Scholarlink

<p>Unmanned aerial vehicles (UAVs), are rapidly used for aerial imagery, object detection, and segmentation tasks. Occlusion degrades the performance of UAVs with obscured areas in aerial scenes. Existing methods like diffusion-based inpainting, transformer-based generative adversarial networks (GANs) models do not restore missing contents explicitly. This work introduces a unified framework that integrates gradient-based optimization, diffusion-based inpainting, contextual attention mechanism, and Bayesian uncertainty modeling in handling issues of dynamic occlusion. It also outperforms other image inpainting methods with better performance, minimizing temporal stability and gradient computation. Experimental results illustrate high-fidelity reconstruction of missing parts across diverse aerial scenes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Gradient-Optimized Framework for Occluded Aerial Image Reconstruction using Generative Adversarial Networks

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Pankaj Singh Jamwal,
  • Ayush Trivedi,
  • Robin Singh Bhadoria

摘要

Unmanned aerial vehicles (UAVs), are rapidly used for aerial imagery, object detection, and segmentation tasks. Occlusion degrades the performance of UAVs with obscured areas in aerial scenes. Existing methods like diffusion-based inpainting, transformer-based generative adversarial networks (GANs) models do not restore missing contents explicitly. This work introduces a unified framework that integrates gradient-based optimization, diffusion-based inpainting, contextual attention mechanism, and Bayesian uncertainty modeling in handling issues of dynamic occlusion. It also outperforms other image inpainting methods with better performance, minimizing temporal stability and gradient computation. Experimental results illustrate high-fidelity reconstruction of missing parts across diverse aerial scenes.