Secant Deep Hyperbolic Cosine Bio Inspired Whale Optimization for Building Detection From Satellite Images

Why It Matters

Accurate, rapid building detection enhances GIS applications, disaster response, and urban planning, giving stakeholders a more reliable satellite‑based intelligence source. The novel optimization reduces computational overhead, making large‑scale analyses more feasible.

Key Takeaways

• SDBN‑HCWO combines Secant DBN with Hyperbolic Cosine Whale Optimization.
• Three hidden layers detect edges, link edges, and identify buildings.
• Outperforms benchmarks on PSNR, false positives, accuracy, and speed.
• Reduces convergence epochs, enhancing training efficiency for satellite GIS.

Pulse Analysis

Satellite‑derived building footprints are a cornerstone for modern geographic information systems, supporting everything from city planning to emergency management. Traditional object‑detection pipelines struggle with the high variability of urban textures, illumination changes, and sensor noise. While deep learning has narrowed the accuracy gap, its performance hinges on optimal hyper‑parameter tuning and effective loss‑function minimization—areas where conventional gradient‑based methods often stall in local minima. Bio‑inspired metaheuristics, such as whale optimization, offer a promising alternative by mimicking natural foraging behaviors to explore solution spaces more globally.

The newly proposed SDBN‑HCWO framework marries a Secant Deep Belief Network with a Hyperbolic Cosine Whale Optimization engine. The Secant DBN provides a layered probabilistic model, where three hidden layers specialize in edge detection, edge linking, and final object classification. Hyperbolic cosine functions shape the whale’s encircling and spiral update dynamics, ensuring a smoother balance between exploration and exploitation. This synergy enables the system to escape sub‑optimal traps and converge faster, delivering sharper feature maps and more reliable building masks even under noisy conditions.

Benchmarking against state‑of‑the‑art detectors reveals that SDBN‑HCWO achieves higher peak signal‑to‑noise ratios, lower false‑positive rates, and improved structural similarity indices, all while cutting classification time and convergence epochs. For practitioners, these gains translate into faster turnaround for large‑scale mapping projects and more dependable inputs for downstream analytics like damage assessment after natural disasters. As satellite constellations proliferate and data volumes surge, optimization techniques that streamline deep‑learning pipelines will become increasingly vital, positioning SDBN‑HCWO as a forward‑looking solution for the GIS community.