Description
The dataset includes non-expansive deep neural network (DNN) denoisers, each trained to remove additive white Gaussian noise with zero mean and a specified standard deviation. The denoisers underpin the Plug-and-Play (PnP) algorithm HyperAIRI for hyperspectral image formation in radio astronomy.
The HyperAIRI denoiser takes as input the channel image of interest, its two adjacent channel images, an estimated spectral index map, and frequency ratios as side information, and outputs the denoised channel image.
The shelf of denoisers consists of six DNNs, each trained for noise levels in the set {3.2×10⁻⁴, 1.6×10⁻⁴, 8×10⁻⁵, 4×10⁻⁵, 2×10⁻⁵, 1×10⁻⁵}.
All DNNs are provided in ONNX format and can be deployed on images of arbitrary size, limited only by hardware memory constraints.
The HyperAIRI denoiser takes as input the channel image of interest, its two adjacent channel images, an estimated spectral index map, and frequency ratios as side information, and outputs the denoised channel image.
The shelf of denoisers consists of six DNNs, each trained for noise levels in the set {3.2×10⁻⁴, 1.6×10⁻⁴, 8×10⁻⁵, 4×10⁻⁵, 2×10⁻⁵, 1×10⁻⁵}.
All DNNs are provided in ONNX format and can be deployed on images of arbitrary size, limited only by hardware memory constraints.
| Date made available | 2025 |
|---|---|
| Publisher | Heriot-Watt University |