Refined Freeman-Durden for harvest detection using POLSAR data
Date
2012-10-02
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
To keep up with an ever increasing human population, providing food is one of the main challenges of the current century. Harvest detection, as an input for decision making, is an important task for food management. Traditional harvest detection methods that rely on field observations need intensive labor, time and money. Therefore, since their introduction in early 60s, optical remote sensing enhanced the process dramatically. But having weaknesses such as cloud cover and temporal resolution, alternative methods were always welcomed. Synthetic Aperture Radar (SAR) on the other hand, with its ability to penetrate cloud cover with the addition of full polarimetric observations could be a good source of data for exploration in agricultural studies. SAR has been used successfully for harvest detection in rice paddy fields. However, harvest detection for other crops without a smooth underlying water surface is much more difficult. The objective of this project is to find a fully-automated algorithm to perform harvest detection using POLSAR image data for soybean and corn.
The proposed method is a fusion of Freeman-Durden and H/A/α decompositions. The Freeman-Durden algorithm is a decomposition based on three-component physical scattering model. On the other hand, the H/A/α parameters are mathematical parameters used to define a three-dimensional space that may be subdivided with scattering mechanism interpretations. The Freeman-Durden model has a symmetric formulation for two of its three scattering mechanisms. On the other hand the surface scattering component used by Freeman-Durden model is only applicable to Bragg surface scattering fields which are not the dominant case in agricultural fields. H/A/α can contribute to both of these issues. Based on the RADARSAT-2 images incidence angle, our field based refined Freeman-Durden model and a proposed roughness measure aims to discriminate harvested from senesced crops. We achieved 99.08 percent overall accuracy for cropped corn and 78.76 percent overall accuracy for soybean detection in a two step decision tree. The final conclusion was that C-band SAR was more than adequate for corn discrimination but soybean needs additional source of data.
Description
Keywords
Engineering--Agricultural, Remote Sensing
Citation
Taghvakish, S. (2012). Refined Freeman-Durden for harvest detection using POLSAR data (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. doi:10.11575/PRISM/27570