Method and system for intelligent generation of cultural tourism video scene description text

Abstract

The present invention discloses a method and a system for intelligent generation of cultural tourism video scene description text, comprising: extracting feature vectors through video frame sequences, processing image data using a convolutional neural network to obtain scene element classification results; querying a preset template knowledge base based on the scene element classification results, screening description frameworks matching the classification using a template matching algorithm to determine a basic narrative structure; if the basic narrative structure does not fully correspond to the unique details of the video, analyzing specific video frames through a keypoint detection algorithm to obtain a differentiated element list; filling placeholders in the basic narrative structure using the obtained differentiated element list to generate a preliminary description text draft; performing coherence verification based on the preliminary description text draft, evaluating sentence fluency using a natural language processing model to determine a final narrative language version. The present invention achieves an organic fusion of universal templates and personalized details, significantly improving the accuracy, coherence, and naturalness of video description text.