The presence of Arctic sea ice increases navigation risks and introduces numerous instability factors, representing a major safety hazard for Arctic Route. This paper addresses the frequent sea ice hazards in Arctic Route by proposing a deep learning-based rapid SAR sea ice image segmentation algorithm, named Fast-SISENet (Fast-Sea Ice Segment Network), to achieve real-time acquisition and processing of Arctic ice conditions. This method improves upon the DeeplabV3+ backbone network by modifying the original ASPP module into a parallel dual-branch structure, allowing it to process features at different scales simultaneously. This modification reduces the number of data transmissions through the network and enhances computation speed. Additionally, a specialized adaptive feature fusion module and a lightweight channel attention mechanism are designed to integrate local and global features, providing more comprehensive feature information to the decoder. Compared to other mainstream segmentation methods, this approach achieves the fastest inference speed while maintaining high classification accuracy. The inference speed of this method can be integrated with SAR quick view systems to enable real-time acquisition and classification of SAR sea ice images, offering a reliable reference for real-time path planning and emergency response to sea ice hazards in Arctic Route.