The invention relates to a deep neural network model inversion attack defense method and equipment, and proposes a model inversion attack defense method based on generative adversarial network and fake samples. First, the generative adversarial network is used to generate false samples, and according to the false samples to ensure the effectiveness of the target victim model, the parameters of the target victim model are finetuned to realize the purpose of defending model inversion attack. The invention can effectively combat model inversion attacks, protect data privacy, and ensure the high availability of the model. 本发明涉及一种深度神经网络模型反演攻击防御方法及设备，提出了一种基于生成对抗网络和假样本的模型反演攻击防御方法，先利用生成对抗网络生成虚假样本，在根据虚假样本在保证目标受害者模型的有效性的基础上，微调目标受害者模型参数，从而实现防御模型反演攻击的目的。本发明可以有效地对抗模型反演攻击，保护数据隐私性，同时确保模型的高可用性。