paper & code

reference paper:

• [1711.06025] Learning to Compare: Relation Network for Few-Shot Learning (arxiv.org)
• Remote Sensing | Free Full-Text | Deep Relation Network for Hyperspectral Image Few-Shot Classification (mdpi.com)

reference code:

• floodsung/LearningToCompare_FSL: PyTorch code for CVPR 2018 paper: Learning to Compare: Relation Network for Few-Shot Learning (Few-Shot Learning part) (github.com)
• gokling1219/RN-FSC: Deep Relation Network for Hyperspectral Image Few-Shot Classification (github.com)
• EnayatAria/ICA-based-band-selection-HSI: Independent component analysis for dimensionality reduction of hyperspectral images (github.com)
• nshaud/DeepHyperX: Deep learning toolbox based on PyTorch for hyperspectral data classification. (github.com)

environment

laptop windows 11

env：Miniconda / Python 3.9.6 / Cuda 11.6

GPU：NVIDIA GeForce GTX 1650 4GB

CPU：lntel(R) Core(TM) i5-9300H CPU @ 2.40GHz

memory：32GB

Python 3.9.16 (main, Mar  8 2023, 10:39:24) [MSC v.1916 64 bit (AMD64)] on win32

torch                             1.12.0+cu116
torchvision                       0.13.0+cu116
scikit-learn                      1.2.2
numpy                             1.24.2
visdom                            0.2.4

h5py                              3.8.0
scipy                             1.10.1
spectral                          0.23.1
mat73                             0.60

jupyter                           1.0.0
ipykernel                         6.22.0
ipython                           8.12.0

server Ubuntu 20.04

AutoDL-品质GPU租用平台-租GPU就上AutoDL

Image： PyTorch 1.11.0 / Python 3.8(ubuntu20.04) / Cuda 11.3

GPU：RTX 3090(24GB) * 1

CPU：24 vCPU AMD EPYC 7642 48-Core Processor

memory：80GB

directory

.
|-- DeepHyperX
|-- HSI_FSC_0_basic
|-- HSI_FSC_result
|-- ICA-based-band-selection-HSI
|-- LICENSE
|-- README.md
|-- RN_FSC_modify
`-- requirements.txt

run Meta-DRN

1. visdom

local

python -m visdom.server

server

# 步骤一：在服务器上install visdom
pip install visdom

# 步骤二：服务器上启动visdom 
python -m visdom.server

# 步骤三：本地ssh连接服务器并映射端口
ssh -L <本地端口>:localhost:8097 -p <ssh访问服务器的端口> <服务器用户名>@<ssh访问服务器的ip>
# 本地端口号可以随便设置，服务器用户名和ssh访问服务器的ip都可以在AutoDL中查看到
ssh -L 8080:localhost:8097  -p 23844 root@region-11.autodl.com 

# 步骤四：全部完成，在本地网页中输入" localhost:8080 "

> [服务器visdom的本地显示_autodl vis_江南綿雨的博客-CSDN博客](https://blog.csdn.net/weixin_43702653/article/details/127273564)

2. band select

cd ICA-based-band-selection-HSI

# band select
python ICA-based_for_BS_all.py

# selected bands sorted
python bandselect_name_bands_sorted.py

3. generate dataset

3.1 generate source dataset

cd HSI_FSC

python .\generate_source_dataset.py --datasetname HS
python .\generate_source_dataset.py --datasetname BO
python .\generate_source_dataset.py --datasetname KSC
python .\generate_source_dataset.py --datasetname CH

python .\generate_meta_dataset.py

3.2 generate target dataset

python .\generate_target_dataset --dataset XX
# XX: SA/IP/UP/PC/XZ

4. meta train

python .\meta_train_EM_RN.py

5. fewshot train

python .\fewshot_train.py --datasetname XX

6. test

python .\test.py --datasetname XX

Step Merge use 'auto.sh' can merge the step 4, 5 and 6

7. generate predict image

python .\display_result_with_visdom.py

contrast experiment

 cd .\DeepHyperX\

 # windows 
 .\auto.bat
 
 # linux
 bash auto.sh

adjust the parameters

Leaning Rate

├─HSI_FSC_1_learningrate

Dropout

├─HSI_FSC_0_basic
├─HSI_FSC_2_Dropout

BatchNorm

├─HSI_FSC_0_basic
├─HSI_FSC_3_BatchNorm

C way

├─HSI_FSC_4_5way
├─HSI_FSC_4_10way
├─HSI_FSC_4_15way
├─HSI_FSC_0_basic
├─HSI_FSC_4_25way
├─HSI_FSC_4_30way

K shot N query

├─HSI_FSC_0_basic
├─HSI_FSC_5_support5_test15
├─HSI_FSC_5_support10_test10
├─HSI_FSC_5_support15_test5

git push error!

# 设置代理
(base) PS D:\Document\DevelopProject\Develop_DeepLearning\HSI\HSI-FSC> git config --global http.proxy http://127.0.0.1:7890
(base) PS D:\Document\DevelopProject\Develop_DeepLearning\HSI\HSI-FSC> git config --global https.proxy http://127.0.0.1:7890
# 查看代理
(base) PS D:\Document\DevelopProject\Develop_DeepLearning\HSI\HSI-FSC> git config --global https.proxy
http://127.0.0.1:7890
(base) PS D:\Document\DevelopProject\Develop_DeepLearning\HSI\HSI-FSC> git config --global http.proxy
http://127.0.0.1:7890
# 取消代理
git config --global --unset http.proxy 
git config --global --unset https.proxy