Image-Text Correlation Based Remote Sensing Image Retrieval
Prem Shanker Yadav, Sachin Dube · 2023
Remote sensing (RS) photos have greatly increased recently as a result of satellite technology developments. Designing an accurate image retrieval model to obtain the most pertinent pictures depending on the query is therefore one of the crucial research areas. However, the complex content of RS pictures cannot be well described by the visual descriptors. An innovative approach for picture retrieval based on image captions is developed to address this issue. The objective is to create textual visuals that clearly define the relationships between items and their titles. In order to get the best match between image captions and the search text, hybrid techniques such as frequency-inverse document frequency (TF-IDF) and bag of words are used. Convolutional Neural Network (CNN), Long Short Term Memory (LSTM), and Vision Transformer are used for image captioning and applied on Remote Sensing Image Captioning and Sydney-Caption dataset. Vision Transformer is out performing compared to LSTM+CNN(VGG16) and LSTM+CNN(ResNet50).