For the test image classification u

For the test image classification using the SVM, 1000 pixels from each class were randomly selected as the training samples. To match with the characteristics of the training samples of F-CNNs model, the training samples of the SVM classifier consists of normalised data and samples were collected from the Landsat images which used as input image in our proposed F-CNNs model. The mixture of spectral information from different images incorporates the spectral variation of class types in the training samples. The classification results are listed in Figure 9. The corresponding accuracy measures are listed in Figure 10 and Figure 11. By observing the classified images in Figure 9 few important points can be summarised: