TripletWatershed_Code

This repository if the official implementation of Triplet-Watershed for Hyperspectral Image Classification. [HAL] [arXiv].

Abstract: Hyperspectral images (HSI) consist of rich spatial and spectral information, which can potentially be used for several applications. However, noise, band correlations and high dimensionality restrict the applicability of such data. This is recently addressed using creative deep learning network archi- tectures such as ResNet, SSRN, and A2S2K. However, the last layer, i.e the classification layer, remains unchanged and is taken to be the softmax classifier. In this article, we propose to use a watershed classifier. Watershed classifier extends the watershed operator from Mathematical Morphology for classification. In its vanilla form, the watershed classifier does not have any trainable parameters. In this article, we propose a novel approach to train deep learning networks to obtain representations suitable for the watershed classifier. The watershed classifier exploits the connectivity patterns, a characteristic of HSI datasets, for better inference. We show that exploiting such characteristics allows the Triplet-Watershed to achieve state-of-art results in supervised and semi-supervised contexts. These results are validated on Indianpines (IP), University of Pavia (UP), Kennedy Space Center (KSC) and University of Houston (UH) datasets, relying on simple convnet architecture using a quarter of parameters compared to previous state-of-the-art networks. The source code for reproducing the experiments and sup- plementary material (high resolution images) is available at https://github.com/ac20/TripletWatershed Code.

Requirements

To install requirements:

conda env create -f environment.yml

Training

Use Main.py for training. Output of python Main.py -h.

usage: Main.py [-h] [--dataset {indianpines,paviaU,ksc,houston}] [--seed SEED] [--train_size TRAIN_SIZE]
               [--embed_dim EMBED_DIM] [--patch_size PATCH_SIZE] [--semi_supervised]

optional arguments:
  -h, --help            show this help message and exit
  --dataset {indianpines,paviaU,ksc,houston}
                        Dataset to work on. Default:indianpines
  --seed SEED           Set the seed for train/test split of dataset. Default:42
  --train_size TRAIN_SIZE
                        Train Size. Default:0.1
  --embed_dim EMBED_DIM
                        Embedding Dimension. Default:64 
  --patch_size PATCH_SIZE
                        Patch size, Default:64
  --semi_supervised     To use semi-supervised split or not. Default:False

Simple example to train on Indian Pines dataset.

python Main.py --dataset indianpines --seed 42

Output files - ./dump/results_006ea0a4.txt, ./dump/indianpines_006ea0a4.png and ./dump/weights_model_006ea0a4.pth will be created. 006ea0a4 is the token generated for this configuration. Sample output of the results:

Train    OA : 1.0 
Train    AA : 1.0 
Train Kappa : 1.0 
Test     OA : 0.9992413568873957 
Test     AA : 0.9990659493549783 
Test  Kappa : 0.9991351157339922 
        MAP : 0.9958027753801199 
 Train time : 517.2645268440247 
  Test time : 3.1228127479553223 

Results

Results acheived on using 10% of the data for training. See tables II - V in the article for details.

Dataset	OA	AA	Kappa
Indian Pines	99.57 ± 0.0026	99.62 ± 0.0029	0.9951 ± 0.0030
University of Pavia	99.98 ± 0.001	99.97 ± 0.001	0.9998 ± 0.001
Kennedy Space Center	99.72 ± 0.0023	99.62 ± 0.0032	0.9969 ± 0.0026
University of Houston	99.25 ± 0.0039	99.32 ± 0.0031	0.9919 ± 0.0042

Results using Semi-Supervised split - 30 training points per class. See tables VI, VII in the article for details.

Dataset	OA	AA	Kappa
Indian Pines	96.74 ± 0.0194	98.53 ± 0.0098	0.9627 ± 0.0221
University of Pavia	99.20 ± 0.0129	98.95 ± 0.0165	0.9894 ± 0.0170

Remark 1 : To run in the semi-supervised mode use (for example above)

python Main.py --dataset indianpines --seed 42 --semi_supervised

Remark 2 : In the following images - the bright patches indicate the differences of the Triplet Watershed results with the ground-truth.

Indian Pines

Ground Truth	Triplet-Watershed Result

University of Pavia

Ground Truth	Triplet-Watershed Result

Kennedy Space Centre

Ground Truth	Triplet-Watershed Result

University of Houston

Ground Truth

Triplet-Watershed Result