GeoBorderNet: An EfficientNet-B4–Vision Transformer Framework for Satellite-Based Border Surveillance
Contributors
Devanathan M
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Satellite imagery enables repeated observation of broad and difficult-to-access regions, but automated interpretation requires both local structural information and broader spatial context. GeoBorderNet combines EfficientNet-B4 and a Vision Transformer (ViT) through late feature fusion for five-class satellite-image classification. EfficientNet-B4 produces a 1280-dimensional local representation and the ViT a 768-dimensional contextual representation; concatenation forms a 2048-dimensional feature vector. Sentinel-2 imagery is considered for Normal Border Region, Border Infrastructure, Unauthorized Activity Region, Terrain Change, and Environmental Change. The methodology emphasizes deterministic preprocessing, training-only augmentation, geographic and temporal leakage control, staged transfer learning, and controlled ablation. This short paper presents only the methodology and does not report experimental results.