Add sentinel hub example

examples/README.md

@@ -11,3 +11,7 @@
 
 - [Prebuilt ResNet50 Usage](nets/resnet.py)
 - [LeNet in MXNet](nets/SageMaker_mx-lenet.ipynb)
+
+## Other Examples
+
+- [Preparing Data with Sentinel Hub](sentinel-hub.md)

examples/sentinel-hub.md

@@ -0,0 +1,33 @@
+# Preparing Data with Sentinel Hub
+
+[Sentinel Hub](https://www.sentinel-hub.com/) is an easy way to access Sentinel imagery for use in machine learning applications. As an example, here is a configuration file for creating building segmentation masks in Valencia, Spain from Sentinel-2 imagery:
+
+```json
+{
+  "country": "spain",
+  "bounding_box": [
+    -0.745697021484375,
+    39.28010491220614,
+    -0.3076171875,
+    39.625788248139436
+  ],
+  "zoom": 14,
+  "classes": [
+    { "name": "Building", "filter": ["has", "building"] }
+  ],
+  "imagery": "https://services.sentinel-hub.com/ogc/wms/[WMS_ID]?service=WMS&request=GetMap&layers=1_TRUE_COLOR&styles=&format=image%2Fpng&transparent=true&version=1.1.1&showlogo=false&name=Sentinel-2%20L1C&width=256&height=256&pane=activeLayer&maxcc=100&time=2018-07-15%2F2018-07-15&srs=EPSG%3A3857&bbox={bbox}",
+  "background_ratio": 1,
+  "ml_type": "segmentation"
+}
+
+```
+
+We've chosen zoom 14 because it roughly corresponds to the maximum resolution of Sentinel imagery (~9.547m vs 10m). Also make sure to replace `[WMS_ID]` with your Sentinel Hub WMS in the `imagery` link above.
+
+Here are some example labels created from this configuration:
+
+![Labeled imagery in Valencia, Spain](images/valencia-example-1.png)
+![Labeled imagery in Valencia, Spain](images/valencia-example-2.png)
+![Labeled imagery in Valencia, Spain](images/valencia-example-3.png)
+
+While the resolution might not support accurate single building footprint mapping, it can be used to create a classifier for built-up or urban areas.