Abstract: Land cover (LC) and land use (LU) have commonly been classified separately from remotely sensed imagery, without considering the intrinsically hierarchical and nested relationships between them. A novel joint deep learning framework is proposed and demonstrated for LC and LU classification. The proposed Joint Deep Learning (JDL) model incorporates a multilayer perceptron (MLP) and convolutional neutral network (CNN), and is implemented via a Markov process involving iterative updating. In the JDL, LU classification conducted by the CNN is made conditional upon the LC probabilities predicted by the MLP. In turn, those LU probabilities together with the original imagery are re-used as inputs to the MLP to strengthen the spatial and spectral feature representation. This process of updating the MLP and CNN forms a joint distribution, where both LC and LU are classified simultaneously through iteration.

Abstract: An object-based convolutional neural network (OCNN) method and system for urban land use classification from VFSR imagery are described. In the OCNN, segmented objects consisting of linearly shaped objects (LS-objects) and other general objects (G-objects), are utilized as functional units. The G-objects are precisely identified and labelled through a single large input window (128×128) CNN with a deep (eight-layer) network to perform a contextual object-based classification. Whereas the LS-objects are each distinguished accurately using a range of small input window (48×48) CNNs with less deep (six-layer) networks along the objects' lengths through majority voting. The locations of the input image patches for both CNN networks are determined by considering both object geometry and its spatial anisotropy, such as to accurately classify the objects into urban land use classes.

Abstract: An object-based convolutional neural network (OCNN) method and system for urban land use classification from VFSR imagery are described. In the OCNN, segmented objects consisting of linearly shaped objects (LS-objects) and other general objects (G-objects), are utilized as functional units. The G-objects are precisely identified and labelled through a single large input window (128×128) CNN with a deep (eight-layer) network to perform a contextual object-based classification. Whereas the LS-objects are each distinguished accurately using a range of small input window (48×48) CNNs with less deep (six-layer) networks along the objects' lengths through majority voting. The locations of the input image patches for both CNN networks are determined by considering both object geometry and its spatial anisotropy, such as to accurately classify the objects into urban land use classes.

Abstract: Land cover (LC) and land use (LU) have commonly been classified separately from remotely sensed imagery, without considering the intrinsically hierarchical and nested relationships between them. A novel joint deep learning framework is proposed and demonstrated for LC and LU classification. The proposed Joint Deep Learning (JDL) model incorporates a multilayer perceptron (MLP) and convolutional neutral network (CNN), and is implemented via a Markov process involving iterative updating. In the JDL, LU classification conducted by the CNN is made conditional upon the LC probabilities predicted by the MLP. In turn, those LU probabilities together with the original imagery are re-used as inputs to the MLP to strengthen the spatial and spectral feature representation. This process of updating the MLP and CNN forms a joint distribution, where both LC and LU are classified simultaneously through iteration.