A two-period image change detection method based on a visual language large model

Abstract

The application discloses a two-period image change detection method based on a visual language large model, comprising the following steps: acquiring and processing the pre-period image and the post-period image of the same area to obtain an image block set; for the post-period image block, target detection is performed on the post-period image block by using an open vocabulary target detection model according to the input text prompt to obtain a candidate box set; the candidate box is used as a prompt to segment the post-period image block by using a visual segmentation large model to obtain a binary mask of the post-period image; the mask in the post-period image is reversely mapped to the corresponding position of the pre-period image, the features of the pre-period image and the post-period image in the mask area are extracted, the feature similarity is calculated, and if the similarity is lower than a preset threshold, it is determined that the target is a newly added change; all the masks determined as changes are subjected to binarization processing, geographical space splicing and vectorization processing, and a vector change detection result with geographical coordinate information is output. The application solves the detection demand of low cost, high precision and wide adaptation.

Images