Pipeline and library for processing hyperspectral images. The pipeline was developed as part of the MASSIMAL project; "Mapping of Algae and Seagrass using Spectral Imaging and MAchine Learning". The project acquired a large set of hyperspectral images of shallow-water areas in Norway, which were used for mapping marine habitats.
The pipeline was developed based on data from a Resonon airborne remote sensing system, with a Pika L hyperspectral camera and a Flame-S-VIS-NIR spectrometer with a cosine corrector for measuring downwelling irradiance. Parts of the pipeline are custom-built for this system. However, other parts can be used for any hyperspectral image in the ENVI file format.
If you use this software in your own work, please cite it as
- Title: "MassiPipe - a data processing pipeline for MASSIMAL hyperspectral images"
- Author: Martin Hansen Skjelvareid
- DOI: 10.5281/zenodo.14748766
MassiPipe uses a number of external libraries. Some of these are easily installed using pip, but GDAL (the Geospatial Data Abstraction Library) and rasterio (which builds on GDAL) can be more difficult to install. We recommend using conda for creating a virtual environment and installing GDAL and rasterio, and then installing the MassiPipe package with pip.
Create environment with GDAL and rasterio, installing from conda-forge channel (change "massipipe" environment name and python version to your preference):
conda create -n massipipe -c conda-forge python=3.10 gdal rasterio
Download massipipe from the massipipe GitHub repository (download as zip or use git clone). Navigate to the root folder of the repo and install using pip ("." indicates installation of package in current directory):
conda activate massipipe
pip install .
If you're a developer, you may want to install massipipe in "editable" mode instead, and also install optional dependencies that are used in development:
pip install -e .[dev]
To register the virtual environment for use with Jupyter notebooks, run
conda activate massipipe
python -m ipykernel install --user --name=massipipe
Datasets which have already been processed by MassiPipe and have been exported/published are organized as follows:
├── 1a_radiance
│ ├── rgb
| | ├── <DatasetName>_<ImageNumber>_radiance_rgb.tiff
| | └── ...
│ ├── <DatasetName>_<ImageNumber>_irradiance.spec
│ ├── <DatasetName>_<ImageNumber>_irradiance.spec.hdr
│ ├── <DatasetName>_<ImageNumber>_radiance.bip
│ ├── <DatasetName>_<ImageNumber>_radiance.bip.hdr
│ └── ...
└── imudata
│ ├── <DatasetName>_<ImageNumber>_imudata.json
│ └── ...
└── orthomosaic
│ └── <DatasetName>_<ImageType>_rgb.tiff
└── config.seabee.yaml
The config.seabee.yaml contains the parameters used to process the data before publication. To create additional image products based on the dataset (e.g. reflectance images), either modify the YAML file directly, or load the original parameters and modify them before running the pipeline.
The example below shows how a pipeline is used to process a published dataset at path
dataset_dir, using the "config" object to specify that reflectance images and glint
corrected radiance images should be created:
from massipipe import Pipeline
pipeline = Pipeline(dataset_dir)
pipeline.config.reflectance.create = True
pipeline.config.radiance_gc.create = True
pipeline.run()
After processing, the images are can be found in the folders "1b_radiance_gc" and "2a_reflectance".
The pipeline can use raw images (not yet calibrated to spectral radiance units) combined with a set of calibration files as the starting point for processing. Raw datasets should be organized as follows:
├── 0_raw
│ ├── <Raw data folder 1>
│ │ ├── <ImageSetName>_downwelling_1_pre.spec
│ │ ├── <ImageSetName>_downwelling_1_pre.spec.hdr
│ │ ├── <ImageSetName>_Pika_L_1.bil
│ │ ├── <ImageSetName>_Pika_L_1.bil.hdr
│ │ ├── <ImageSetName>_Pika_L_1.bil.times
│ │ └── <ImageSetName>_Pika_L_1.lcf
│ ├── <Raw data folder 2>
│ └── ...
└── calibration
│ ├── <downwelling_calibration_file>.dcp
│ └── <radiance_calibration_file>.icp
└── config.seabee.yaml
Note that the structure of the 0_raw folder is the structure created directly by the Pika L camera. The config.seabee.yaml file is used to modify the processing of the specific dataset. If no such config file is present, a template with defaults is created when a Pipeline object is instantiated.
Use a pipeline to process a dataset at path dataset_dir, using only default
parameters:
from massipipe import Pipeline
pipeline = Pipeline(dataset_dir)
pipeline.run()
A config file with name "seabee.config.yaml" is created in the root directory of the dataset. The pipeline runs and produces all default output image products ("quicklook" images, calibrated irradiance spectra, calibrated radiance images, RGB mosaic)
After processing, the dataset has the following structure (files not shown), with
processed files in folders 1a_radiance, geotransform, imudata, mosaics, and
quicklook. Logs from each processing are saved to the logs folder.
├── 0_raw
│ ├── <Raw data folder 1>
│ ├── <Raw data folder 2>
│ └── ...
├── 1a_radiance
│ ├── <DatasetName>_<ImageNumber>_irradiance.spec
│ ├── <DatasetName>_<ImageNumber>_irradiance.spec.hdr
│ ├── <DatasetName>_<ImageNumber>_radiance.bip
│ ├── <DatasetName>_<ImageNumber>_radiance.bip.hdr
│ └── ...
├── calibration
│ ├── downwelling_calibration_spectra
│ └── radiance_calibration_frames
├── geotransform
├── imudata
├── logs
├── mosaics
├── quicklook
└── config.seabee.yaml
Note that if some data is missing (e.g. downwelling irradiance), some of the data products will not be created.
This dataset was collected as part of the "MASSIMAL" project (Mapping of Algae and Seagrass using Spectral Imaging and Machine Learning). The project was conducted in the period 2020-2024, and data collection and field work was performed at various locations along the Norwegian coast in the period 2021-2023.
The project was financed by the Norwegian Research Council (8 MNOK) and by UiT the Arctic University of Norway (600 kNOK), and was a collaboration between
- UiT the Arctic University of Norway ("UiT")
- Norwegian Institute for Water Research ("NIVA")
- Nord University ("Nord")
The following researchers were the main contributors to the project:
- Martin Hansen Skjelvareid (principal investigator), UiT
- Katalin Blix, UiT
- Eli Rinde, NIVA
- Kasper Hancke, NIVA
- Maia Røst Kile, NIVA
- Galice Guillaume Hoarau, Nord
All UAV flights were piloted by Sigfinn Andersen at FlyLavt AS, Bodø.
Additional information about the project can be found on the following websites:
- UiT project page
- Cristin research database project page
- Norwegian Research Council project page
- SeaBee data portal with Massimal data
MassiPipe is distributed under the terms described in the LICENSE file. Please refer to the LICENSE file for complete details about usage and distribution rights.
Contributions of all kinds are welcome! If you'd like to contribute to this project, please fork the repository and submit a pull request with your proposed changes. Before starting work on a significant change, consider opening an issue to discuss your ideas.
Bug reports, feature suggestions, documentation improvements, and any other ideas for improving MassiPipe are welcome. If you have a dataset collected using a different hyperspectral camera than that used here (Resonon Pika L) and want to use MassiPipe to process the data, get in touch - we are interested in example datasets for expanding the pipeline to other hyperspcetral cameras. Thanks!