Spot The Pineapple: Mapping Land Use Using an Object-Based Classification Technique

Abstract

An object-based classification and a pixel-based classification were compared for classifying land use using SPOT5 satellite imagery. Object-based classification can potentially offer more rapid and realistic capture of land use features than the more traditional method of pixel-based classification. To test the capability of the object-based classification, we chose pineapple crops because they are an intensive crop, they have characteristic spatial features and they are common in south-east Queensland, Australia. A region-merging multi-resolution segmentation and a hierarchical classification was used for the object-based classification using Definiens Professional version 5 and a supervised maximum likelihood classifier (MLC) was employed for the pixel-based classification. The panchromatic band of the SPOT5 image was enhanced by using a Fast Fourier Transform to optimise segmentation of pineapple crops. The Kappa statistic for the object-based classification was 0.84 compared to 0.74 for the pixel-based classification, however less than a day of editing improved the Kappa statistic of the object-based classification to 0.95. We believe this shows that object-based classification offers an improved method for classifying horticultural crops, because of its ability to segment satellite imagery into real-world objects using hierarchical scales, feature shape and context, as well as allowing the user to include expert knowledge and manually edit results.