Pixel-based Classification Workflow with

Prerequisites

• Please refer to the prerequisites section in our guide for more information. This sample demonstrates how to do export training data and model inference using ArcGIS Image Server. Alternatively, they can be done using ArcGIS Pro as well.
• If you have already exported training samples using ArcGIS Pro, you can jump straight to the training section. The saved model can also be imported into ArcGIS Pro directly.

Introduction

As an example, we use land cover classification to demonstrate of workflow of pixel-based image classification using arcgis.learn. To begin with, we need input imagery as well as labels for each pixel. With the ArcGIS platform, these datasets are represented as layers, and are available in our GIS. This guide notebook showcases an end-to-end to land cover classification workflow using ArcGIS API for Python. The workflow consists of three major steps: (1) extract training data, (2) train a deep learning image segmentation model, (3) deploy the model for inference and create maps. To better illustrate this process, we will use World Imagery and high-resolution labeled data provided by the Chesapeake Conservancy land cover project.

Figure 1. A subset of of the labeled data for Kent county, Delaware

Export training data for deep learning

Import ArcGIS API for Python and get connected to your GIS

from arcgis import GIS

gis = GIS("home")

Preprocess Training data

To export training data, we need a labeled imagery layer that contains the class label for each location, and a raster input that contains all the original pixels and band information. In this land cover classification case, we will be using a subset of the one-meter resolution Kent county, Delaware, dataset as the labeled imagery layer and World Imagery: Color Infrared as the raster input.

Notes:

• The raster and labeled imagery layer both are required to be '8 bit Unsigned' when the raster is a RGB layer such as world imagery.
• The Labeled imagery layer should be a thematic raster with pixel values corresponding to the label class value.
• The pixel values should range from 1 to n, where n is the total number of classes.
• Any NoData value should be mapped to 0, as portions of image with nodata values would be ignored in while exporting training data.

label_layer = gis.content.search("Kent_county_full_label_land_cover")[0] # the index might change
label_layer

Kent_county_full_label_land_cover
land cover labels for Unet sampleImagery Layer by portaladmin
Last Modified: May 09, 2019
0 comments, 35 views

m = gis.map("Kent county, Delaware")
m

Now let's retrieve the World Imagery layer.

world_imagery_item = gis.content.search("WorldImagery_AOI_NewYork")[0] # the index might change
world_imagery_item

WorldImagery_AOI_NewYork
To be used internally only. Not allowed for distribution outsideImagery Layer by portaladmin
Last Modified: June 12, 2019
0 comments, 5 views

world_imagery_layer = world_imagery_item.layers[0]
m.add_layer(world_imagery_layer)

m.add_layer(label_layer)

Specify a folder name in raster store that will be used to store our training data

Make sure a raster store is ready on your raster analytics image server. This is where the output sub images, also called chips, labels and metadata files are going to be stored.

from arcgis.raster import analytics

ds = analytics.get_datastores(gis=gis)
ds

<DatastoreManager for https://datascienceadv.esri.com/server/admin>

ds.search()

[<Datastore title:"/nosqlDatabases/AGSDataStore_bigdata_bds_4c9tuc3o" type:"nosql">,
 <Datastore title:"/nosqlDatabases/AGSDataStore_nosqldb_tcs_l6mh5mhm" type:"nosql">,
 <Datastore title:"/enterpriseDatabases/AGSDataStore_ds_b6108wk9" type:"egdb">,
 <Datastore title:"/rasterStores/LocalRasterStore" type:"rasterStore">]

rasterstore = ds.get("/rasterStores/LocalRasterStore")
rasterstore

<Datastore title:"/rasterStores/LocalRasterStore" type:"rasterStore">

samplefolder = "landcover_sample_world_imagery"
samplefolder

'landcover_sample_world_imagery'

Export training data using arcgis.learn

With the feature class and raster layer, we are now ready to export training data using the export_training_data() method in arcgis.learn module. In addition to feature class, raster layer, and output folder, we also need to specify a few other parameters such as tile_size (size of the image chips), strid_size (distance to move each time when creating the next image chip), chip_format (TIFF, PNG, or JPEG), metadata format (how we are going to store those training labels). More detail can be found here.

Depending on the size of your data, tile and stride size, and computing resources, this opertation can take 15mins~2hrs in our experiment. Also, do not re-run it if you already run it once unless you would like to update the setting.

import arcgis
from arcgis import learn
arcgis.env.verbose = True

export = learn.export_training_data(input_raster = world_imagery_layer,
                                    output_location = samplefolder,
                                    input_class_data = label_layer.url, 
                                    chip_format = "PNG", 
                                    tile_size = {"x":400,"y":400}, 
                                    stride_size = {"x":0,"y":0}, 
                                    metadata_format = "Classified_Tiles",                                        
                                    context = {"startIndex": 0, "exportAllTiles": False, "cellSize": 2},
                                    context = context,
                                    gis = gis)

Submitted.
Executing...
Start Time: Wednesday, May 22, 2019 9:34:15 AM
Running script ExportTrainingDataforDeepLearning...
No cloud raster store.
Exporting...
Succeeded at Wednesday, May 22, 2019 9:56:49 AM (Elapsed Time: 22 minutes 33 seconds)

Now let's get into the raster store and look at what has been generated and exported.

from arcgis.raster.analytics import list_datastore_content

samples = list_datastore_content(rasterstore.datapath + '/' + samplefolder + "/images", filter = "*png")
# print out the first five chips/subimages
samples[0:5]

Submitted.
Executing...
Start Time: Wednesday, May 22, 2019 1:04:14 PM
Running script ListDatastoreContent...

['/rasterStores/LocalRS/landcover_sample/images/000000000.png',
 '/rasterStores/LocalRS/landcover_sample/images/000000001.png',
 '/rasterStores/LocalRS/landcover_sample/images/000000002.png',
 '/rasterStores/LocalRS/landcover_sample/images/000000003.png',
 '/rasterStores/LocalRS/landcover_sample/images/000000004.png']

labels = list_datastore_content(rasterstore.datapath + '/' + samplefolder + "/labels", filter = "*png")
# print out the labels images for the first five chips
labels[0:5]

Start Time: Wednesday, May 22, 2019 1:04:19 PM
Running script ListDatastoreContent...
Completed script ListDatastoreContent...
Succeeded at Wednesday, May 22, 2019 1:04:19 PM (Elapsed Time: 0.05 seconds)

['/rasterStores/LocalRS/landcover_sample/labels/000000000.png',
 '/rasterStores/LocalRS/landcover_sample/labels/000000001.png',
 '/rasterStores/LocalRS/landcover_sample/labels/000000002.png',
 '/rasterStores/LocalRS/landcover_sample/labels/000000003.png',
 '/rasterStores/LocalRS/landcover_sample/labels/000000004.png']

Model training

If you've already done part 1, you should already have the training chips. Please change the path to your own export training data folder that contains "images" and "labels" folder.

from arcgis.learn import UnetClassifier, prepare_data

data_path = r'to_your_data_folder'

data = prepare_data(data_path, batch_size=16)

Visualize training data

To get a sense of what the training data looks like, arcgis.learn.show_batch() method randomly picks a few training chips and visualize them.

data.show_batch()

Load model architecture

We will be using U-net, one of the well-recognized image segmentation algorithm, for our land cover classification. U-Net is designed like an auto-encoder. It has an encoding path (“contracting”) paired with a decoding path (“expanding”) which gives it the “U” shape. However, in contrast to the autoencoder, U-Net predicts a pixelwise segmentation map of the input image rather than classifying the input image as a whole. For each pixel in the original image, it asks the question: “To which class does this pixel belong?”. U-Net passes the feature maps from each level of the contracting path over to the analogous level in the expanding path. These are similar to residual connections in a ResNet type model, and allow the classifier to consider features at various scales and complexities to make its decision.

Figure 2. Architecture of a Unet model

model = UnetClassifier(data)

Train a model through learning rate tuning and transfer learning

Learning rate is one of the most important hyperparameters in model training. Here we explore a range of learning rate to guide us to choose the best one.

model.lr_find()

Based on the learning rate plot above, we can see that the loss going down dramatically at 1e-4. Therefore, we set learning rate to be a range from 3e-5 to 1e-4, which means we will apply smaller rates to the first few layers and larger rates for the last few layers, and intermediate rates for middle layers, which is the idea of transfer learning. Let's start with 10 epochs for the sake of time.

model.fit(10, lr=slice(3e-5, 1e-4))

Visualize classification results in validation set

Now we have the model, let's look at how the model performs.

model.show_results()

As we can see, with only 10 epochs, we are already seeing reasonable results. Further improvement can be achieved through more sophisticated hyperparameter tuning. Let's save the model for further training or inference later. The model should be saved into a models folder in your folder. By default, it will be saved into your data_path that you specified in the very beginning of this notebook.

model.save('stage-1-10')

Deployment and inference

Locate model package

If you have finished Part 2 of this notebook, you should have a model folder saved already. The model package is the "stage-1-10.zip" zip file in the folder. The model package file includes a few files:

1. A model definition file with the extension .emd which includes information like model framework (e.g. tensorflow, pytorch). ArcGIS needs it to interpret your model.
2. A model file in binary format that we have trained in Part 2.
3. If your framework is not supported yet, a custom python raster function has to be added as well.

model_package = "Path_To_Your_Model_Package"

classify_land_model_package = gis.content.add(item_properties={"type":"Deep Learning Package",
                                                               "typeKeywords":"Deep Learning, Raster",
                                                               "title":"Land_Cover_Kent_DL_Model_Updated",
                                                               "tags":"deeplearning", 'overwrite':'True'}, 
                                              data=model_package)

classify_land_model_package

Land_Cover_Kent_DL_Model_Updated
Deep Learning Package by portaladmin
Last Modified: September 24, 2019
0 comments, 0 views

Now we are ready to install the mode. Installation of the deep learning model item will unpack the model definition file, model file and the inference function script, and copy them to "trusted" location under the Raster Analytic Image Server site's system directory.

from arcgis.learn import Model, list_models

land_cover_model = Model(classify_land_model_package)

land_cover_model.install()

'[resources]models\\raster\\f4e6090320014fefa233b7c47d44c8ea\\land_cover_full_60.emd'

Model inference

To test our model, let's get a new raster image by specifying a spatial extent.

from arcgis.learn import classify_pixels

ext = {'spatialReference': {'latestWkid': 3857, 'wkid': 102100},
       'xmin': -8411742.088727,
       'ymin': 4703820.069424,
       'xmax': -8395632.940157,
       'ymax': 4720903.740949}

context = {'cellSize': 2,
           'processorType':'GPU',
           'extent': ext,
           'batch_size': 9}

out_classify = classify_pixels(input_raster = world_imagery_layer, # make sure pass in the layer not url
                               model = land_cover_model,
                               output_name = "land_cover_sample_inference_result_24092019_1",
                               context = context,
                               gis = gis)

out_classify

land_cover_sample_inference_result_24092019_1
Analysis Image Service generated from ClassifyPixelsUsingDeepLearningImagery Layer by portaladmin
Last Modified: September 24, 2019
0 comments, 0 views

Visualize land cover classification on map

from arcgis.raster.functions import colormap

result_map = gis.map('Kent County, Delaware')
result_map.basemap = 'satellite'

# applying color map [value, red, green, blue]
land_cover_colormap=[[0, 0, 0, 0],
                     [1, 0, 197, 255],
                     [2, 0, 168, 132],
                     [3, 38, 115, 0],
                     [4, 76, 230, 0],
                     [5, 163, 255, 115],
                     [6, 255, 170, 0],
                     [7, 255, 0, 0],
                     [8, 156, 156, 156],
                     [9, 0, 0, 0],
                     [10, 115, 115, 0],
                     [11, 230, 230, 0],
                     [12, 255, 255, 115],
                     [13, 0, 0, 0],
                     [14, 0, 0, 0],
                     [15, 0, 0, 0]]

result_map.add_layer(colormap(out_classify.layers[0], 
                              colormap = land_cover_colormap, 
                              astype='u8'),
                    {'opacity':0.2})
result_map