osamabinIaggin/project-vision

Deep-learning + GIS pipeline detecting flood-driving drainage encroachment from 5 cm aerial imagery — U-Net semantic segmentation, multi-temporal change detection (Accra, Ghana) # VISION — Geospatial Deep Learning for Flood-Driver Detection > Computer vision on centimetre-scale aerial imagery to localise the *human* causes > of urban flooding — drainage encroachment and obstructed waterways — that > conventional satellite remote sensing cannot resolve. **VISION** (*Vulnerability Inference for Submersion-prone Informal settlements via Orthoimagery and Networks*) is an applied geospatial-AI system that couples **deep-learning semantic segmentation** — a **U-Net convolutional neural network** — with **multi-temporal change detection** and **topology-aware spatial reasoning** to detect and quantify the anthropogenic drivers of pluvial flooding in Accra, Ghana, at a resolution roughly two orders of magnitude finer than the satellite imagery on which prior work has relied. ### Technical approach - **Semantic segmentation (U-Net / CNN)** — pixel-wise extraction of buildings, drainage, and encroachment from 5 cm RGB orthomosaics - **Multi-temporal change detection** — 2020 vs 2024 epochs to quantify the *growth* of encroachment onto watercourses - **Topographic flood-susceptibility modelling** — gradient-boosted / random-forest ensembles over terrain morphometrics (slope, flow accumulation, TWI) - **Topology-aware geospatial overlay** — reconciling segmented structures against the hydrographic network to rank hazard loci - **Open-channel hydraulics** — Manning conveyance capacity of field-surveyed drain cross-sections, and its collapse under progressive siltation - **Corridor observation** — drain-aligned image chips cut at native 2–5 cm along every surveyed segment, to observe channel condition rather than assume it - **Reproducible by construction** — open data, scripted acquisition, and a version-controlled methodology rather than committed binary payloads --- ## Abstract Recurrent and frequently catastrophic inundation in metropolitan Accra is, on the preponderance of the evidence, an anthropogenic rather than a climatological phen …