Urban Feature Extraction within a Complex Urban Area with an Improved 3D-CNN Using Airborne Hyperspectral Data

Airborne hyperspectral data has high spectral-spatial information. However, how to mine and use this information effectively is still a great challenge. Recently, three-dimensional convolutional neural network (3D-CNN) provides new effective way of classification. its capability mining in complex urban areas, especially cloud shadow areas not been validated. Therefore, 3D-1D-CNN model was proposed for feature extraction with images affected by shadows. Firstly, spectral composition parameters, vegetation index, texture characteristics were extracted from data. Secondly, the parameters fused segmented into many S × B patches which would be input 3D-CNN classifier areas. Thirdly, Support Vector Machine (SVM), Random Forest (RF),1D-CNN, 3D-CNN, 3D-2D-CNN classifiers also carried out comparison. Finally, confusion matrix Kappa coefficient calculated accuracy assessment. The overall 96.32%, 23.96%, 11.02%, 5.22%, 0.42%, much higher than that SVM, RF, 1D-CNN, or respectively. results indicated could spatial-spectral effectively, grass highway missing In future, used green spaces.