抄録
We propose a multimodal CNN that reconstructs conductivity distribution images by combining features from multiple sets of potential data. A comparison with a single modal CNN, which reconstructs images from a single set of potential data, was performed. The reconstructed images were evaluated for the size and position of foreign objects. While the multimodal CNN did not show significant improvement in terms of size accuracy compared to the single modal CNN, it demonstrated superior performance in position estimation. Additionally, we examined reconstructions of polygonal objects and objects with rotation. The multimodal CNN exhibited higher resolution in capturing object shapes and rotations compared to the single modal CNN. Future work will focus on applying this method to the non-destructive inspection of building materials, beyond simulations.
