Model inversion attacks (MIA) are increasingly becoming a significant concern in the field of privacy protection, as they reconstruct sensitive training data by analyzing the outputs of public models. Existing MIA techniques utilize Generative Adversarial Networks (GANs) to reconstruct high-dimensional data, such as facial images. In traditional model inversion attacks, the Adam optimizer is directly applied to the optimization process, which may lead to limitations in the search space and dependence on local optima, thereby affecting the accuracy of the model inversion attack and the authenticity of the generated images. To address these challenges, this study proposes a two-stage multi-strategy integrated model inversion attack method within the framework of existing MI attacks, aimed at enhancing the performance of MIA. Firstly, the Particle Swarm Optimization (PSO) algorithm is introduced, which is used for optimization search in the latent vector space, effectively avoiding the problem of traditional optimization methods easily falling into local optimal solutions, and is used to find the optimal latent vectors. Secondly, by integrating the Genetic Algorithm (GA), the diversity of the population is increased, preventing the algorithm from converging prematurely to local optima, while enhancing the global exploration ability to explore new and promising areas. Finally, the Adam gradient descent algorithm is used to fine-tune the latent vectors, achieving more precise optimization. Experiments have shown that the proposed attack method significantly improves the attack performance on various datasets and models, with higher quality and more realistic recovered images.