Opportunities and limitations of object-based image analysis for detecting urban impervious and vegetated surfaces using true-colour aerial photography

Abstract

Monitoring soil sealing in urban environments is of great interest as a key indicator of sustainable land use. Many studies have attempted to automatically classify surface impermeability by using satellite or aerial imagery. Air photo interpretation (API) has been used as a method to verify their accuracy. However, independent accuracy assessments of API have not been widely reported. The aims of this research are, firstly, to investigate independent accuracy assessments of API. Secondly, to determine whether object-based image analysis could replace manual interpretation for the detection of sealed soil and vegetated surfaces at the residential garden plot level. Four study areas, representing the industrial, commercial and residential parts of Cambridge, UK were manually digitised and classified by API. The same areas were automatically segmented and manually classified with the use of eCognition. The two methods were compared and the average overall mapping agreement was estimated to be 92%. The disagreement was qualitatively analysed and the advantages and disadvantages of each method were discussed. The very high agreement between the two methods in conjunction with the benefits of the automated method led to the conclusion that automated segmentation using eCognition could replace the manual boundary delineation when true-colour aerial photography is used. Future work will examine automated image classification methods, using eCognition, as a replacement for normal image interpretation methods.