Date of Publication

12-6-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electronics and Communications Engineering

Subject Categories

Electrical and Computer Engineering

College

Gokongwei College of Engineering

Department/Unit

Electronics And Communications Engg

Thesis Advisor

Edwin Sybingco

Defense Panel Chair

Argel A. Bandala

Defense Panel Member

Robert Kerwin C. Billones

Renann G. Baldovino

Marielet A. Guillermo

Mark Angelo C. Purio

Abstract (English)

The determination of rice-growing areas and their extent is an important endeavor that aids in formulating data-driven strategies for ensuring food security and economic stability, especially in leading rice-producing and -consuming countries such as the Philippines. Recent studies have utilized a wide spectrum of approaches to rice mapping, from simple classical classification algorithms to more complex deep learning methods. However, existing techniques still struggle to fully leverage the characteristic backscatter dynamics of rice crops, are very sensitive to noise, are constrained by sequential processing, or are too loaded with architectural components that prevent broad usage. In this study, a temporal convolutional network (TCN)-based model is proposed to address these challenges and provide a novel approach to rice mapping with improved accuracy and efficiency over other deep learning methods. It was implemented using Sentinel-1 SAR time series and applied in provinces with diverse agro-ecological conditions. Results show that the model can accurately and efficiently map paddy rice fields, particularly in a province with large and contiguous fields. A limitation of the proposed approach was observed when mapping provinces with small-sized and fragmented fields as they tend to be undetected by Sentinel-1. Overall, the model has demonstrated statistically significant performance improvements compared to other deep learning solutions, including season-to-season generalizability and the potential to be used as a framework for SAR-based rice mapping.

Abstract Format

html

Abstract (Filipino)

Ang kaalaman patungkol sa mga lugar at lawak ng mga taniman ng palay ay mahalagang gawain na nakatutulong sa pagbuo ng mga estratehiyang batay sa datos upang siguruhin ang seguridad sa pagkain at katatagan ng ekonomiya, lalo na sa mga bansang nangunguna sa produksyon at pagkonsumo ng bigas kagaya ng Pilipinas. Ilang mga bagong pag-aaral ang gumamit ng iba't ibang pamamaraan upang tukuyin ang mga taniman ng palay, mula sa simpleng classification algorithms hanggang sa kumplikadong deep learning. Gayunpaman, ang mga kasalukuyang pamamaraang ito ay hirap pa rin samantalahin ang natatanging backscatter dynamics ng mga palay, madaling maapektuhan ng noise, pinababagal ng sequential na proseso, o kaya naman ay maraming arkitektural na bahagi na pumipigil sa malawakang paggamit nito. Sa pag-aaral na ito, isang temporal convolutional network (TCN)-based model ang iminumungkahi upang matugunan ang nabanggit na mga problema at makapagbigay ng makabagong pamamaraan sa pagtukoy ng mga taniman ng palay nang may mas mataas na accuracy at efficiency kaysa sa ibang mga deep learning models. Ito ay naisagawa sa pamamagitan ng SAR time series mula sa Sentinel-1 at ginamit sa mga probinsya na may iba't ibang agro-ecological na kondisyon. Ayon sa kinalabasan ng pag-aaral na ito, ang model ay may kakayahang tukuyin ang mga taniman ng palay nang accurate at efficient, lalo na sa lugar na may malalaki at malalawak na taniman. Isang limitasyon naman ang naobserbahan sa iminumungkahing pamamaraan kapag ito ay ginagamit sa lugar na may maliliit at magkakalayong taniman dahil na rin maaaring hindi nakita ang mga ito gamit ang Sentinel-1. Sa kabuuan, ang model ay nakapagpamalas ng statistically significant na mas mahusay na pagganap kumpara sa ibang deep learning models, nakapagtala ng season-to-season generalizability, at nagpakita ng potensyal na magamit bilang pamamaraan sa pagtukoy ng mga taniman ng palay gamit ang teknolohiya ng SAR.

Abstract Format

html

Language

English

Format

Electronic

Keywords

Synthetic aperture radar

Upload Full Text

wf_yes

Embargo Period

12-12-2025

Download

Share

COinS