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

Building detection from satellite images is a demanding task and also considered as a hot research topic over the past few years. To be more specific, the satellite images become very significant for geographic information system (GIS) application, building detection and disaster monitoring. The key issue is how to recognize the objects of interest concerning building in satellite images swiftly, accurately with minimum falsification and noise. Deep Learning (DL) is one of the most efficient techniques for ensuring experience-based decision making in an automatic fashion. With these techniques expected output are said to be produced for the unseen inputs from learned patterns. Therefore, DL-based models are in the recent years being applied for object detection. One of the biggest challenges in the DL-based models learning is the selection of parameter and optimization process. Also numerous models have been proposed using bio-inspired optimization solutions to solve this problem. In this work, to avoid local optima and ensuring a smooth balance of exploration and exploitation involved in building detection, a method called, Secant Deep Belief Network-based Hyperbolic Cosine Whale Optimization (SDBN-HCWO) is proposed. The bio-inspired Hyperbolic Cosine Whale Optimization processes works under the Secant Deep Belief Network. The Secant Deep Belief Network consists of visible and hidden layer. In our work, three hidden layers are employed to detect best edges in the first hidden layer by means of Hyperbolic Cosine Prey Encircling-based best edge Identification model, linking best edges in the second hidden layer via Shrinking Encircle and Spiral Update-based optimal edge linking model and robust building detection in the third hidden layer by employing Secant Object Detection model. The performance of the SDBN-HCWO method is evaluated quantitatively and qualitatively based on the best fitness values. The experimental outcomes show that the proposed SDBN-HCWO method yields better performance results in terms of PSNR, false positive rate, classification accuracy, classification time, structural similarity index (SSIM), feature similarity index (FSIM) and convergence epochs for significant building detection than the other state-of-the-art methods.