@misc{Gasparyan_Hayk._A._A,
 author={Gasparyan, Hayk. A.},
 howpublished={online},
 publisher={Изд-во НАН РА},
 abstract={Salient object detection (SOD) aims to identify the most visually prominent objects in images, crucial for tasks like image segmentation, visual tracking, autonomous navigation, and photo cropping. While SOD has been extensively studied in natural scene RGB images, detecting salient objects in remote sensing images remains underexplored due to varying spatial resolutions and complex scenes. This paper presents a novel framework for SOD called Multispectral Decomposition Network (MSD-Net) in remote sensing 3-band RGB images, combining Multispectral Decomposition and Frequency-based Saliency detection. The framework includes three key steps: (i) Multispectral Decomposition: Decomposing a 3-band RGB image into 32 multispectral bands to enhance feature capture across spectral domains; (ii) Synthetic RGB Reconstruction: Using a new entropy-based measure to select the most informative bands in salient regions by analyzing frequency domain and constructing synthetic RGB image; and (iii) Saliency Fusion and Object Detection: training a segmentation network on the fusion of synthetic RGB image and input image for improved accuracy. Comprehensive evaluations of public datasets demonstrate that the proposed method performs better than state-of-the-art (SOTA) models and offers a robust solution for detecting salient objects in complex remote sensing images by integrating multispectral and frequency-based techniques.},
 type={Հոդված},
 title={A Multispectral Decomposition and Frequency-Based Framework for Salient Object Detection in Remote Sensing},
 keywords={Mathematical cybernetics, Computer science},
}