Within this research, aimed to solve the problem of face recognition in setting of partial face occlusion by personal protective equipment an approach was proposed, which assumed training of a neural network model FaceNet on a specialized dataset Masked VGGFace2. This paper presents a strategy of training dataset augmentation based on the convolutional neural network model MTCNN and MaskTheFace software. Using this suite, based on the VGGFace2 dataset, an extended dataset Masked VGGFace2 was established, including images of human faces, partially occluded by personal protective equipment. Within this research, a series of experiments was performed, aimed to qualitative assessment of predictions, obtained with this solution, in comparison to the FaceNet model, which was trained on the original VGGFace2 dataset. The performed experiments revealed, that the proposed solution demonstrates a significantly higher recognition quality of partially occluded faces (AP = 0.9488)-recognition accuracy increase as per average precision (AP) metric was over 24%, compared to the original model. On test subset from the original dataset VGGFace2, the value of the AP metric was 0.9635, what outperformed the respective metric of the original model by 2.8% and justified a high genericity of the proposed solution in terms of face recognition problem.