MULTI-SENSOR DATA FUSION FOR LAND COVER PARAMETER RETRIEVAL AND CLASSIFICATION

Abstract

With the availability of many di erent sensors, information retrieval from multisensor data and information fusion has become an important aspect in current era. Imaging sensors play a vital role in surveillance and monitoring. Airborne sensors and satellite borne sensors are nowadays employed in day-to-day life for several important applications like land cover monitoring, weather forecasting, cross border surveillance, disaster management etc. Drone is the most used airborne device along with multispectral, hyperspectral sensors and/or thermal sensors in the current decade, and satellite borne sensors includes sensors like Landsat 1-8, Sentinel 1-3, MODIS aqua & terra, ALOS PALSAR 1 & 2, Scatsat-1, SMOS, SMAP etc. These sensors are broadly classi ed as active and passive sensors depending on their mode of imaging and has properties like spatial, spectral and temporal resolution. Sensors are selected based on its type and properties depending on the speci c application, for example in agriculture monitoring optical data with high spatial and high temporal resolution will be best suited. However, such requirements are not satis ed easily due to certain limitations and the concept of using complementary information from di erent sensors through fusion can provide useful information. Combining information from di erent sensors are quite challenging, if the sensors properties like acquisition time, spatial resolution, temporal resolution etc., are di erent. Sensors working in microwave wavelength captures the texture and dielectric property of the target, whereas sensors working in visible wavelength provides re ectance of the target at various spectral bands. Drone based sensors can provide very ne information compared to the satellite based sensors and data can be acquired as many times as the users need. These properties can be exploited for monitoring various land cover types and agriculture applications such as crop-types, soil moisture, leaf area index, vegetation health. Researcher have developed several methods to retrieve these parameters from di erent sensor, however challenge persists in usage of minimum a priori information. Therefore, in this thesis an attempt has been made to use multi sensor data to retrieve land cover parameter with minimum a priori information.