Skip to Main content Skip to Navigation
Theses

Processing of time series of high spatial resolution satellite images through multi-source information fusion for monitoring agricultural practices. the case of sugarcane harvest on reunion island.

Mahmoud El Hajj
Abstract : Time series of satellite images acquired at high spatial resolution constitute an important source of information for monitoring changes over large areas. These data are particularly interesting for agricultural applications. They can be used to track field scale changes of land surface induced by farming practices. However, operational processing of the time series is often limited by the irregularity of acquisitions and the recurring need for expert intervention. This thesis presents an innovative methodology that addresses this issue. The studied application concerns the sugarcane harvest detection on Reunion Island using SPOT images (Kalideos ISLE-Réunion©). To ensure the comparability of the images, we first developed a method for relative radiometric normalization based on an automatic technique of invariant targets selection. Then, we designed and developed a real-time decision system that processes the time series by integrating information from crop growth modeling and expert knowledge. The information fusion formalism we used is based on linguistic modeling and fuzzy logic. This formalism copes with the imprecision, the uncertainty and the heterogeneous nature of the data. It also allows building interpretable decision rules that implement human reasoning. Two methods of rule generation are proposed: the first is manual and relies on expert knowledge, whereas the second is automatic and requires learning data. System performances were evaluated on different study sites and different years. Results are satisfactory: the overall accuracy reaches 98.8 % and the percentage of good harvest detections is 96.1%. The analysis of the contribution of the different sources showed, among other things, that the crop model provides up to 6.4 % of additional precision of harvest detection. The contribution of expert knowledge was difficult to assess because this source is involved in different parts of the system. The methodology presented in this thesis is very promising. It can be applied to other crops and transferred to other agricultural applications like the dynamic mapping of land cover.
Complete list of metadatas

https://pastel.archives-ouvertes.fr/pastel-00005085
Contributor : Ecole Agroparistech <>
Submitted on : Friday, July 24, 2009 - 8:00:00 AM
Last modification on : Monday, October 19, 2020 - 11:11:04 AM
Long-term archiving on: : Sunday, November 27, 2016 - 12:41:10 AM

Identifiers

  • HAL Id : pastel-00005085, version 1

Collections

Citation

Mahmoud El Hajj. Processing of time series of high spatial resolution satellite images through multi-source information fusion for monitoring agricultural practices. the case of sugarcane harvest on reunion island.. Sciences of the Universe [physics]. AgroParisTech, 2008. English. ⟨NNT : 2008AGPT0103⟩. ⟨pastel-00005085⟩

Share

Metrics

Record views

311

Files downloads

4021