Enhanced Superpixel-Based Multiscale CNN Mechanism for UAV Image Segmentation

Unmanned Aerial Vehicles (UAVs) are advanced remote sensing tools that have the potential to revolutionize variety applications,including environmental monitoring, urban planning, agriculture, and disaster management. These airborne sensors provide high resolution, real-time data used for traditional remote sensing methods often struggle to capture, particularly in inaccessible or large-scale areas. To address challenges in misclassification in complex urban aerial imagery, we proposed a super-pixel-aided multiscale Convolutional Neural Network (CNN) architecture. This approach integrates an attention mechanism, and the SLICO algorithm. The attention mechanism enhances the model’s focus on crucial image regions, optimizing feature extraction. The SLICO algorithm generates super-pixels to reduce computational costs and refine boundary detection. This integrated approach effectively addresses scale variance in aerial imagery, resulting in more precise segmentation. We evaluated the model using UAV-based dataset: the Urban Drone Dataset (UDD). The proposed model significantly outperformed several state-of-the-art methods, achieving impressive Intersection over Union (IoU) scores on dataset.In recent years, unmanned aerial vehicles (UAVs) have gained significant attention across a wide range of domains, including urban planning, precision agriculture, disaster response, environmental monitoring, and infrastructure surveillance. Their ability to capture high- resolution images at low operational costs, coupled with flexible deployment and maneuverability, makes them a powerful tool in the field of remote sensing. The wealth of visual data collected by UAVs provides valuable insights for analysis, yet this same abundance introduces unique computational and methodological challenges.