Abstract
As an advanced technique in remote sensing, hyperspectral target detection (HTD) is widely concerned in civilian and military applications. However, the limitation of prior and mixed pixels phenomenon makes HTD models sensitive to data corruption under various interference from environment. In this work, a novel two-stage detection framework based on adversarial learning is proposed, which extracts spectral features in latent space through background reconstruction under weak supervision. To address the issues of insufficient utilization of both background information and limited prior knowledge, the generative adversarial network (GAN) is applied to estimate background in a weakly supervised manner with target-based constraints and channel-wise attention, which produces the detection proposal in the first stage. Then, a refined result is produced in the second stage, in which the input data consists of the refined data and refined feature map based on previous detection proposal. To provide samples for weakly supervised learning (WSL), the pseudo datasets are produced by a coarse sample selection procedure, which makes full use of limited prior information. Finally, an exponential constrained nonlinear function is adopted to acquire pixel-level prediction via suppressing the background and combining features from different stages. Experiments on real hyperspectral images (HSIs) captured by different sensors at various scenes verify the effectiveness of the proposed framework.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
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.