Talk by Dr. Pedram Ghamisi: Bridging the Gap between Earth Observation and Machine Learning: Do We Really Need Deep Learning for Earth Observation?

Speaker: Dr. Pedram Ghamisi is the head of the Machine Learning group at Helmholtz-Zentrum Dresden-Rossendorf, Helmholtz Institute Freiberg for Resource Technology, Freiberg, Germany, and the CTO and co-founder of VasoGnosis Inc, Milwaukee, WI, USA.

Title: Bridging the Gap between Earth Observation and Machine Learning: Do We Really Need Deep Learning for Earth Observation?

Abstract: The field of remote sensing, or Earth observation, provides the possibility to map objects or areas of the Earth from a distance, typically from aircraft or satellites. We are now facing an entirely different scale of the challenge in image interpretation because of the enormous volume and variety of data being generated by Earth observation missions (e.g., multispectral, hyperspectral, RADAR, passive microwave, thermal, and LiDAR). As a consequence, the number of data produced by sensing devices has increased exponentially in the last few decades, creating the “Big Data” phenomenon, and leading to the creation of the new field of “data science”, including the popularization of “machine learning” and “deep learning” algorithms to deal with such data. In contrast with machine learning which is a well-established and ever-growing field of research in the remote sensing community, deep learning at remote sensing is a very young research topic. Although young, a huge number of complex deep learning-based algorithms have been developed in the remote sensing community to tackle a variety of applications such as time-series remotely-sensed data analysis, scene classification, and multi-sensor data fusion.
Surprisingly, ALL the papers, which are recognized as “most popular” by the key journals in the remote sensing community (e.g., IEEE TGRS, IEEE GRSM, and IEEE GRSL), are on the very topic of deep learning in remote sensing (and nothing else)! Several questions now spring to mind, e.g., do we really need deep learning in remote sensing or we only have a tendency toward new, eye-catching trends? How is the performance of new deep approaches compared to well-established machine learning techniques for the analysis of remote sensing data with unique nature? This short presentation tries to answer the aforementioned questions by providing several relevant examples in which deep learning plays a vital role in analyzing remotely-sensed data.

More information about the speaker: http://pedram-ghamisi.com/