LREN

We provide a Tensorflow implementation of LREN: Low-Rank Embedded Network for Sample-Free Hyperspectral Anomaly Detection (AAAI2021).

Framework of LREN:

Prerequisites

• Linux 18.04 LTS
• Python 3.7.8
• Tensorflow 1.5.1
• CUDA 10.2
• Scipy 1.2.1
• Numpy 1.18.5
• Matplotlib 3.3.0

Citation

If you use this code for your research, please cite:

Jiang, K., Xie, W., Lei, J., Jiang, T., & Li, Y. (2021). LREN: Low-Rank Embedded Network for Sample-Free Hyperspectral Anomaly Detection. Proceedings of the AAAI Conference on Artificial Intelligence, 35(5), 4139-4146.

Running Code

In this code, you can run our models on on four benchmark hyperspectral datasets, including SanDiego, Hydice, Coast, and Pavia.

Usage

python run_main_LREN.py

Result

Hyperspectral Datasets

For the ease of reproducibility. We provide experimental results on hyperspectral datasets as belows:

Dataset	AUC(P_d, P_f)	AUC(P_f, \tau)
SanDiego	0.9897	0.0134
Hydice	0.9998	0.0102
Coast	0.9982	0.0276
Pavia	0.9925	0.0433
Average	0.9951	0.0236

Extension

Since our approach is based on the following three properties:

1. The background (i.e., the normal instances) still preserves a low-rank property lying in a low-dimensional manifold.
2. The presence probability of the anomaly is much lower than that of the background (i.e., the normal instances).
3. The latent representation serves the anomaly estimation, which optimally updates the parameters of the deep latent space.

LREN is applicable to anomaly detection tasks that satisfy these three properties. We conducted experiments on Outlier Detection DataSets (ODDS) to demonstrate the effectiveness of LREN in other anomaly detection tasks.

Dataset	AUC(P_d, P_f)	AUC(P_f, \tau)	Precision	Recall	F1
Thyroid	0.9910	0.0980	0.8571	0.6452	0.7362
Arrhythmia	0.8353	0.0490	0.6389	0.451	0.5287