99 full papers submitted by students, a panel of 50 reviewers selects
the 10 finalist papers, and the decision of the 2015 Mikio Takagi
Student First Prize was to the paper written by Filipe Condessa, with
his two advisors, José Bioucas-Dias (IST) and Jelena Kovacévič (CMU).
“It feels great to receive this recognition from the Geoscience and
Remote Sensing community,” says Filipe Condessa. IGARRS is the world's
premier symposium on the subject of remote sensing.
Filipe Condessa is on his 4th year as a dual degree doctoral student in
Electrical and Computer Engineering (ECE), at IST and CMU. The paper
“Supervised Hyperspectal Image Classification with Rejection” introduces "a framework for robust image classification of remote sensing images (acquired by satellite or aircrafts),” explains Filipe Condessa. The
first prize was endowed with US$1000.00.
Hosted by the IEEE Geoscience and Remote Sensing Society, the International Geoscience and Remote Sensing Symposium 2015 (IGARSS 2015
) was held from July 26th through Friday July 31th, 2015, at the Convention Center in Milan, Italy. This is the major event in remote sensing and provides an ideal forum for obtaining up-to-date information about the latest developments, exchanging ideas, identifying future trends, and networking with the international geoscience and remote sensing community.CMU Portugal Program: What did you felt when you received the 2015 Mikio Takagi Student First Prize (MTStP)?Filipe Condessa (FC):
I'm very happy for this achievement. It feels great to receive this recognition from the Geoscience and Remote Sensing community. Being awarded this prize while representing my two homes, Carnegie Mellon University and Instituto Superior Técnico (IST-UL), is wonderful. It is reassuring to repay the confidence CMU and IST-UL and my advisors, Jelena Kovacévič and José Bioucas-Dias, gave me for the past four years.CMU Portugal Program: Could you explain the main findings of your paper, which was written with your advisors José Bioucas-Dias (IST) and Jelena Kovačević (CMU)?FC:
The paper introduces a framework for robust image classification of remote sensing images (acquired by satellite or aircrafts), namely hyperspectral images with very rich spectral content. Hyperspectral images are very large and prohibitively expensive to classify manually, so automatic image classification is a very popular area of research. However, automatic (and even manual) classification is a very hard task, and will not infrequently fail. The framework introduced aims to improve classification performance by selectively abstaining from classifications where misclassifications can be expected through the use of rejection, and harnessing contextual information from the image.CMU Portugal Program: What were the main challenges of writing the paper?FC:
There is often a shorter writing cycle for papers submitted to competition. On the other hand, in competitions, you want not only to have a very good paper, but the best paper. This adds some pressure both to the paper writing stage, and to the presentation stage.CMU Portugal Program: What makes this paper distinctive from other in this scientific area?FC:
The paper introduces the use of classification with rejection in the hyperspectral image classification domain, and combines classification with rejection with classification with context, which allows for very interesting results and behaviors at the expense of a higher complexity of the model.CMU Portugal Program: How is this paper related with your Ph.D. studies?FC:
My Ph.D. studies are focused on the design of robust classification systems using rejection and context applied to image classification (biomedical and remote sensing). This paper is very central to my Ph.D. as it presents succinctly the main ideas I'm working on applied to the hard task of hyperspectral image classification.