Scholarlink

<p>The Intelligent Transportation System (ITS) is essential for effective roadway management, particularly given the increasing demand for rapid ambulance response in growing urban populations. This paper addresses the challenges of ambulance detection using contemporary computer vision techniques, which often encounter difficulties due to the limited distinguishing features of emergency vehicles. We suggest a new deep convolutional neural network (DCNN) that uses advanced transfer learning models like MobileNet and Xception to improve prediction accuracy and lower the number of parameters. To further optimize the model, we apply Harris Hawk Optimization (HHO) to fine-tune hyperparameters, thereby maximizing prediction performance. Our approach also incorporates preprocessing algorithms for input images and evaluates various optimization techniques during testing. We utilize integrated datasets to mitigate the lack of publicly available resources, including a new dataset of 4749 images. The results show that our method significantly outperforms fourteen existing state-of-the-art techniques, including a wide variety of CNN models and vision transformations, achieving an accuracy of 99.86%, a precision of 100%, a recall of 99.7%, an F1-score of 99.85%, and a Matthews correlation coefficient (MCC) of 99.72% while maintaining an efficient parameter structure suitable for mobile and edge devices. The outcomes showcase an applicable framework that enhances ITS applications.</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

Innovative Convolutional Neural Networks Based on the Harris Hawk Optimization Algorithm for Ambulance Detection 

  • 期刊论文

  • 发表日期：

    2025年12月11日

查看全文
  • Ahmed M. Gab Allah,
  • Amany M. Sarhan,
  • Mahmoud Abdelwahab

摘要

The Intelligent Transportation System (ITS) is essential for effective roadway management, particularly given the increasing demand for rapid ambulance response in growing urban populations. This paper addresses the challenges of ambulance detection using contemporary computer vision techniques, which often encounter difficulties due to the limited distinguishing features of emergency vehicles. We suggest a new deep convolutional neural network (DCNN) that uses advanced transfer learning models like MobileNet and Xception to improve prediction accuracy and lower the number of parameters. To further optimize the model, we apply Harris Hawk Optimization (HHO) to fine-tune hyperparameters, thereby maximizing prediction performance. Our approach also incorporates preprocessing algorithms for input images and evaluates various optimization techniques during testing. We utilize integrated datasets to mitigate the lack of publicly available resources, including a new dataset of 4749 images. The results show that our method significantly outperforms fourteen existing state-of-the-art techniques, including a wide variety of CNN models and vision transformations, achieving an accuracy of 99.86%, a precision of 100%, a recall of 99.7%, an F1-score of 99.85%, and a Matthews correlation coefficient (MCC) of 99.72% while maintaining an efficient parameter structure suitable for mobile and edge devices. The outcomes showcase an applicable framework that enhances ITS applications.