Recovery of impenetrable rough surface profiles via CNN-based deep learning architecture

Aydin, Izde; Budak, Güven; Sefer, Ahmet; Yapar, Ali

doi:10.1080/01431161.2022.2105177

Cited by 5 publications

(3 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To conquer the above problems, the inverse scattering problem is generally converted to an optimization problem, which can be solved by the 1) objective function method (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17) or 2) neural network learning method. (8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22) The objective function is generally used to find the extreme value of the problem, e.g., to calculate the minimum error of the scattering field and the measured scattering field. The gradient method is more suitable for small-area searches, (4) whereas the global algorithm is suitable for full-area searches.…”

Section: Introductionmentioning

confidence: 99%

“…Regarding the neural network learning method, (8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22) Yao et al used a two-stage neural network, in which the preliminary permittivity from the scattering field in the first stage and then the neural network was used to calculate the true permittivity. (10) Wei and Chen proposed a deep learning method to improve the results.…”

Section: Introductionmentioning

confidence: 99%

“…A deep learning convolutional neural network architecture for the shape imaging of perfectly electrically conducting rough surfaces was addressed in Ref. 22. The algorithm was numerically tested with various scenarios and proved to be very effective.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Microwave Imaging of Conductors by Direct Sampling Method and U-Net

Chiu¹,

Li²,

Chien³

et al. 2023

Sensors and Materials

View full text Add to dashboard Cite

Electromagnetic imaging is an emerging technology widely applied in many fields, such as medical imaging, biomedical imaging, and nondestructive testing. In this study, we place transmitter and receiver antennas around an unknown object. We can use the direct sampling method (DSM) to reconstruct the material size and shape of the unknown object on the basis of the scattered field. We apply U-Net to reconstruct electromagnetic images of perfect conductors. Perfect conductors in free space are studied by irradiating a transverse magnetic (TM) polarization wave. Using the scattered electric field measured outside the object together with the boundary conditions on the conductor surface, a set of nonlinear integral equations can be derived and further converted into matrix form by the method of moments. Since an iterative algorithm is computationally expensive and time-consuming, a real-time electromagnetic imaging technique combining deep learning neural networks is proposed for reconstructing the perfect conductors. The initial shapes of the conductors are first computed by DSM by using the scattered electric field measured outside the object. The initial shapes of the conductors are then input to U-Net for training. Numerical results show that U-Net is capable of reconstructing accurate conductor shapes. Therefore, artificial intelligence techniques can reconstruct shapes more accurately than iterative algorithms, when combined with DSM.

show abstract