The stability of bank slopes in pumped storage power stations is crucial, particularly in regions where frequent water level fluctuations occur. This study aims to investigate the degradation mechanism of bank slope under such fluctuating conditions, focusing on granite residual soil from the pumped storage power stations in Fujian, China. To explore the effects of drying-wetting cycles and matric suction on soil shear strength, drying and wetting cycles were conducted with unsaturated triaxial shear tests. The results revealed that the shear parameter strengthening effect occurs when the matric suction increases from 50 kPa to 200 kPa. Moreover, during the first five drying-wetting cycles, soil shear strength decreased sharply, with cohesion and internal friction angle decreasing by approximately 15.4% and 11.2%, respectively. This degradation trend stabilized in the later cycles. Scanning Electron Microscopy (SEM) analysis of the soil microstructure showed an evolution from a dense structure to a penetrating cavity during the cycles. This change reflects that the strength degradation characteristics of granite residual soils are controlled by the synergistic effects of structural and frictional mechanisms, manifesting as initial degradation followed by stabilization. Additionally, by fitting the nonlinear characteristics of the experimental data, shear strength evolution functions for matric suction and drying-wetting cycles were established, revealing the effect of these factors on strength degradation. These findings provide a theoretical basis for the stability analysis of bank slopes in pumped storage power stations, offering insights into soil behavior under fluctuating water levels.

Surface cracks are critical indicators of deterioration in flood-control infrastructure, yet automated detection from inspection imagery remains challenging due to complex backgrounds, elongated geometries, and variations in apparent scale. This study aims to develop a lightweight detector for accurate dike crack detection while maintaining computational efficiency. A hybrid CNN–Transformer detector, termed YOLO-Defender, was built on YOLOv8n. The proposed framework incorporates a C2F-FTB block in the backbone for local–global feature representation, a BiFF-PAN neck for enhanced multi-scale feature fusion, and coordinate attention for improved spatial feature localization. The model was evaluated on an in-house dike crack dataset containing 1217 images collected from flood-defense structures and related crack imaging scenarios using a unified training and testing protocol. YOLO-Defender achieved 89.6% mAP@0.5 and 69.2% mAP@0.5:0.95, improving YOLOv8n by 3.0 and 5.1 percentage points, respectively. Compared with YOLOv8n, the proposed model reduced parameters and GFLOPs by 41.2% and 18.5%, respectively, while achieving an inference speed of 153 FPS on a desktop GPU. The results indicate that task-oriented architectural design can improve crack detection accuracy and localization performance while preserving lightweight characteristics. These findings support efficient screening of hydraulic infrastructure inspection imagery.