Hyperspectral imaging provides a useful tool for extracting complex information when visual spectral bands are not enough to solve certain tasks. However, processing hyperspectral images is usually computationally expensive due to the great amount of both spatial and spectral data they incorporate. Here, we present a low-cost convolutional neural network called Hyper3DNet designed for hyperspectral image classification. Its architecture consists of two parts: a series of densely connected 3-D convolutions used as a feature extractor, and a series of 2-D separable convolutions used as a spatial encoder.
We used three well-known remote sensing HSI datasets: Indian Pines (IP), Pavia University (PU), and Salinas (SA).
We also experimented with the EuroSAT dataset (EU) in spite of the fact that it is not a hyperspectral but a multispectral dataset, so that we validate the usefulness of our network for images with just a few spectral channels. The original EUROSAT dataset can be downloaded from here. Alternatively, a pre-processed ready-to-use dataset that combines all the images in a single ".h5" file can be downloaded from here.
Furthermore, we use an in-greenhouse controlled HSI dataset of Kochia leaves in order to classify three different herbicide-resistance levels (herbicide-susceptible, dicamba-resistant, and glyphosate-resistant). A total of 76 images of kochia with varying spatial resolution and 300 spectral bands ranging from 387.12 to 1023.5 nm were captured. From these images, which were previously calibrated and converted to reflectance values, we manually extracted 6,316 25x25 pixel overlapping patches. The Kochia dataset can be downloaded from here.
Python requirements can be installed from requirements.txt:
pip install -r requirements.txt
This repository contains the following scripts:
TrainCrossval.py: Load data and trained the selected network using 10-fold cross-validation. The weights are saved in the \weigths folder.CalculateMetrics-Plot.py: Load data, calculate the performance metrics of the selected network, and plot segmentation results. The weights are laoded from the \weigths folder.TestDemo.ipynb: Jupyter file that evaluates the performance metrics of a selected network.network.py: Contains all the network architectures used in this work.utils.py: Additional methods used to transform the data and calculate metrics.
Use this Bibtex to cite this repository
@article{10.1117/1.JRS.14.036519,
author = {Giorgio Morales and John W. Sheppard and Bryan Scherrer and Joseph A. Shaw},
title = {{Reduced-cost hyperspectral convolutional neural networks}},
volume = {14},
journal = {Journal of Applied Remote Sensing},
number = {3},
publisher = {SPIE},
pages = {1 -- 18},
keywords = {convolution neural networks, hyperspectral classification, deep learning, parameter reduction},
year = {2020},
doi = {10.1117/1.JRS.14.036519},
URL = {https://doi.org/10.1117/1.JRS.14.036519}
}