A Comparative Study of RGB and Multispectral Sensor Based Cotton Canopy Cover Modelling Using Multi-Temporal UAS Data

Recent years have witnessed enormous growth in Unmanned Aerial System (UAS) and sensor technology, which made it possible to collect high temporal and spatial resolution data over the crops throughout the growing season. We present a comparative study of multispectral and RGB sensor based cotton canopy cover modelling using multi-temporal UAS Data. Moreover, we propose a multi-thresholding based canopy cover model using RGB sensor. For this experiment, the whole study area is divided into approximately one square meter size grids. Later, grid-wise percentage canopy cover is computed using both RGB and multispectral sensors over ten epochs during the growing season of the cotton crop. Initially, ‘Canopeo’ canopy cover classification algorithm is used to extract canopy using RGB images and an empirically determined NDVI value is used as a threshold to extract canopy from multispectral images. Both canopy cover models were validated using manually measured canopy cover. Multispectral sensor based canopy cover model turned out to be more stable and estimated canopy cover accurately. Whereas, RGB based canopy cover model was very unstable and failed to identify canopy when it started changing color. We proposed a multi-thresholding approach to improve RGB based canopy cover modeling to account for change in color in the canopy later in the growing season. As multispectral sensors are more sensitive and expensive, proposed canopy cover model provides an affordable alternate to agriculture scientists and breeders.