Automatic classification of BW aerial orthophotos

摘要

The paper presents automatic classification of black and white aerial orthophotos. The classification is based on object-oriented image analysis in the eCognition software. The methodology where channels calculated from texture measures were used to enlarge signature space for individual classes and smoothing filters were applied to create suitable segment shapes for future classifications, is described. The methodology was verified on detailed forest ground truth data. The results have proved the accuracy of about 80 to more than 90 per cent for individual classes. The classes comprised forest classes and non-forest classes. The paper presents results from two different detailed classifications.