Joint deep learning for land cover and land use classification
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.
1. A computer implemented method of jointly determining land cover and land use classifications of land from remotely sensed imagery of said land, the method comprising:
for an input image illustrating a patch of land to be classified:
i) segmenting objects within the image;
ii) determining for one or more pixels in the input image a first conditional probability of a first land cover classification from a plurality of predefined land cover classifications using a machine learning network of a first type;
iii) determining for segmented objects in the input image a second conditional probability of a first land use classification from a plurality of predefined land use classifications using a machine learning network of a second type; and
iv) iterating steps ii) and iii) above, using the second conditional probability as an input to the first determining step ii);
wherein the iteration process produces land cover classification data for the one or more pixels in the input image and land use classification data for the segmented objects in the input image.
2. A method according to claim 1 , wherein the machine learning network of the first type is a multilayer perceptron.
3. A method according to claim 1 , wherein the machine learning network of the second type is an object based convolutional neural network.
4. A method according to claim 1 , and further comprising generating an output image corresponding to the input image, the output image comprising the input image illustrating the patch of land visually augmented to indicate the land use classification determined for the segmented objects in the input image and/or the land cover classification determined for the one or more pixels in the input image.
5. A method according to claim 4 , wherein the visual augmentation comprises overlaying a color wash on to the segmented objects in the input image, the overlaid color being selected in accordance with a predetermined color mapping of color to land use classification.
6. A method according to claim 1 , wherein the iteration is repeated for at least 5 times, and more preferably for at least 8 times, and most preferably 10 times.
7. A method according to claim 1 , wherein the iteration is repeated no more than 10 times.
8. A computer system for jointly determining land cover and land use classifications of land from remotely sensed imagery of said land, the system comprising:
one or more processors;
at least computer readable storage medium storing one or more computer programs so arranged such that when executed by the processors they cause the computer system to:
for an input image illustrating a patch of land to be classified:
i) segment objects within the input image;
ii) determine for one or more pixels in the input image a first conditional probability of a first land cover classification from a plurality of predefined land cover classifications using a machine learning network of a first type;
iii) determine for segmented objects in the input image a second conditional probability of a first land use classification from a plurality of predefined land use classifications using a machine learning network of a second type; and
iv) iterate steps ii) and iii) above, using the second conditional probability as an input to the first determining step ii);
wherein the iteration process produces land cover classification data for the one or more pixels in the input image and land use classification data for the segmented objects in the input image.
9. A system according to claim 8 , wherein the machine learning network of the first type is a multilayer perceptron.
10. A system according to claim 8 , wherein the machine learning network of the second type is an object based convolutional neural network.
11. A system according to claim 8 , and further comprising generating an output image corresponding to the input image, the output image comprising the input image of the patch of land visually augmented to indicate the land use classification determined for the segmented objects in the input image and/or the land cover classification determined for the one or more pixels in the input image.
12. A system according to claim 11 , wherein the visual augmentation comprises overlaying a color wash on to the segmented objects in the input image, the overlaid color being selected in accordance with a predetermined color mapping of color to land use classification.