Object-based Urban Environment Mapping with High Spatial Resolution Ikonos Imagery

Advances in remote sensing such as increasing spatial/spectral resolutions have strengthened its ability of urban environmental analysis. Unfortunately, high spatial resolution imagery also increases internal variability in landcover / use unit, which can cause consequent classification result showing a “salt and pepper” effect. To overcome this problem, region-based classification has been used. In such a classification, image-object (IO) is used rather than pixel as a classification unit. Using IKONOS high spatial resolution imagery, in this study, we propose to test whether the IO technique can significantly improve classification accuracy when applied to urban environmental mapping with high spatial resolution imagery compared to pixel-based method in Tampa Bay, FL, USA. We further evaluate the performance of artificial neural network (ANN) and Maximum Likelihood Classifier (MLC) in urban environmental classification with high resolution data and test the effect of number of extracted IO features on urban classification accuracy. Experimental results indicate that, in this particular study, a statistically significant difference of classification accuracy is proved between using pixel-based and IO-based data; ANN outperforms MLC when both using 9 features pixel-based data; and using more features (30 vs. 9 features) can increase IO classification accuracy, but seems not statistically significant at the 0.9 confidence level at this study.