University of Twente Student Theses


Mapping Crop Types in Smallholder Farming Areas using SAR Imagery with Dynamic Time Warping

Gella, Getachew Workineh (2020) Mapping Crop Types in Smallholder Farming Areas using SAR Imagery with Dynamic Time Warping.

[img] PDF
Abstract:Crop type related information is very essential for various planning and decision support activities in everyday life especially for early forecast and monitoring of food production. Though smallholder farming areas are profound food producers, mapping crop types is mainly constrained by their inherent characteristics like fragmentation (small farm size), rugged terrain, and presence of thick clouds in the growing season. More importantly, crops mixed dominance of the landscape coupled with fragmented holdings, crops behave different phenological characteristics which mostly constrains conventional mapping techniques for crop type mapping. Therefore, the main objective of this study was mapping crop types using all-weather time-series Synthetic Aperture Radar (SAR) with time-weighted Dynamic Time Warping that accounts for phenological development of crops. The study has used Sentinel-1 dual polarimetry (VV, VH) and TerraSAR-X single polarimetry (HH) images. Basic Registration of Crop Plots (BRP) dataset was used as a reference for training and validation. Obtained imagery was passed through a series of pre-processing operations. As Sentinel-1 imagery has dual polarimetry bands, derived features (Ratio, Modified Radar Vegetation Index, and Dual Polarimetric Soil Vegetation Index) were computed. Additionally, polarimetric decomposition was also undertaken. Within stated broader objective, a detailed analysis was done to know crop-specific responses for incident radar signal, to understand the capability of Time-Weighted Dynamic Time Warping for crop type mapping, implications of using either only backscatter bands and inclusion of derived features and decomposed polarimetric features on mapping accuracy of crops. In addition to these, under broader dynamic time warping, two further model improvement strategies (Variable Time Weight Dynamic Time Warping and Angular Metric for Shape Similarity) were also tested for performance. More importantly, the study has investigated an ensemble classifier that integrates TerraSAR-X and Sentinel-1 classification outputs for synergistic use of both sensing systems for crop type mapping. From these analyses, the study has come up with promising findings that show potentials of SAR imagery with time-weighted Dynamic Time Warping for crop type mapping. It has also clearly demonstrated predictive capabilities of either using dual polarimetry or single polarimetry SAR datasets for mapping crops in smallholder farming areas. Finally, by considering achieved outputs and existing caveats on this study, to refine the findings, further works were also recommended.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page