Object-Oriented Semisupervised Classification of VHR Images by Combining MedLDA and a Bilateral Filter

Table 1

Comparison of the each class accuracy of four classifying results.

Class	SVM	SVM + MV	ssLDA	ssMedLDA
Asphalt	92.504901%	95.370231%	97.602172%	98.386367%
Meadows	98.219744%	99.892756%	98.847123%	99.941016%
Gravel	66.22201%	72.748928%	64.173416%	72.987137%
Trees	90.013055%	81.070496%	85.411227%	96.083551%
Metal sheets	99.107807%	97.32342%	98.66171%	100%
Bare soil	62.278783%	64.028634%	78.146749%	94.690793%
Bitumen	83.308271%	96.842105%	99.097744%	99.172932%
Bricks	89.788159%	97.311244%	98.777838%	98.859316%
Shadows	100%	98.944034%	96.937698%	99.788807%
OA	88.6972%	90.9729%	92.7887	97.0723%
Kappa	0.8625	0.8896	0.9130	0.9649