Fast and Automatic Parametric Modeling of Antenna Structure Using CNN-LSTM Network for Efficient Data Generation

In recent years, deep learning-assisted methods for antenna design, such as surrogate models, have gained popularity due to their potential to improve design efficiencies by replacing time-consuming electromagnetic simulations. However, these methods require a large number of diverse and representative training data samples, which can be challenging to acquire using traditional manual modeling methods. To address this issue, our study proposes an intelligent and image-based parametric modeling approach that leverages deep learning to accelerate the acquisition of antenna samples with different physical structures. Our method takes an image of the antenna structure as input, typically available in scientific publications, and automatically generates the corresponding modeling codes using Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks. The CNN captures features from antenna structure images, while the LSTM generates the modeling codes. Through extensive training, our proposed model achieves fast and automatic data acquisition of antenna physical structures based solely on antenna images. Experimental results demonstrate a significant speed enhancement compared to the manual modeling approach. This approach lays the foundation for efficient data acquisition, which is crucial for building robust surrogate models in the future.