Diagnosis of array antennas based on near-field data using Faster R-CNN

In this paper, a source reconstruction method for detecting failures in array antenna elements using near-field data based on Faster region-convolutional neural network (Faster R-CNN) is introduced. The proposed method leverages the convergence and optimization of deep learning to eliminate the ill-posedness of the inverse equation of the electric field integral equation (EFIE). The near-field pattern images of an array are collected, divided into training and test datasets, and then fed into the Faster R-CNN to learn the inverse operator of the EFIE formula. Four different excitation states are trained using the same neural network to compare the effects of excitation faults on feature extraction and classification accuracy. A prototype of the array antenna is fabricated and experimentally verified, with good detection results. Simulation and measurement results show that the four types of array excitation state can be classified and located, and with a mean average precision (mAP) of 93.57%, a training time of 24 minutes, and an efficiency of 0.14 seconds per image. The proposed method realizes the classification and location diagnosis of the excitation faults of array antenna with low computing resource requirements, providing a high efficiency and accuracy evaluation method for the microwave array. • Using a source reconstruction method based on Faster R-CNN achieves diagnosis of array faults. • High-precision and efficient detection locates and classifies back-feeding faults in the array. • Design comparative experiments to verify the superiority of the method. • The experimental results verify the feasibility of the diagnosis method.