TDA-SSNN: Efficient Hybrid Top-Down Attention Siamese Sparrow Neural Network for Automatic Drone Navigation

Unmanned aerial vehicles (UAVs) are increasingly employed in diverse applications such as military operations, delivery services, traffic and architectural monitoring, and aerial surveillance. These drones often encounter challenges in navigating complex and dynamic environments, including irregular object shapes, occlusions, and varying contrasts with backgrounds. To overcome these obstacles, this study presents a novel hybrid approach that integrates Top-Down Attention with a Siamese Sparrow Neural Network (TDA-SSNN) for efficient and data-driven drone navigation. The methodology begins with sourcing input images from publicly available datasets, including those from Udacity's project. To enhance image quality, the Robust Generalized Maximum Blake-Zisserman Total Correntropy (RMB-ZTC) adaptive filter is employed for preprocessing. Feature selection is performed using a novel Gorilla Troops Optimizer with Artificial Hummingbird Optimization (GTO-AMO). For automatic drone navigation, the TDA-SSNN is utilized, enhanced by an advanced Sparrow Search Algorithm to optimize the performance of the Top-Down Attention-based Siamese Neural Network. The proposed approach achieves a remarkable accuracy of 99.6%.