Images collected by remote sensors placed on airplanes or satellites are useful in many different ways, but they can trick sometimes if errors cannot be screened.
This was a concern to Filipe Condessa, co-author of the paper "Supervised Hyperspectral Image Classification with Rejection”. It received the “Best Student Paper Award” in the IEEE International Geoscience and Remote Sensing Symposium (IGARSS). The paper introduces a framework for robust image classification of remote sensing aerial images. The goal is to classify the image components according to their composition (for example to separate wheat fields from soy fields). This is a way to render the procedure robust to sensor noise and to human errors in the training examples, by considering the option to reject parts of the classification (to abstain from erroneous classifications) and by using contextual cues.
The improved classification performance was not indifferent to Geoscience and Remote Sensing community, among the 99 submitted full papers, and left Filipe Condessa along with his two advisors, José Bioucas-Dias (IST) and Jelena Kovacévič (CMU), happy and proud.
Lisbon Machine Learning School, organized jointly by IST, the Instituto de Telecomunicações and the Spoken Language Systems Lab – L2F of INESC-ID, has started on the 16th July and will go on until the 23rd.
The topic of the school is Natural Language Understanding and participants are quite happy with the speakers, the lectures, which are considering “very exciting” and lab sessions.
This is an event that no long requires promotion as people already search for it even before registrations begin. A participant even said that for ones working and with contact in the area “it is the World Championship for us, it is impossible not to hear about this, everybody talks about it”.
In the halfway of this Machine Learning School, some participants already consider their expectations exceeded.
The event is hosting participants from all 5 continents.