With the types and forms of resumes becoming increasingly complex, the workload and difficulty of the human resources department in extracting and organizing key information from resumes have increased. Traditional resume information extraction methods mainly rely on optical character recognition (OCR) technology, but the recognition accuracy of OCR is often low when dealing with resumes with complex layouts or mixed text and images. In recent years, advances in deep learning techniques, especially pretrained language models such as BERT and GPT, have made significant progress in language processing tasks. However, these models have not been specifically optimized for resume data, so their extraction performance in specific fields may not be satisfactory. This article proposes an efficient multimodal resume recognition model that improves information extraction by preprocessing resume layout formats, using YOLO layout analysis, and combining the VI-LayoutXLM model and paddle OCR technology. The experimental results show that the optimized model performs well in resume information extraction tasks, with an F1 score increasing from 0.9 to 0.92.