Stitching of Video Sequences for Weed Mapping

Abstract

Weeds are among the most significant and costly environmental threats in Australian agriculture. The distribution, size, density and species of weeds are often heterogeneous in cropping lands. Instead of uniformly spray the same type of herbicide to a whole farm land, using weed maps to do selective spray can reduce herbicide usage and therefore can reduce the serious problems of herbicide resistance, soil damage and food safety. Making weed maps is challenging. The general weed mapping methods heavily rely upon interpolation and hence loss of accuracy. Image stitching technologies have been adopted to make digital mapsm however they have not been well studied for weed mapping. This paper describes a weed mapping method using image stitching technologies to stitch video sequences. In the generated weed maps, the sampling points are continuous and interpolation methods are used at the minimum level and therefore it provide a more accurate way of weed mapping.