TransPUNet: Transformer-CNN 하이브리드 모델을 활용한 InSAR의 Phase Unwrapping 정밀도 향상 모델

초록

Phase unwrapping, a critical preprocessing step in analyzing InSAR (Interferometric Synthetic Aperture Radar) data, aims to reconstruct absolute phase values from interferometric phase measurements. However, conventional methods are often vulnerable to noise and phase discontinuities. In this paper, we propose TransPUNet, a hybrid model that combines CNN and Transformer modules in parallel, and introduce a composite loss function that integrates Mean Squared Error (MSE) and Structural Similarity Index Measure (SSIM) to preserve structural integrity. Experimental results using Digital Elevation Map (DEM)-based simulated datasets demonstrate that the proposed model achieves a mean Root Mean Square Error (mRMSE) that is approximately 51% lower than that of existing deep learning- based approaches, thereby validating its superior reconstruction accuracy.

키워드