This repository contains our research on the innovative application of Attentive Cross-Domain Few-Shot Learning (ACDFSL) in Hyperspectral Image (HSI) Classification. This study specifically tackles the challenging aspect of limited labeled data in various environments.
Our approach, which is unique in its integration of attention mechanisms into few-shot learning models, introduces a deep learning architecture of four convolution blocks incorporating Squeeze-and-Excitation (SE) attention and Residual elements. This strategy marks a significant shift from conventional methodologies.
CUDA: Version 10.0
Python: Version 3.7
PyTorch: Version 1.5
scikit-learn (sklearn): Version 0.23.2
numpy: Version 1.19.2
Please make sure to have the specific versions installed to ensure the code runs correctly.
Target: https://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes
Source: https://opendatalab.com/Chikusei_Dataset
After satisfying the required dependencies, you can execute the script for training and prediction on the Salinas dataset using the following steps:
Clone this repository on your local machine.
Navigate to the repository's directory using your terminal.
Run the Salinas-train-predict.py script by entering the following command in the terminal:
This command will start the training process using the Salinas dataset and, upon completion, perform predictions. Please ensure that the Salinas dataset is correctly placed in the directory and the path is accurately specified within the Salinas-train-predict.py script.
Feel free to explore the script for understanding the training and prediction process, and adjust any parameters if necessary.