Deep Multi-Scale Contextual Encoding in U-Net for Per-Pixel Classification of High-Variability Microscopic Images
Procedia Computer Science · 2026 · 283 · 62-70
DOI: 10.1016/j.procs.2026.06.095 ↗Kurz gefasst
Fachartikel in der Zeitschrift Procedia Computer Science, 2026. Verfasst von B. Sirisha und 3 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Microscopic Image analysis is instrumental in many scientific and medical fields, requiring accurate semantic segmentation of microscopic structures for effective interpretation. Convolutional neural networks (CNNs) have shown remarkable success in this task, with U-Net being prominent architecture for Per-Pixel Classification tasks. Baseline U-Net architectures may struggle to capture deep multi-scale features essential for segmenting objects of varying sizes in microscopic images. Herein, we put forward an improvised U-Net architecture specifically tailored for microscopic image segmentation by incorporating deep multi-scale skip connections. Unlike traditional skip-connections limited to corresponding encoder and decoder layers, our approach integrates skip connections between layers at different scales, facilitating the extraction of deep multi-scale features crucial for the semantic segmentation of microscopic structures. Our approach’s effectiveness is demonstrated by experimental findings on a variety of datasets, with a significant improvement in Dice Score from 0.9100934 for standard U-Net to 0.9440559 for our improvised deep multi-scale U-Net architecture. The observed performance enhancement underscores the importance of capturing scale-invariant features in microscopic image segmentation, highlighting the robustness and effectiveness of our proposed approach.
Zitieren
APA
Sirisha, B., Jyothishree, M. V. R., Prasad, P. P., & Lakshmi, T. (2026). Deep Multi-Scale Contextual Encoding in U-Net for Per-Pixel Classification of High-Variability Microscopic Images. Procedia Computer Science, 283, 62-70. https://doi.org/10.1016/j.procs.2026.06.095
Vancouver
Sirisha B, Jyothishree MVR, Prasad PP, Lakshmi T. Deep Multi-Scale Contextual Encoding in U-Net for Per-Pixel Classification of High-Variability Microscopic Images. Procedia Computer Science. 2026;283:62-70. doi:10.1016/j.procs.2026.06.095
BibTeX
@article{Sirisha2026,
author = {Sirisha, B. and Jyothishree, M V R and Prasad, P Phani and Lakshmi, T},
title = {Deep Multi-Scale Contextual Encoding in U-Net for Per-Pixel Classification of High-Variability Microscopic Images},
journal = {Procedia Computer Science},
year = {2026},
volume = {283},
pages = {62--70},
doi = {10.1016/j.procs.2026.06.095}
}
EUVIMED