Classification of land based on analysis of remotely-sensed earth images

Patent metadata

Field	Value
Publication number	US-9619734-B2
Application number	US-201514837628-A
Country	US
Kind code	B2
Filing date	Aug 27, 2015
Priority date	Sep 11, 2013
Publication date	Apr 11, 2017
Grant date	Apr 11, 2017

Abstract

Land classification based on analysis of image data. Feature extraction techniques may be used to generate a feature stack corresponding to the image data to be classified. A user may identify training data from the image data from which a classification model may be generated using one or more machine learning techniques applied to one or more features of the image. In this regard, the classification module may in turn be used to classify pixels from the image data other than the training data. Additionally, quantifiable metrics regarding the accuracy and/or precision of the models may be provided for model evaluation and/or comparison. Additionally, the generation of models may be performed in a distributed system such that model creation and/or application may be distributed in a multi-user environment for collaborative and/or iterative approaches.

First claim

What is claimed is:

1. A land classification system operable for analysis of very high resolution (VHR) remotely-sensed multispectral Earth imagery, comprising:
an image store containing image data corresponding to VHR remotely-sensed multispectral Earth images;
at least one feature extraction module in operative communication with the image store, the feature extraction module being operable to produce feature data regarding at least a portion of image data;
a feature stack comprising the image data and the feature data;
a client interface operable to receive training data regarding one or more pixels of the image data from a user regarding a land class to which the one or more pixels belong, wherein the land class is selected from a plurality of pixel-level land classes;
a classification compute module operable to generate a classification model at least in part based on a portion of the feature stack corresponding to the one or more pixels and the training data, wherein the classification model relates to classification of pixels of image data into one of the plurality of pixel-level land classes, wherein the classification compute module is operable to apply the classification model to the image data to classify each pixel of the image data into a respective one of the plurality of pixel-level land classes based on an analysis of the feature stack in relation to the classification model; and
a post classification analysis module operable to reclassify at least one pixel of the image data, that has been classified by the classification compute module into a first pixel-level land class of the plurality of pixel-level land classes, into a second pixel-level land class of the plurality of pixel-level land classes different than the first pixel-level land class based on at least one post classification rule, wherein the post classification rule is not the same as the classification model and utilizes classification information regarding at least one other pixel to reclassify the at least one pixel from the first pixel-level land class into the second pixel-level land class.

2. The system of claim 1 , wherein the post classification rule comprises at least one of a minimum mapping unit rule, a surrounded by rule, a topological relation rule, a majority analysis rule, or a smoothing rule.

3. The system of claim 1 , wherein the VHR remotely-sensed multispectral Earth image data comprises spectral band data corresponding to at least 8 multispectral bands, wherein the multispectral bands collectively range from at least about 300 nanometers in wavelength to at least about 2400 nanometers in wavelength.

4. The system of claim 3 , wherein the multispectral bands comprise a plurality of short wave infrared (SWIR) bands from at least about 1100 nanometers in wavelength to at least about 2400 nanometers in wavelength.

5. The system of claim 1 , wherein the feature stack comprises a plurality of data layers, wherein each data layer includes different feature data for each pixel of the image data, wherein the feature stack includes at least one spectral feature layer, at least one morphological feature layer, and at least one textural feature layer.

6. The system of claim 5 , wherein the at least one spectral feature layer comprises data values for each pixel of the image data based on spectral band data collected by a remote image acquisition platform.

7. The system of claim 6 , wherein at least one spectral feature layer comprises a relative measure between at least two spectral band data layers.

8. The system of claim 5 , wherein the at least one morphological feature layer comprises for each given pixel of the image data information based on the arrangement of adjacent pixels relative to the given pixel.

9. The system of claim 5 , wherein the textural feature layer comprises information for a given pixel based on a spatial distribution of tonal variations within one or more spectral band data layers relative to a given pixel.

10. The system of claim 1 , wherein the client interface comprises a plurality of distributed client interfaces operable to receive training data from a plurality of users, wherein the classification model is at least partially based on training data received from a plurality of users.

11. The system of claim 1 , further comprising a plurality of feature extraction modules each operable to generate a feature data layer of the feature stack.

12. The system of claim 1 , wherein the classification compute module is operable to utilize at least one machine learning algorithm to generate the classification model.

13. The system of claim 12 , wherein the classification compute module is operable to utilize a plurality of machine learning algorithms to generate the classification model.

14. The system of claim 12 , wherein a user is operable to specify at least one of a plurality of machine learning algorithms for use in generation of the classification model.

15. The system of claim 1 , wherein the classification model comprises at least one model parameter corresponding to at least one of a geographic region or a temporal period, wherein the classification computation module model is operable to determine the classification model for classification of image data based on the at least one model parameter and image data metadata, wherein the one or more classes correspond to at least one of a land cover, a land use, or temperature.

16. The system of claim 1 , further comprising:
a model evaluation module operable to provide quantifiable evaluation data regarding the classification module based on performance of the classification module with respect to a portion of training data received by a user, wherein the model evaluation module is operable to generate a classifier accuracy value for the classification module.

17. The system of claim 16 , wherein the model evaluation module is operable to generate at least one precision measurement for the classification module.

18. The system of claim 16 , wherein the model evaluation module is operable to compare a first model to a second model to determine if the performance of the first model is statistically significant over the performance of the second model.

19. The system of claim 1 , wherein the feature extraction module is operable to first produce feature information for pixels belonging to the training data prior to generation of the classification model by the classification compute model and is operable to second produce feature information for pixels other than the training data prior to application of the classification model to classify pixels other than the training data.

Assignees

• Digitalglobe Inc

Inventors

• Marchisio Giovanni B
• Tusk Carsten
• Koperski Krzysztof
• Tabb Mark D
• Shafer Jeffrey D

Classifications

• G06F18/2411

  based on the proximity to a decision surface, e.g. support vector machines · CPC title

• G06V20/13 Primary

  Satellite images · CPC title

• G06K9/4604

  Physics · mapped topic

• G06K9/0063

  Physics · mapped topic

• G06K9/4642

  Physics · mapped topic