.png){kind=icon}
.png){kind=icon}
Return
Incorporating difference information into curriculum learning for multitemporal image classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.127070.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multitemporal remote sensing image classification aims at exploiting the available information on one image with a training set (source image, SI) for classifying newly obtained images in the identical region (target image, TI). However, existing studies do not fully account for robust difference information (DI) among multitemporal images. For further exploiting the difference and learning a robust classifier, this paper proposes a difference-based curriculum learning scheme for classifying multitemporal images. In the proposed method, the robust difference information transfer technique is designed to reduce the errors of the unchanged class for improving the accuracy of the transferred labels. Then curriculum learning is utilized to formulate a learning objective with allocating a weight per sample for measuring the reliability. In the process of curriculum learning, a tunable curriculum function incorporating difference information is designed to constrain the curriculum region in an implicit way. Experimental performance over ten datasets (five areas) indicates that our proposed model achieves remarkable classification performance, thus confirming its suitability when applied to remote sensing images.
Keywords:
Curriculum learning
Superpixel segmentation
Multitemporal image classification
Remote sensing
Change detection
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
Organization
No organization information available