Updating Farmland Satellite Imagery Using High Resolution Aerial Images

Abstract

This paper presents a new approach for updating farmland satellite imagery by registering it with high resolution aerial images. It is based on Field Boundary Junction (FBJ) as feature for registration. Fields connected to FBJ are grouped together to form FBJ descriptor. Fields are described by sampling Field Boundaries (FB) relative to FBJ locations. Every field gets different description for each FBJ connected to it. It makes FBJ descriptor unique and helps in finding the correct FBJ matches reliably. It is invariant to scaling, in-plane rotation and translation. Experimental results also show its invariance to non rigid transformations, splitting/merging of fields and localization errors in FBs and FBJs.