SVMIRE - An Open Source SVM Image Retrieval with Relevance Feedback System For Earth Observation Data Classification

Abstract

The continuous increase of Earth Observation image acquisitions requires new weakly supervised algorithms for classification and image retrieval. In this paper, we present the architecture of SVMIRE (SVM Image REtrieval with Relevance Feedback) which is a flexible, modular, and fast data mining system based on a relevance feedback approach that increases the performance of the Support Vector Machine (SVM) classifiers. The proposed system has the capability of storing and reusing the obtained classification model, and results. The functionalities of the SVMIRE system are tested on two datasets: one Landsat 8 OLI/TIRS (Operational Land Imager/Thermal Infrared Sensor) and one Sentinel-2 MSI (Multi-Spectral Instrument).