Scholarlink

<p>This study leverages advances in sensor technology and platforms to support sustainable development by extracting building information from remotely sensed data. The novelty of this study lies in utilizing the Whale Optimization Algorithm (WOA) to optimize the hyperparameters of U-Net with a ResNet backbone (ResU-Net) for extracting buildings from unmanned aerial vehicle (UAV) orthomosaics. The hyperparameters (learning rate, optimizer, epochs, loss, and activation functions) were encoded within defined ranges and evaluated through a fitness function based on the intersection over union (IoU) score. The proposed model was evaluated using datasets from four distinct test areas, focusing on accuracy, precision, recall, mean intersection over union (MIoU), and F1 scores. The comparative assessment indicated that the proposed WOA-ResU-Net consistently enhanced accuracy and improved the balance between recall and precision compared to U-Net and ResU-Net across all test areas. WOA-ResU-Net attained accuracy, precision, recall, F1-score and MIoU scores of 0.9770, 0.9873, 0.9318, 0.9588, and 0.9447 for Test Area-1, 0.9403, 0.7436, 0.9108, 0.8188, and 0.8087 for Test Area 2, 0.9391, 0.9489, 0.9274, 0.9963, and 0.9381 for Test Area-3, and 0.9687, 0.9458, 0.9377, 0.9417, and 0.9240 for Test Area-4, respectively. The theoretical significance and practical implication of this study include advancing the use of metaheuristics in optimizing deep learning algorithms and improved building extraction for urban planning, infrastructure monitoring, and disaster management.</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

WOA-ResU-Net: a whale optimization algorithm enhanced U-Net with ResNet backbone for building segmentation from UAV orthomosaics

  • 期刊论文

  • 发表日期：

    2025年11月29日

查看全文
  • Richmond Akwasi Nsiah,
  • Saviour Mantey,
  • Yao Yevenyo Ziggah

摘要

This study leverages advances in sensor technology and platforms to support sustainable development by extracting building information from remotely sensed data. The novelty of this study lies in utilizing the Whale Optimization Algorithm (WOA) to optimize the hyperparameters of U-Net with a ResNet backbone (ResU-Net) for extracting buildings from unmanned aerial vehicle (UAV) orthomosaics. The hyperparameters (learning rate, optimizer, epochs, loss, and activation functions) were encoded within defined ranges and evaluated through a fitness function based on the intersection over union (IoU) score. The proposed model was evaluated using datasets from four distinct test areas, focusing on accuracy, precision, recall, mean intersection over union (MIoU), and F1 scores. The comparative assessment indicated that the proposed WOA-ResU-Net consistently enhanced accuracy and improved the balance between recall and precision compared to U-Net and ResU-Net across all test areas. WOA-ResU-Net attained accuracy, precision, recall, F1-score and MIoU scores of 0.9770, 0.9873, 0.9318, 0.9588, and 0.9447 for Test Area-1, 0.9403, 0.7436, 0.9108, 0.8188, and 0.8087 for Test Area 2, 0.9391, 0.9489, 0.9274, 0.9963, and 0.9381 for Test Area-3, and 0.9687, 0.9458, 0.9377, 0.9417, and 0.9240 for Test Area-4, respectively. The theoretical significance and practical implication of this study include advancing the use of metaheuristics in optimizing deep learning algorithms and improved building extraction for urban planning, infrastructure monitoring, and disaster management.