Using AWS SageMaker/Labelmaker to process Satellite Imagery

This solution shows how to process map imagery using AWS SageMaker and Labelmaker to build an AI Model to predict buildings. This readme updates an article "Use Label Maker and Amazon SageMaker to automatically map buildings in Vietnam" by ZHUANGFANG NANA YI referenced below and provides a more basic step by step process.

First we'll build an EC2 Instance for downloading and preprocessing map images using labelmaker. We'll then transfer the map data to S3. Once on S3 we'll start a Jupyter Notebook using AWS SageMaker to build and deploy a model. We can then use the Jupyter Notebook to predict buildings in the satellite imagery.

Configure AWS EC2 Instance

Use the AWS Console to configure the EC2 Instance for processing map data. This is a step by step process.

AWS EC2 Dashboard

Select Instances

Instances

Click on "Launch Instance"

Choose AMI

Ubuntu Server 16.04 LTS (ami-43a15f3e)

Click on "Select"

Choose Instance Type

m5.4xlarge

Click on "Next"

Configure Instance
Click on "Advanced Details

User data
Select "As file"
Click on "Choose File" and Select "cloud-init" from project cloud-deployment directory 

Next

Add Storage
Next

Add Tags
Next

Configure Security Group
Select "Select an existing security group"
Select "default"

Review
Click on "Launch"

Obtain MapBox Access Token

You need to create an account on MapBox and obtain your access token from the Account section.

Preprocess Satellite Imagery on AWS EC2 Instance

You will need to ssh into the AWS EC2 Instance you created above. This is a step by step process.

Check to insure cloud-init has completed

See contents of "/tmp/install-label-maker" it should say "installation complete".

Download the labelmaker json configuration file

Download config.json and replace <ACCESS TOKEN> with API Key you created in your Mapbox Account.

wget https://raw.githubusercontent.com/kskalvar/aws-sagemaker-labelmaker-satellite-imagery/master/labelmaker-config/config.json  

Preprocess Satellite Imagery using Labelmaker

Run the following label-maker commands:

label-maker download
label-maker labels
label-maker preview
label-maker images
label-maker package

Configure AWS CLI and copy results to S3

aws configure

AWS Access Key ID []:
AWS Secret Access Key []:

Note: Replaced with your AWS Account Number.  Example: data-<Your AWS Account Number>

aws s3api create-bucket --bucket data-<Your AWS Account Number> --region us-east-1  
aws s3 cp data s3://data-<Your AWS Account Number> --recursive

Configure AWS SageMaker

Use the AWS Console to configure a SageMaker Instance for processing map data. This is a step by step process.

AWS SageMaker Dashboard

Click on "Create notebook instance"

Notebook instance name: labelmaker
Notebook instance type: ml.t2.medium
IAM Role: Create a new role

S3 buckets you specify:
Select Specific S3 buckets
Enter: data-<Your AWS Account Number>
Click on "Create role"

Click on "Create notebook instance"

Display Notebook instances using the SageMaker Dashboard

Notebook/Notebook instances
Name: labelmaker
Action: Open # it will show pending until it's ready to open

This will open the Jupyter Notebook in a new tab on your browser.

Upload Jupyter Notebook

Click on "Upload" and Select "SageMaker_mx-lenet.ipynb" from project jupyter-notebook directory

Once the notebook is uploaded, click on the notebook to open it.
Run each cell Step by Step

Note: data-754487812300 should be replaced with your AWS Account Number in some of the cells. Example: data-<Your AWS Account Number> to match your S3 bucket above.

References

Use Label Maker and Amazon SageMaker to automatically map buildings in Vietnam https://developmentseed.org/blog/2018/01/19/sagemaker-label-maker-case/

Mapbox API Key
http://www.mapbox.com