PCAHyperspectralClassifier

Classification of Hyperspectral Images ( HSIs ) with Principal Component Analysis ( PCA ) preprocessing exploiting CUDA ( cuBLAS ). The code, explicitly designed for the NVIDIA Jetson Nano Developer kit, can run on any PC with NVIDIA GPU, Python3, and the necessary modules.

References

This work uses the DeepHyperX toolbox based on the following paper in Geoscience and Remote Sensing Magazine :

N. Audebert, B. Le Saux and S. Lefevre, "Deep Learning for Classification of Hyperspectral Data: A Comparative Review," in IEEE Geoscience and Remote Sensing Magazine, vol. 7, no. 2, pp. 159-173, June 2019.

For the PCA, this work uses the algorithm of this paper :

M. Andrecut, "Parallel GPU Implementation of Iterative PCA Algorithms*," 2009, https://www.researchgate.net/publication/26829736_Parallel_GPU_Implementation_of_Iterative_PCA_Algorithms.

as implemented on https://github.com/nmerrill67/GPU_GSPCA

As 3D CNN, this work uses the model described in this paper :

• 3D CNN (Spectral–Spatial Classification of Hyperspectral Imagery with 3D Convolutional Neural Network, Li et al., Remote Sensing 2017)

Bibtex format :

@article{8738045, author={N. {Audebert} and B. {Le Saux} and S. {Lefèvre}}, journal={IEEE Geoscience and Remote Sensing Magazine}, title={Deep Learning for Classification of Hyperspectral Data: A Comparative Review}, year={2019}, volume={7}, number={2}, pages={159-173}, doi={10.1109/MGRS.2019.2912563}, ISSN={2373-7468}, month={June},}

@misc{pca, author={M. Andrecut}, title= {Parallel GPU Implementation ofIt is based on the PyTorch deep learning and GPU computing framework Iterative PCA Algorithms}, journal={Research Gate}, year={2009}, note ={\url{https://www.researchgate.net/publication/26829736_Parallel_GPU_Implementation_of_Iterative_PCA_Algorithms}}}

@Article{rs9010067, AUTHOR = {Li, Ying and Zhang, Haokui and Shen, Qiang}, TITLE = {Spectral–Spatial Classification of Hyperspectral Imagery with 3D Convolutional Neural Network}, JOURNAL = {Remote Sensing}, VOLUME = {9}, YEAR = {2017}, NUMBER = {1}, ARTICLE-NUMBER = {67}, URL = {https://www.mdpi.com/2072-4292/9/1/67}, ISSN = {2072-4292}, DOI = {10.3390/rs9010067} }

DATASET

Several public hyperspectral datasets are available at UPV/EHU wiki. For the testing of this code, we used the following public datasets:

PaviaU
    ├── PaviaU_gt.mat
    └── PaviaU.mat
    
IndianPines
    ├── IndianPines_gt.mat
    └── IndianPines_corrected.mat
    
Salinas
    ├── Salinas_gt.mat
    └── Salinas_corrected.mat

SETUP FOR UBUNTU LINUX OS

1) Update :

sudo apt update

2) Install pip3 for Python3:

sudo apt install python3-pip  && python3 -m pip install --upgrade pip

3) Install Python3 modules :

python3 -m pip install -r requirements.txt

4) Check Cuda version or Install CUDA

nvcc -V

5) Install CuPy module ( e.g., for CUDA 12.1 )

python3 -m pip install cupy-cuda12x

6) Install gsl library:

sudo apt-get install libgsl-dev

7) Compile PCA with CUDA compiler

nvcc -Xcompiler -fPIC -shared -o pca.so main.cpp kernel_pca.cu -lcublas -lm -lgsl -lgslcblas

USAGE EXAMPLE

1) Set VISDOM environment in a separate shell to view images and plots in a browser

python3 -m visdom.server

2) Train the Model

python3 main.py --model li --dataset IndianPines --training_sample 0.7  --epoch 200 --cuda 0 --pca 10

3) Inference

python3 inference.py --cuda 0 --image IndianPines --checkpoint models/ip/10_IP.pth --model li --pca 10

To sample milliWatts absorbed every second on Jetson Nano, run on a separate terminal

sudo watch -t -n 1 "(cat /sys/bus/i2c/drivers/ina3221x/6-0040/iio:device0/in_power1_input) | tee -a consumpPCA.txt"