Geoalert / aeronetlib

A Remote Sensing data handling library for Deep Learning

Aeronet

Python library to work with geospatial data

List of content

• Aim and scope
• Modules
• Quickstart example
• Requirements and installation
• Documentation and wiki
• Citing
• License

Aim and scope

As a part of Aeronetlib, which is designed to make it easier for the deep learning researchers to handle the remote sensing data, Aeronet_raster provides an interface to handle geotiff raster images.

Modules and classes

• .raster

  • Band | BandCollection
  • BandSample | BandSampleCollection

• .collectionprocessor

  • CollectionProcessor
  • SampleWindowWriter
  • SampleCollectionWindowWriter

• .visualization

  • add_mask

Quickstart example

# Split multiband image to single-bands
IMAGE_FILE = '/path/to/image.tif'
OUT_PATH = '/path/to/dataset/'
channels = ['RED', 'GRN', 'BLU']

bc = split(IMAGE_FILE, OUT_PATH, channels)
print(f"num_bands is {bc.count}, shape is {bc.shape}")

# bc.show() returns RGB image as numpy array
# undersampling specifies resize coefficient
plt.imshow(bc.show(undersampling=32))

#############################################################################

# Load several singleband files
path_to_folder = 'data/handlabeled'
channels = ['RED', 'GRN', 'BLU']
labels = ['101', '901', '902']  # mask channels, each one in separate file
extension = '.tif'

paths = [os.path.join(path_to_folder, ch + extension) for ch in channels+labels]
bc =BandCollection(path)

# Inspect area at pixel coordinates (xmin = 4000, xmax=5000, ymin=4000, ymax=5000)
# labels = numbers of bands with masks to show over RGB
plt.imshow(bc.show((4000, 4000, 5000, 5000), labels = (3,4,5)))

# Get specified area as numpy array
# ch_axis - channel axis in resulting array
sample = bc.numpy((4000, 4000, 5000, 5000), ch_axis=0)
print(sample.shape) # returns (6, 1000, 1000)

# Get specified area as BandSampleCollection
sample = bc.sample(y, x, height, width)

# Process BandCollection

from aeronetlib_raster.aeronet_raster import CollectionProcessor

OUT_CHANNELS = ['RED', 'GRN', 'BLU']

def processing_fn(sample):
    # Processing function, makes image brighter
    # Here can be segmentation model prediction, etc.
    return np.clip(sample+50, 0, 255).astype(np.uint8)

# Wrap the function into Predictor
# `bound` means the width of samples overlap
predictor = CollectionProcessor(input_channels = (0, 1, 2),
                                output_labels=OUT_CHANNELS,
                                processing_fn=processing_fn,
                                sample_size=(2048,2048),
                                bound=512)

# Open the imagery and process it
result_bc = predictor.process(bc, 'result')
plt.imshow(result_bc.show(undersampling=32))

Requirements and installation

1. python 3
2. rasterio >= 1.0.0
3. shapely >= 1.7.1
4. rtree>=0.8.3
5. opencv-python>=4.0.0
6. tqdm >=4.36.1

Pypi package: .. code:: bash

$ pip install aeronet[all]

for partial install:

Raster-only .. code:: bash

$ pip install aeronet[raster]

Vector-only .. code:: bash

$ pip install aeronet[vector]

Source code: .. code:: bash

$ pip install git+https://github.com/aeronetlab/aeronetlib

Contributing We accept pull-requests and bug reports at github page

You can use `make build to build the libraries and make upload` to update them at pypi (authorization required).

Testing 1. Create and activate virtual environment 2. `make prepare to install all requirements in the venv 3. make test` to run all tests

Documentation and wiki

The project wiki contains some insights about the background of the remote sensing data storage and processing and useful links to the external resources. Latest documentation is available at Read the docs

Citing

@misc{Yakubovskiy:2019,
  Author = {Pavel Yakubovskiy, Alexey Trekin},
  Title = {Aeronetlib},
  Year = {2019},
  Publisher = {GitHub},
  Journal = {GitHub repository},
  Howpublished = {\url{https://github.com/aeronetlab/aeronetlib}}
}

License

Project is distributed under MIT License.