Super-Resolusi Berbasis Diffusion-Wavelet Pada Citra UAV Untuk Segmentasi Gulma

Fadillah, Aurelia Nuril (2026) Super-Resolusi Berbasis Diffusion-Wavelet Pada Citra UAV Untuk Segmentasi Gulma. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002221086-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Pemetaan gulma berbasis citra UAV penting untuk mendukung pertanian presisi melalui aplikasi herbisida yang tepat sasaran. Namun, keterbatasan kualitas citra UAV dapat menyebabkan hilangnya detail yang diperlukan untuk membedakan gulma dan tanaman padi. Penelitian ini mengembangkan metode super-resolusi berbasis DiffusionWavelet yang mengombinasikan Local Entropy-Based Coupled Anisotropic Diffusion dan 2-Level Discrete Wavelet Transform sebagai tahap pra-pemrosesan citra UAV untuk meningkatkan performansi segmentasi gulma menggunakan YOLOv8-Seg. Tahapan super-resolusi dari metode Diffusion-Wavelet memproses citra low-resolution 384 × 384 menjadi 512×512 melalui tahap upsampling awal dengan metode Iterative Back Projection (IBP) yang dikombinasikan dengan Bilateral Total Variation (BTV), kemudian dilakukan dekomposisi wavelet dua level dan diterapkan Anisotropic Diffusion berbasis local entropy pada sub-band level-2. Metode Anisotropic Diffusion dilengkapi fidelity term dan penguatan adaptif dengan konfigurasi parameter optimal hasil grid search. Selanjutnya citra direkonstruksi kembali menggunakan Inverse Discrete Wavelet Transform (IDWT) sehingga menghasilkan citra super-resolusi berukuran 512×512. Evaluasi dilakukan pada 734 citra dataset WeedyRice-RGBMS-DB menggunakan metrik PSNR dan SSIM untuk kualitas rekonstruksi, serta metrik IoU pada 148 citra uji untuk kinerja segmentasi dengan dua skenario, yaitu image Non SR (512 × 512) dan image SR (768 × 768). Metode Diffusion-Wavelet dengan IBP+BTV sebagai tahap upsampling menghasilkan nilai PSNR sebesar 26,3215 dB dan SSIM sebesar 0,8653. Pada evaluasi segmentasi, penerapan image SR menghasilkan peningkatan Mean IoU dari 0,5438 menjadi 0,6339 (∆ = +0,0901), menunjukkan bahwa metode yang diusulkan mampu meningkatkan akurasi sekaligus konsistensi segmentasi gulma.
======================================================================================================================================
UAV-based weed mapping is crucial for supporting precision agriculture through targeted herbicide application. However, limitations in UAV image quality can result in the loss of details necessary to distinguish between weeds and rice plants. This study develops a ”Diffusion-Wavelet” super-resolution method—combining Local EntropyBased Coupled Anisotropic Diffusion and a 2-Level Discrete Wavelet Transform—as a pre-processing step for UAV imagery to enhance weed segmentation performance using YOLOv8-Seg. The super-resolution stage of the Diffusion-Wavelet method processes lowresolution 384 × 384 images into 512 × 512 images via an initial upsampling phase using Iterative Back Projection (IBP) combined with Bilateral Total Variation (BTV); this is followed by a two-level wavelet decomposition and the application of local entropybased Anisotropic Diffusion to the level-2 sub-bands. The Anisotropic Diffusion method incorporates a fidelity term and adaptive enhancement, utilizing optimal parameter configurations determined through grid search. Subsequently, the image is reconstructed using the Inverse Discrete Wavelet Transform (IDWT) to produce a 512 × 512 superresolution image. Evaluation was conducted on 734 images from the WeedyRice-RGBMSDB dataset using PSNR and SSIM metrics to assess reconstruction quality, and the IoU metric on 148 test images to evaluate segmentation performance across two scenarios: non-SR images (512 × 512) and SR images (768 × 768). The Diffusion-Wavelet method, employing IBP+BTV for the upsampling stage, achieved a PSNR value of 26.3215 dB and an SSIM value of 0.8653. In the segmentation evaluation, the application of SR images resulted in an increase in Mean IoU from 0.5438 to 0.6339 (∆ = +0.0901), demonstrating that the proposed method is capable of improving both the accuracy and consistency of weed segmentation.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Super-resolusi citra, UAV, Diffusion-Wavelet, segmentasi gulma, pertanian presisi, Image super-resolution, UAV, Diffusion-Wavelet, weeds segmentation, precision agriculture
Subjects:	Q Science Q Science > QA Mathematics Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA403.3 Wavelets (Mathematics) T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Aurelia Nuril Fadillah
Date Deposited:	27 Jul 2026 01:32
Last Modified:	27 Jul 2026 01:32
URI:	http://repository.its.ac.id/id/eprint/137534

Actions (login required)

View Item