SD-RSIC: Summarization-Driven Deep Remote Sensing Image Captioning

Deep neural networks (DNNs) have been recently found popular for image captioning problems in remote sensing (RS). Existing DNN-based approaches rely on the availability of a training set made up high number RS images with their captions. However, captions may contain redundant information (they can be repetitive or semantically similar to each other), resulting deficiency while learning mapping from domain language domain. To overcome this limitation, article, we present novel summarization-driven (SD-RSIC) approach. The proposed approach consists three main steps. first step obtains standard by jointly exploiting convolutional (CNNs) long short-term memory (LSTM) networks. second step, unlike existing methods, summarizes ground-truth into single caption sequence and eliminates redundancy set. third automatically defines adaptive weights associated combine summarized based semantic content image. This is achieved weighting strategy defined context LSTM Experimental results obtained RSCID, UCM-Captions, Sydney-Captions data sets show effectiveness compared state-of-the-art approaches. code publicly available at https://gitlab.tubit.tu-berlin.de/rsim/SD-RSIC .