Influenced by the different spectrum with same object and the same spectrum with different object

when classifying remote sensing images

there will be a limitation on the improvement of the classification accuracy if merely relies on spectral signatures.However

this influence could be impaired by the characteristics of local spatial statistics

which are capable of describing and analyzing the spatial autocorrelation.In this paper

the effects which different Local Spatial Statistics indices （Moran’s I

Getis-Ord G i

Geary’s C）

neighborhood rule

and lags have on the classification accuracy of the high resolution remote sensing image:SPOT 5 are discussed.Firstly

band 1 is analyzed regarding to local spatial statistics

and the results are added to original spectral bands as textural ones;Secondly

supervised classification is carried out with both spectral band and textural band;Finally

test samples are selected to assess the classification accuracy and different accuracies on different conditions are juxtaposed with each other to figure out the relations and disciplines between the classification accuracy and the parameters.Results indicate that G i statistics should be the optimum choice to raise the overall classification accuracy

with an increase from 87.74% to 95.12%.