Material level measurement is a core requirement in industrial automation management, and millimeter-wave radar imaging technology serves as a key approach for achieving precise measurement. The imaging quality of the material surface profile directly determines the reliability and accuracy of monitoring. However, existing super-resolution methods struggle to effectively reconstruct clear material surface profiles and their performance deteriorates significantly under low signal-to-noise ratio (SNR) conditions. This paper proposes an imaging method for material surface profiles based on the joint optimization of iterative adaptive approach (IAA) and total variation (TV) regularization. First, a corner reflector calibration method is employed to compensate for gain and phase inconsistencies across the actual array channels, thereby ensuring the reliability of the echo data. Secondly, an IAA-guided Iteratively reweighted least squares (IRLS) robust algorithm is employed to mitigate the generation of false contours—a common defect of conventional IRLS-based TV regularization solutions caused by noise interference. Within the IRLS framework, the non-smooth regularization problem is converted into a series of weighted least squares problems that can be solved iteratively. The gradient information derived from the IAA-estimated angular power spectrum is then utilized as spatial prior information to assess the probability of material edge presence at different angular positions. This enables the adaptive adjustment of the penalty intensity within the TV weights, effectively eliminating false contours in low signal-to-noise ratio SNR environments. Furthermore, a feedback mechanism based on echo residuals and covariance matrix residuals is introduced to update the regularization parameters, overcoming the instability associated with manual hyperparameter tuning. Simulation and experimental results demonstrate that the proposed method enhances the imaging quality of material surface profiles and remains robust under low signal-to-noise ratio conditions. Compared with the Iterative Adaptive Approach IAA, the root mean square error is reduced by approximately 76%, and the image cross-entropy is reduced by about 7.8%; compared with TV-Sparse, the root mean square error is reduced by approximately 6%, and the image cross-entropy is reduced by about 3%.