Jiasheng Wu

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.

Canopy fractional vegetation cover (FVCc) is a critical indicator for evaluating the effectiveness of ecological restoration, and its accurate estimation provides valuable data for regional ecological management. In this study, Sentinel-2 and MODIS data were integrated to develop an angular relationship model for MODIS reflectance, which was then used to estimate Sentinel-2 reflectance at a 45° viewing angle. Background reflectance at a 10 m spatial resolution was derived using a four-scale model, and total and shrub-herb fractional vegetation cover were estimated using a pixel dichotomy model. Finally, an empirical model tailored to the characteristics of the study area was developed to retrieve FVCc. Cross-validation results demonstrated that the multi-angle retrieval method proposed in this study achieved higher accuracy than the single-angle approach. The spatial distribution of FVCc in Changting County is characterized by higher values in peripheral areas and lower values in the central region. Temporal transitions among fractional vegetation cover classes were predominantly upward, indicating an overall trend of continuous improvement. These findings provide important technical support and a scientific basis for estimating and monitoring dynamic changes in forest canopy fractional vegetation cover.

Understory shrub–grass coverage is a key indicator of forest ecosystem structure and function, and its accurate retrieval via remote sensing is essential for regional ecological assessments. To address the critical limitation in existing multi-angle remote sensing inversion methods: high-resolution images lack angular information while multi-angle datasets suffer from low spatial resolution, making it difficult to achieve large-scale and fine-grained inversion of understory shrub–grass coverage. Here, we developed an inversion method for estimating understory shrub–grass coverage by integrating multi-angle Moderate Resolution Imaging Spectroradiometer data with high-resolution Sentinel-2 imagery to produce 10 m resolution coverage maps; we then used this method to analyze spatiotemporal changes in Changting County from 2016 to 2025. The results demonstrated that the method achieved high accuracy (R² = 0.8418, RMSE = 0.07), meeting the requirements for quantitative understory shrub–grass coverage estimation. Understory shrub–grass coverage exhibited a concentric decreasing pattern from the surrounding mountainous areas toward the central plain, with high-coverage zones concentrated primarily in the west, southwest, and south. Over the 2016–2025 period, understory shrub–grass coverage displayed a fluctuating upward trend: approximately 60% of the study area showed improvement, while 37.73% experienced slight degradation. The change persistence was dominated by positive trends, with an area proportion of 54.14%, which was close to that of the anti-persistent type (44.87%). This study provides technical support for the high-resolution inversion of understory vegetation structure.

Severe erosion persists in the red soil region of southern China despite dense vegetation. Stratified vegetation volume (SVV), which integrates horizontal and vertical vegetation density, better captures understory structure than fractional cover. Here, we established and surveyed 75 forest stands (10 m × 10 m) spanning an erosion-intensity gradient in Changting County, Fujian Province, China. Within each stand, soil was sampled by depth (0–20 cm), and living and dead vegetation volumes in the canopy, shrub–herb, and litter layers were quantified to derive SVV. Relative to slightly eroded soils, moderate and severe erosion reduced the soil water content by 38–41% and soil organic matter by 19–34%, while increasing bulk density by 25–30% and pH by 6–8%. Severe erosion increased the sand content by 20–31% and decreased the gravel content by ≤15%. SVV declined sharply with erosion, with the largest loss in the shrub–herb layer (66–97%). Erosion was most strongly associated with shrub–herb SVV, soil water content, organic matter, and bulk density (r = 0.5–0.6, p < 0.05). The shrub–herb layer was the key component resisting surface erosion. Overall, understory degradation accelerates erosion and soil coarsening, reinforcing a constrained vegetation–soil system; restoring native shrubs and grasses, coupled with targeted canopy thinning, may improve soil and water conservation.

Spoil heaps have become a major source of anthropogenic soil erosion, but the hydrological responses and erosion mechanisms of in situ slopes under rainstorms remain poorly understood. We performed simulated rainfall experiments at real estate (Site A), railway (Site B), and railway station (Site C) construction sites, as well as spoil sites (Site D) in China’s Yangtze River Delta. Rainfall parameters, surface runoff, interflow, vertical soil moisture profiles, and sediment yield were monitored: (1) Hydrological responses differed significantly across the sites due to soil structure complexity; stable erosion after the first rainfall event was not achieved at any site except Site C. Soil erosion was the strongest at Site C, followed by Sites D, B, and A. After the second rainfall event, erosion was stable, increasing, and decreasing at Sites A, B and C, and D, respectively. (2) Runoff and the soil loss rate were positively correlated (R² > 0.7), and the slopes of the fitted regression lines were highest for Sites B and C, followed by Sites D and A. (3) Soil erodibility values based on field data were 0.0029, 0.1164, 0.1974, and 0.0989 t·hm²·h·hm⁻²·MJ⁻¹·mm⁻¹ for Sites A, B, C, and D, respectively. (4) The soil bulk density, gravel content, and silt content were key factors contributing to the severe erosion of field spoil heaps. Spoil heaps from different project types exhibited distinct hydrological and erosional behaviors, which necessitates targeted mitigation strategies to reduce severe erosion and landslide risks.