Scholarlink

<p>Accurately predicting the resonant frequencies of microstrip antennas is crucial for efficient antenna design and optimisation, yet traditional analytical and numerical methods often face challenges in handling complex parameter interactions. This paper presents a novel approach to predict the resonant frequencies of microstrip antennas using convolutional neural networks (CNNs) and image-based encoding of antenna parameters. The proposed method encodes the key design parameters—length (<i>L</i>), width (<i>W</i>), height (<i>h</i>), and relative permittivity (<i>ε</i><sub><i>r</i></sub>)—into 2 × 2 and 4 × 4 RGB images, where each parameter is mapped to specific colour channels or derived spatial features. These encoded images are utilized as inputs to a CNN architecture tailored for regression tasks, predicting the resonant frequency as a continuous output. The model demonstrates superior prediction accuracy for training and testing on a comprehensive dataset of microstrip antenna designs, achieving a low average percentage error (APE). The CNN effectively captures the complex relationships between antenna parameters and their corresponding resonant frequencies by leveraging spatial and feature-derived patterns in the RGB-encoded images. This approach offers a novel perspective on antenna design optimisation, enabling a highly accurate, automated, and scalable solution to predict antenna performance. The results underscore the potential of image-based encoding in enhancing the rapid design and optimisation of microstrip antennas.</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

Enhancing antenna frequency prediction using convolutional neural networks and RGB parameters mapping

  • 期刊论文

  • 发表日期：

    2025年10月6日

查看全文
  • Ritika Singh,
  • Aditya Singh,
  • Ashok Jangid,
  • Meghna Sharma,
  • Pratibha Rashmi,
  • Manu Pratap Singh

摘要

Accurately predicting the resonant frequencies of microstrip antennas is crucial for efficient antenna design and optimisation, yet traditional analytical and numerical methods often face challenges in handling complex parameter interactions. This paper presents a novel approach to predict the resonant frequencies of microstrip antennas using convolutional neural networks (CNNs) and image-based encoding of antenna parameters. The proposed method encodes the key design parameters—length (L), width (W), height (h), and relative permittivity (εr)—into 2 × 2 and 4 × 4 RGB images, where each parameter is mapped to specific colour channels or derived spatial features. These encoded images are utilized as inputs to a CNN architecture tailored for regression tasks, predicting the resonant frequency as a continuous output. The model demonstrates superior prediction accuracy for training and testing on a comprehensive dataset of microstrip antenna designs, achieving a low average percentage error (APE). The CNN effectively captures the complex relationships between antenna parameters and their corresponding resonant frequencies by leveraging spatial and feature-derived patterns in the RGB-encoded images. This approach offers a novel perspective on antenna design optimisation, enabling a highly accurate, automated, and scalable solution to predict antenna performance. The results underscore the potential of image-based encoding in enhancing the rapid design and optimisation of microstrip antennas.