Image captioning aims to describe the content of an image through a textual description including attributes and relationships of detected objects. In remote sensing (RS) community IC is becoming an interesting solution for the study of very high spatial resolution (VHR) images that are characterized by high-level information detail. RSIC systems are developed in a supervised way where annotated samples are needed to train the system. However, obtaining a large amount of annotated samples is time-consuming and costly. To address this issue, in this work we propose an active learning solution to select the most important samples to label and include in the training set with the aim of maintaining the system's accuracy as high as possible while using a few amount of training samples. The most important samples are selected based on decision uncertainty and diversity criteria. Experimental results show that the proposed active learning solution represents a good trade-off between the number of training samples and the accuracy of the system.

A New Active Image Captioning Fusion Strategy / Hoxha, G.; Munari, A.; Melgani, F.. - ELETTRONICO. - (2022), pp. 1-4. (Intervento presentato al convegno 2022 IEEE Mediterranean and Middle-East Geoscience and Remote Sensing Symposium, M2GARSS 2022 tenutosi a Virtual Conference nel 7-9, March 2022) [10.1109/M2GARSS52314.2022.9840136].

A New Active Image Captioning Fusion Strategy

Hoxha G.;Melgani F.
2022-01-01

Abstract

Image captioning aims to describe the content of an image through a textual description including attributes and relationships of detected objects. In remote sensing (RS) community IC is becoming an interesting solution for the study of very high spatial resolution (VHR) images that are characterized by high-level information detail. RSIC systems are developed in a supervised way where annotated samples are needed to train the system. However, obtaining a large amount of annotated samples is time-consuming and costly. To address this issue, in this work we propose an active learning solution to select the most important samples to label and include in the training set with the aim of maintaining the system's accuracy as high as possible while using a few amount of training samples. The most important samples are selected based on decision uncertainty and diversity criteria. Experimental results show that the proposed active learning solution represents a good trade-off between the number of training samples and the accuracy of the system.
2022
2022 IEEE Mediterranean and Middle-East Geoscience and Remote Sensing Symposium, M2GARSS 2022 - Proceedings
Piscataway, NJ USA
Institute of Electrical and Electronics Engineers Inc.
978-1-6654-2795-1
Hoxha, G.; Munari, A.; Melgani, F.
A New Active Image Captioning Fusion Strategy / Hoxha, G.; Munari, A.; Melgani, F.. - ELETTRONICO. - (2022), pp. 1-4. (Intervento presentato al convegno 2022 IEEE Mediterranean and Middle-East Geoscience and Remote Sensing Symposium, M2GARSS 2022 tenutosi a Virtual Conference nel 7-9, March 2022) [10.1109/M2GARSS52314.2022.9840136].
File in questo prodotto:
File Dimensione Formato  
M2GARSS-Active Captioning.pdf

Solo gestori archivio

Tipologia: Versione editoriale (Publisher’s layout)
Licenza: Tutti i diritti riservati (All rights reserved)
Dimensione 657.38 kB
Formato Adobe PDF
657.38 kB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11572/373012
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 2
  • ???jsp.display-item.citation.isi??? 2
social impact