<p>Nonlinear ill-posed integral equations play a central role in a variety of scientific and engineering applications, including inverse modeling and signal reconstruction. However, their numerical treatment remains challenging due to both the intrinsic instability of the underlying problem and the high computational cost of conventional solution methods, especially in the presence of noisy data. Although the Levenberg–Marquardt (LM) method provides a stable regularization framework, its direct use in multiscale discretization environments results in prohibitively expensive computations, particularly for large-scale systems derived from fine discretizations. In this work, we propose a fast multiscale Galerkin method that effectively combines the Levenberg–Marquardt scheme with matrix compression strategy and multilevel iterative algorithm. The resulting approach, termed the Fast Discretized Levenberg–Marquardt (FDLM) method, substantially improves computational performance relative to earlier implementations such as the Discrete Levenberg–Marquardt (DLM) and Modified Levenberg–Marquardt (MLM) methods. By leveraging matrix compression and multilevel iteration, the FDLM method significantly reduces computational and storage costs while maintaining solution accuracy. Theoretical analysis under suitable assumptions establishes the convergence of approximate solutions generated by the FDLM method and proves order-optimal convergence rates when the balance principle is adopted as the stopping rule. Numerical experiments are provided to validate the efficiency of the algorithm and the effectiveness of the balancing principle, demonstrating that the proposed approach delivers both rapid computation and accurate, stable reconstructions.</p>