研究目的
To address the electromagnetic inverse scattering problem using complex convolutional neural networks for super-resolution electromagnetic imaging.
研究成果
The proposed complex CNN method can perform super-resolution in electromagnetic imaging on both simulation and real data, demonstrating great generalization capacity. It offers a non-iterative and fast method for addressing real-time practical large-scale electromagnetic inverse scattering problems.
研究不足
The traditional iteration methods are inappropriately applied to large-scale EMIS problems due to inherent expensive computational cost. The proposed method aims to address this but may still face challenges in very large-scale applications.
1：Experimental Design and Method Selection
The methodology involves extending traditional convolutional neural networks to the complex field using a specific equation for convolutional operations in complex fields. The network is designed as a series of cascaded CNN modules, each consisting of a three-layer residual convolutional neural network.
2：Sample Selection and Data Sources
The performance of the proposed networks is investigated using the MNIST database. The fields are generated considering a monochromatic radar scattering multistatic/multiview configuration. The dataset consists of 10000 image pairs, divided into 8000 for training and 2000 for blind testing.
3：List of Experimental Equipment and Materials
Not explicitly mentioned in the provided text.
4：Experimental Procedures and Operational Workflow
The investigation domain is illuminated by TM-polarized incident waves at 4 GHz. The region is discretized into 56 by 56 sub-squares, with the relative permittivity of targets set as 3. The reconstructions from the BP algorithm serve as inputs for training networks, with the ground truth as outputs.
5：Data Analysis Methods
The quality of reconstructions is assessed through visual comparison and statistical histograms of SSIM index for reconstructions.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
