Machine Readable Travel Document (MRTD) Quality Inspection System Based on Knowledge Graphs and Multimodal Models

In this paper, we propose the Machine Readable Travel Document (MRTD) Quality Inspection System Based on Knowledge Graphs and Multimodal models which obtains the image to be inspected, production equipment log information, document information, and the corresponding document inspection text prompt for the document to be inspected. Based on the above information, it determines the inspection task for the document to be inspected. It obtains the knowledge graph feature information corresponding to the inspection task, including equipment information, document information, document anomaly information, anomalous solutions, and time information; extract image feature information from the images to be inspected; align the feature information from the knowledge graph, image feature information, and text feature information to obtain feature-aligned embedding features; perform quality inspection on the embedding features based on the inspection task to obtain inspection output data; use a pre-trained large language model, determine the quality inspection results for the document to be inspected based on the best output data. This system can clearly obtain accurate inspection results, determine the causes of document anomalies, and make timely adjustments to the document production process. The model accuracy is 7.3% higher than Qwen VL 2.5 7B-Instruct, thereby improving document production efficiency and quality.