Damage assessment method based on multi-modal data fusion

The invention discloses a damage assessment method based on multi-modal data fusion, and the method comprises the steps: obtaining images before and after a target region is damaged, and carrying out the attribute information expansion of each image, and obtaining attribute text information; according to the damage assessment method based on multi-modal data fusion, the number of image pairs obtained before and after damage is set to a first network structure in a damage assessment network model, and forward reasoning is carried out through the first network structure, a second network structure, a first full connection layer and a feature decoding structure to obtain a damage assessment report. Deep-level multi-modal data fusion is realized to judge a damage result, and a more accurate evaluation result can be obtained; meanwhile, the first network structure is obtained by improving based on a ViT-G/14 model (a visual large model), the feature decoding module is based on an LLaMA-7B large language model, dam