At the Institute of Electrical and Electronics Engineers India’s Geoscience And Remote Sensing Symposium 2021 (IEEE InGARSS 2021) held in December 2021, Maitrik Shah, a PhD scholar at Ahmedabad University, won the third prize for his paper titled ‘Deep Learning-based Emulator for 6S Atmospheric Correction Model’. Maitrik authored this paper together with Professor Mehul S Raval, Professor and Associate Dean - Experiential Learning, and Professor Srikrishnan Divakaran, Associate Professor, School of Engineering and Applied Science, Ahmedabad University.
IEEE is the world’s largest technological institution dedicated to the advancement of technology. IEEE InGARSS 2021 comprised a diverse list of activities including sessions from invited speakers, tutorials, technical sessions, and special sessions for young professionals, women in engineering and NGOs.
Maitrik’s paper was the third best, selected from among 25 papers submitted by renowned institutions across India and the world including Liverpool John Moores University, IISc Bangalore, IIT Mumbai, IIT Kanpur, IIT Roorkee and BITS Pilani, to name a few.
Maitrik’s paper proposed making atmospheric correction simpler and more contemporary. “Atmospheric correction, the process of removing the effect of atmospheric molecules on the electromagnetic energy reflected from the earth and retrieving correct surface reflectance values, is necessary and crucial in many remote sensing applications like crop yield prediction, object recognition and change detection. Existing atmospheric correction models are based on the physics-based radiative transfer codes that involve substantial computational costs,” explains Maitrik.
That’s how he decided to bring deep learning into the picture. “Deep learning is a machine learning technique that teaches computers to do what comes naturally to humans: learn by example,” he says. The model suggested by him will try to emulate the behaviour of one of the famous physics-based atmospheric correction models - 6S but by using deep learning. “The output is similar to the 6S model but less time-consuming and less compute-intensive,” adds Maitrik. The PhD scholar is of the opinion that deep learning based models are able to perform atmospheric correction just as physics-based models and in the future may replace the traditional physics-based approach.