NA-STM Enhanced Energy-Efficient Custom Spiking U-Net Denoiser for Noise-Robust CIFAR-10 Classification
IEEE Open Journal of the Computer Society · 2026 · 7 · 647-658
DOI: 10.1109/ojcs.2026.3681673 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Open Journal of the Computer Society, 2026. Verfasst von Chetali Yadav und 1 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Event-driven convolutional spiking neural network (CSNN) based U-Nets provide a more energy-efficient solution for image denoising and classification applications compared to convolutional U-Net. We introduce a novel approach of noise-adaptive spike threshold modulation (NA-STM) with custom spiking U-Net, batch normalisation through time (BNTT) and CosineAnealingWarmRestarts (CAWR) scheduler to denoise the noisy CIFAR-10 images corrupted by Gaussian noise (GN), speckle noise (SN) and salt & pepper noise (S&PN). We evaluated the denoising performance using peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), mean square error (MSE) and improvement in classification accuracy of the denoised CIFAR-10 images. We achieved an improvement in classification accuracy of $\sigma _{25}$ noise level CIFAR-10 images by 25.35%, 35.16% and 38.9% after denoising in case of GN, SN and S&PN, respectively. Our proposed model achieved better denoising performance and an improvement in classification accuracy compared to other published works. Our proposed custom spiking U-Net’s computational complexity is 24.8% and energy consumption is 4.83% compared to the equivalent CNN U-Net, hence, our proposed spiking U-Net is suitable for deployment on low power edge devices.
Zitieren
APA
Yadav, C., & Reniwal, B. S. (2026). NA-STM Enhanced Energy-Efficient Custom Spiking U-Net Denoiser for Noise-Robust CIFAR-10 Classification. IEEE Open Journal of the Computer Society, 7, 647-658. https://doi.org/10.1109/ojcs.2026.3681673
Vancouver
Yadav C, Reniwal BS. NA-STM Enhanced Energy-Efficient Custom Spiking U-Net Denoiser for Noise-Robust CIFAR-10 Classification. IEEE Open Journal of the Computer Society. 2026;7:647-658. doi:10.1109/ojcs.2026.3681673
BibTeX
@article{Yadav2026,
author = {Yadav, Chetali and Reniwal, Bhupendra Singh},
title = {NA-STM Enhanced Energy-Efficient Custom Spiking U-Net Denoiser for Noise-Robust CIFAR-10 Classification},
journal = {IEEE Open Journal of the Computer Society},
year = {2026},
volume = {7},
pages = {647--658},
doi = {10.1109/ojcs.2026.3681673}
}
EUVIMED