Search

Precise and fast extraction of sea ice leads is the foundation for polar research and ship navigation. The accuracy of traditional methods for sea ice lead extraction is limited, and the efficiency of deep learning methods is difficult to guarantee. Besides, the tedious preprocessing steps complicate the application of existing methods. In this article, we proposed a lightweight semantic segmentation model based on the U-Net framework for sea ice lead extraction, which introduced lightweight blocks and a feature branch. Lightweight blocks took the place of the convolutional layers in U-Net to reduce the network parameters and increase operational speed. With the input of the contrast feature of horizontal-vertical (HV) polarization, the feature branch was used to improve the extraction precision and robustness. Besides, the combination of the lightweight blocks, feature branch and U-Net framework was beneficial to resist preprocessing. In the experiments, the performance of the proposed method on non-preprocessed Sentinel-1 dataset was better than that of the classical semantic segmentation method on preprocessed Sentinel-1 dataset in floating-point operations (FLOPs), parameters, frames per second (FPS), and accuracy evaluation. The results indicate that the proposed network is effective and lightweight for sea ice lead extraction from non-preprocessed data. [ABSTRACT FROM AUTHOR]

Autoencoders (AEs) are widely utilized in hyperspectral unmixing (HU) as an unsupervised learning model. In particular, convolutional AE networks are popular for processing multidimensional hyperspectral features. Nonetheless, the traditional convolutional AE network’s receptive field is constrained in the unmixing task, and establishing the connection between the local spatial neighborhood and the local spectrum fails to improve unmixing performance significantly. To address these limitations, a bilateral global attention network based on both spatial and spectral information is proposed. It enables the network to obtain respective feature dependencies in the two dimensions and achieve optimal fusion of both features. The network comprises two information extraction branches. The spatial information extraction branch uses the Swin Transformer block to acquire the global spatial attention of the overall image, while the spectral information extraction branch designates a simplified spectral channel attention mechanism to gain spectral attention weight maps. The network’s efficacy is demonstrated through a comparative study using a synthetic dataset and two real datasets. The code of this work is available at https://github.com/UPCGIT/SSABN. [ABSTRACT FROM AUTHOR]