Object based image classification: state of the art and computational challenges

As the spatial resolution of satellite remote sensing imagery is advancing towards sub meter, the predominantly pixel based (or single instance) classification methods needs be redesigned to take advantage of the spatial and structural patterns found in the very high resolution imagery. In this work, we look at the advantages of object based image analysis methods through the newer multiple instance learning learning schemes. We analyze these methods in the context of big geospatial data and allude readers to some of the outstanding computational challenges.