Per-Pixel CNN (PCNN) for high-resolution imagery classification

This project aims to classify high-resolution images by ultilizing deep convolutional neural networks. If you find this to be helpful, please cite our paper "Spectral–spatial feature extraction for hyperspectral image classification: A dimension reduction and deep learning approach,IEEE Transactions on Geoscience and Remote Sensing 54 (8), 4544-4554". Thanks!

Step.1: Generate image patch-based traning dataset using "generate_dataset.m"

In this step, 10% of available samples were selected to train CNN models.

Step.2: Train a 5-layer CNN with "vai_train_rs_cnn.m"

Step.3: Predict the whole map's labels with the well-trained CNN