<p>The original method proposed in this work enables the reconstruction of the microstructure of multiphase media with precise interface detection by analyzing its thermal responses. This study proposes a novel approach that combines singular value decomposition (SVD) and level-set methods to accurately identify interfaces and different phases within an unsteady thermal field recorded via infrared thermography. This approach allows for reliable discrimination of interfaces even after thermal diffusion has occurred. Moreover, the use of SVD ensures robustness against measurement noise, which is inherent in all experimental setups. A mathematical formulation based on the SVD of temperature data transforms the dynamic problem into a static one. Accurate microstructure discrimination is achieved by evolving the level curves of a distance function until a cost function is minimized. This cost function corresponds to the squared error between the observed data and the data simulated by solving the direct problem. The velocity of domain evolution is defined by the derivative of the variational formulation with respect to the domain, which depends on solving both the direct and adjoint problems. Numerical experiments are conducted on a biphasic material with complex and random phase structures to demonstrate the effectiveness of the proposed method.</p>