New mapping solutions based on hyperspectral images acquired from an unmanned aerial systems (drone)

Unmanned Aircraft Systems (UAS), also known as drone, is an emerging technology with a tremendous potential to enable new civilian applications. In a near future, the number and use of UAS will increase. They will play a significant role in many public missions such as border surveillance, wildlife surveys, military training, weather monitoring, and local law enforcement. When equipped with camera, UAS are efficient acquisition platforms for regional mapping and surveying at costs far lower than airborne photogrammetric solutions. They display additional advantages like a rapid and easy deployment allowing frequent and up-to-date data acquisition, as well as acquisitions where and when it is needed. In addition, the lower flight altitude of the UAS provides images with higher spatial resolution than airborne and spaceborne platforms.

Hyperspectral remote sensing is concerned with the extraction of information from objects or scenes lying on the Earth surface, based on their radiance acquired by airborne or spaceborne sensors. It has advanced significantly in the past two decades. The unprecedented spectral and spatial resolutions enable a diversity of applications requiring fine identification of materials or estimation of physical or biophysical parameters. The main sources of difficulties related to hyperpsectral imagery are, the high dimensionality and size of the hyperspectral data, the spectral mixing and the degradation mechanisms associated to the measurement process such as geometric distortions and atmospheric effects.

The proposed research project aims at advancing knowledge related to the design and exploitation of new mapping solutions based on hyperspectral images acquired from an UAS (i.e. drone). It focuses mainly on mining and precision farming applications. Experiments and new algorithm developments will be conducted in order to tackle issues related to image mosaicking, high dimensionality of data, and image radiometric distortion. Dedicated classification approaches will also be devised to map specific phenomenon or materials. Given a set of observations (i.e. pixels vectors in a hyperspectral image), the goal of classification is to assign a unique label to each pixel vector so that it is well defined by a given class. Feature mining, correlation, clustering, and kernel methods are among the main, supervised and unsupervised, techniques targeted for this hyperspectral image classification research. Efforts will also be invested in assessing the precision and performances achievable with such technology. This project is currently carried out in collaboration with other research institutions and centres (ex. INRS-ETE) and may involve industrial partners on the short term.

Faculty Supervisor:

Sylvie Daniel

Student:

Priyadarsini Sivaraj

Partner:

Discipline:

Geography / Geology / Earth science

Sector:

University:

Program:

Globalink