| 摘要 |
[Objectives]To develop an assessment model for reservoir dam turf health based on UAV and AI technologies,to achieve automated and precise assessment of slope protection turf health,and to enhance the refined level of dam safety management.[Methods]An assessment indicator system covering four dimensions(coverage rate,weed index,degradation rate,and uniformity)was established.High-resolution images were captured by UAV,and pixel-level land cover classification and recognition were performed using the U-Net semantic segmentation model.The Gray-Level Co-occurrence Matrix(GLCM)was introduced to quantify turf uniformity.The Analytic Hierarchy Process(AHP)was employed to determine indictor weights,and a comprehensive assessment model of the Dam Turf Health Index(DTHI)along with five-level health grading criteria was constructed.[Results]On the test set,the U-Net model achieved a mean pixel accuracy of 97.55%,a Mean Intersection Over Union(MIOU)of 91.13%,and an F 1-score of 0.9545,enabling accurate identification of target turf,weeds,bare soil,and degraded areas.The DTHI model converted multi-dimensional indicators into intuitive health grades(Excellent,Good,Medium,Poor,and Inferior),providing decision support for graded maintenance management.[Conclusions]This assessment model can effectively improve the efficiency and objectivity of turf health monitoring,reduce the safety risks associated with manual inspection,and provide a feasible technical solution for intelligent management and maintenance of dam turf. |