Seiring kemajuan teknologi pertanian, pemanfaatan citra udara dari Unmanned Aerial Vehicles (UAV) atau drone menjadi solusi potensial untuk mendukung deteksi penyakit secara cepat dan efisien. Penelitian ini menerapkan segmentasi citra berbasis U-Net untuk mendeteksi area lahan persawahan yang terinfeksi penyakit menggunakan citra udara. Model U-Net dikembangkan dalam beberapa varian, yaitu U-Net standar, Attention U-Net, dan Attention U-Net dengan backbone EfficientNetB3, guna mengeksplorasi pengaruh mekanisme attention dan backbone pretrained terhadap akurasi segmentasi. Data yang digunakan merupakan citra udara yang dilabeli dengan bantuan indeks vegetasi VARI (Visible Atmospherically Resistant Index), yang digunakan untuk membedakan area sehat dan terindikasi berpenyakit. Proses penelitian mencakup tahapan pra-pemrosesan citra, pelatihan model , serta evaluasi kinerja model menggunakan metrik Intersection over Union (IoU) dan Dice Similarity Coefficient (DSC). Hasil penelitian Tugas Akhir ini menunjukkan bahwa Attention U-Net dengan backbone EfficientNetB3 memberikan performa terbaik dengan nilai IoU sebesar 0,7020 dan DSC sebesar 0,8231 pada data uji, mengungguli model U-Net standar dan Attention U-Net tanpa backbone.
====================================================================================================================================
Rice is a major food commodity in Indonesia that plays a crucial role in national food security. However, rice production often faces serious challenges due to plant disease attacks that can significantly reduce yields. Early detection of infected areas is an important step in mitigating production losses. As agricultural technology advances, the utilization of aerial imagery from unmanned aerial vehicles (UAVs) or drones is a potential solution to support disease detection quickly and efficiently. This research applies U-Net based image segmentation to detect disease-infected rice field areas using aerial imagery. The U-Net model is developed in several variants, namely standard U-Net, Attention U-Net, and Attention U-Net with backbone EfficientNetB3, to explore the effect of attention mechanism and pretrained backbone on segmentation accuracy. The data used is aerial imagery labeled with the help of the VARI vegetation index (Visible Atmospherically Resistant Index), which is used to distinguish healthy and diseased areas. The research process includes image pre-processing, model training, and model performance evaluation using the Intersection over Union (IoU) and Dice Similarity Coefficient (DSC) metrics. The results of this Final Project research show that the Attention U-Net with backbone EfficientNetB3 provides the best performance with IoU value of 0.7020 and DSC of 0.823 on the test data, outperforming the standard U-Net model and Attention U-Net model without backbone.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Segmentasi Citra, U-Net, Attention U-Net, EfficientNetB3, Deteksi Penyakit Padi, Image Segmentation, U-Net, Attention U-Net, EfficientNetB3, Rice Disease Detection.
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Mathematics and Science > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Gloria Laura
Date Deposited:	23 Jul 2025 04:06
Last Modified:	23 Jul 2025 04:06
URI:	http://repository.its.ac.id/id/eprint/120556

Actions (login required)

View Item