Mozi Gao

River ice semantic segmentation is a crucial task that provides essential information for hydrological monitoring and infrastructure protection in cold regions. Previous works mainly focus on global long-range dependency modeling or local feature extraction, while the balance between computational efficiency and fine irregular-boundary preservation is often neglected. In this paper, we propose IceRWKV, an efficient semantic segmentation network for river ice based on the Receptance Weighted Key Value (RWKV). First, the RWKV sequence model is introduced into this task to break the quadratic complexity bottleneck, achieving high-precision global–local feature aggregation with low computation cost. Then, a novel Geometry-Direction Co-sensing Module (GDCM) is adopted to fit irregular ice contours and suppress background noise through an adaptive geometric correction and polarization feature-refinement strategy. Furthermore, Haar wavelet downsampling (HWD) is utilized to replace traditional downsampling operations, effectively mitigating feature aliasing and preserving high-frequency details. We conduct extensive experiments on the NWPU_YRCC_EX, NWPU_YRCC2, and Alberta River Ice Segmentation datasets. Comprehensive experimental results demonstrate that IceRWKV achieves state-of-the-art (SOTA) performance against 10 competing methods. Specifically, on the NWPU_YRCC_EX dataset, our method achieves a Mean Intersection over Union (mIoU) of 93.41% and an inference speed of 37.19 Frames Per Second (FPS) on NWPU_YRCC_EX, demonstrating a favorable trade-off between segmentation accuracy and computational efficiency.

Accurate crack segmentation based on optical sensor imagery is of paramount importance for infrastructure health monitoring and disaster early warning. However, in complex natural scenes, cracks typically exhibit characteristics such as being thin and elongated, multi-branched, and having low contrast. Existing segmentation models struggle to balance low computational overhead with the simultaneous modeling of global topological continuity and the precise extraction of local details. To overcome the aforementioned limitations, we introduce a Structure-Aware Global–Local Mamba (SGL-Mamba). Firstly, the SGL-Mamba Block is designed in the encoding stage, which significantly enhances the joint modeling capacity for continuous topological structures and edge textures through the synergy of a parallel directional scanning mechanism and a local perception branch. Secondly, we design a High–Low Frequency Separation Enhancement (HLFSE) module to reconstruct the skip connections. This module leverages frequency decoupling to adaptively suppress high-frequency background noise and alleviate the semantic gap. Finally, in the decoding stage, the Deformable Large Kernel Attention (D-LKA) is integrated, utilizing a dynamic spatial receptive field to precisely adapt to irregular crack orientations. Extensive experiments on three public datasets (Crack500, DeepCrack, and CrackMap) demonstrate that SGL-Mamba outperforms other state-of-the-art (SOTA) methods, achieving an F1 score (F1) of 0.7982 ± 0.0038 and an mIoU of 0.8011 ± 0.0036 on the Crack500 dataset. While ensuring lightweight architecture and high computational efficiency, the proposed method provides an effective and practical solution for automatic crack detection.