Abstract
Due to the extremely complex composition of remote sensing scenes, REmote Sensing Image Scene Classification (RESISC) is still a challenging task. To further improve classification accuracy, this article introduces a deep-learning detector into RESISC and proposes to classify remote sensing images according to the detected class-specific signature objects. Inspired by the classification procedure of human vision system, we design a classification framework that utilizes class-specific signature objects of scene classes to guide scene classification. When performing image classification, the proposed framework first adopts a deep-learning classifier to create an initial judgment of the scene class for an image and then determines the scene class based on the class-specific signature objects detected from the image. The proposed method can compete with the state-of-the-art methods on three RESISC benchmark datasets, including NWPU-RESISC45, AID, and OPTIMAL-31.
Highlights
W ITH the rapid development of remote sensing technology, the automation level of remote sensing image interpretation is constantly improving
We introduce a deep-learning detector into scene classification to detect class-specific signature objects in remote sensing images
To further improve scene classification accuracy, we investigate utilizing class-specific signature objects to guide scene classification and develop an object-guided method based on using the off-theshelf deep convolutional neural network (CNN) models
Summary
W ITH the rapid development of remote sensing technology, the automation level of remote sensing image interpretation is constantly improving. It is worth noticing that scene classification methods with human-engineering features perform well on some scenes with uniform structures and spatial arrangements, but it is difficult for them to depict the high diversity and the nonhomogeneous spatial distributions in remote sensing images [12]. As reported in [16]–[19], since large-scale datasets have rich image variations, deep-learning methods that directly fine-tune the existing CNNs on RESISC dataset suffered an accuracy degradation. This is mainly because they only utilize the feature from the last layer of CNN to classify images and ignore the features from different hierarchical layers of CNN.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callMore From: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
