Year: 2024
Type: conference-paper
Abstract: Remote sensing image captioning is a task that aims to automatically generate descriptive captions for remote sensing images, helping to interpret and summarize visual content.Although latest works are able to reflect fundermental information of remoter sensing image captions, the object count information still absence.In this paper, we propose a new framework to incorporate object count into remote sensing image captions.The framework we propose contains three models: Vis-Transformer, Cap-Transformer and numeral caption generating.The framework solves the problem of lacking object count information in remote sensing image captions.We set up the experiment to evaluate the performance of our proposed framework on several widely-used remote sensing image captioning datasets.The experiment aims to analyze how effectively object count information is incorporated into the generated captions, as well as the overall caption quality.We compare our method with state-of-the-art models across multiple metrics, including BLEU, ROUGE, CIDEr, SPICE, METEOR, and Num-Meteor.We applied Transformer to the task of integrating the number of objects into captions.The experimental results indicate that our proposed framework significantly improves the accuracy of remote sensing image captions by successfully incorporating object count information and compared to our previous research, we have achieved better results on the RSICD dataset.
Authors: Zihao Ni, Zhaoyun Zong, Yinghao Xu, Peng Ren
Institution: China University of Petroleum, East China
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4412100034