Publications

CoSA: Correlation-Guided Change Attention with Learnable Residual Gating for Remote Sensing Change Detection

Abdirashid Omar, Jonghyuk Park

IEEE Access, 2026

TLDR: A lightweight decoder-side refinement module for remote sensing change detection that uses bi-temporal feature correlation, multi-scale context aggregation, and learnable residual gating to improve changed-region detection.

CHANGE DETECTION, COMPUTER VISION, IEEE ACCESS, REMOTE SENSING
IEEE   DOI   arXiv   Code

UrbanCDNet: Appearance-Robust and Boundary-Aware Bitemporal Change Detection for Korean Urban Building Monitoring

Abdirashid Omar, Jonghyuk Park

arXiv preprint, 2026

TLDR: A task-specific Siamese CNN for sparse Korean urban building change under photometric variation and temporal misalignment, with locked-test F1 of 0.7511 and IoU of 0.6014.

ARXIV, CHANGE DETECTION, REMOTE SENSING, VISION AI
arXiv   GitHub

Compact Autoencoder Representations for Historical Retrieval of African Rainfall Anomalies

Abdulhakin Mohamud Ismail, Abdirashid Omar, Abdikadir Uggas

Deep Learning Indaba, Poster, 2026

TLDR: A Deep Learning Indaba poster studying compact autoencoder representations for retrieving historically similar African rainfall-anomaly patterns from monthly CHIRPS rainfall fields.

AUTOENCODERS, CLIMATE AI, DATA SCIENCE, DEEP LEARNING, POSTER
Poster   Preview

Data-Efficient Crosswalk Segmentation from Overhead CCTV via Confidence- and Geometry-Guided Pseudo-Labeling

Abdirashid Omar, Jonghyuk Park

arXiv preprint, 2026

TLDR: A data-efficient DeepLabV3-ResNet50 pipeline for overhead-CCTV crosswalk segmentation that reaches 88.91% IoU on 40 held-out, manually labelled images using 241 manual annotations and confidence- and geometry-guided pseudo-label selection from 5,926 unlabelled frames.

ARXIV, COMPUTER VISION, DOMAIN ADAPTATION, PSEUDO-LABELING, SEMANTIC SEGMENTATION
arXiv   Code   Project note