Global-Local Detail Guided Transformer for Sea Ice Recognition in Optical Remote Sensing Images

Zhanchao Huang,Wenjun Hong,Hua Su

2024-05-22

Abstract:The recognition of sea ice is of great significance for reflecting climate change and ensuring the safety of ship navigation. Recently, many deep learning based methods have been proposed and applied to segment and recognize sea ice regions. However, the diverse scales of sea ice areas, the zigzag and fine edge contours, and the difficulty in distinguishing different types of sea ice pose challenges to existing sea ice recognition models. In this paper, a Global-Local Detail Guided Transformer (GDGT) method is proposed for sea ice recognition in optical remote sensing images. In GDGT, a global-local feature fusiont mechanism is designed to fuse global structural correlation features and local spatial detail features. Furthermore, a detail-guided decoder is developed to retain more high-resolution detail information during feature reconstruction for improving the performance of sea ice recognition. Experiments on the produced sea ice dataset demonstrated the effectiveness and advancement of GDGT.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges in sea - ice identification, specifically including:

1. **Diversity of Sea - ice Region Scales**: Sea ice has a wide coverage and huge scale differences, which makes existing methods difficult to accurately segment sea - ice regions of different scales, especially those floating ice regions with tortuous and fragmented edge contours.
2. **Fine Identification of Sea - ice Types**: Existing methods have difficulties in identifying different types of sea - ice at a fine - grained level, especially in thin - ice areas where the boundary with seawater is not obvious.
3. **Lack of Spatial Details**: Although the Transformer structure has a large receptive field and can learn non - local related features, it lacks the capture of spatial details, resulting in difficulties in segmenting small sea - ice and identifying different types of sea - ice.
4. **Texture Interference**: The texture of the ice surface and ice on land can also interfere with the segmentation and identification of sea - ice.

To address the above challenges, the paper proposes a global - local detail - guided Transformer method (GDGT), aiming to improve the accuracy of sea - ice identification by fusing global structure - related features and local spatial detail features. Specific contributions include:

1. **Global - Local Feature Fusion Mechanism (GLFF)**: The GLFF module is designed to fuse global structure - related features and local spatial detail features in the decoder to more accurately segment multi - scale sea - ice regions.
2. **Detail - guided Decoder (DGD)**: The DGD module is developed to use the discrete wavelet transform (DWT) to retain high - resolution detail information, thereby improving the performance of small sea - ice identification in the feature reconstruction process.

Through experimental verification, GDGT performs better than existing methods on sea - ice datasets, especially in multi - scale sea - ice segmentation and small sea - ice identification.