【摘要】 在分析不同覆冰类型视觉形态特征基础上，提出了一种基于双边特征融合语义分割的覆冰类型精细辨识方法。首先，通过多尺度语义分割网络提取覆冰图像特征，并利用双边特征融合模块同时捕获覆冰图像的语义特征和空间特征；随后，引入基于级联融合策略的深度聚合池化模块，以增强网络对全局特征的感知能力，进而建立基于预设颜色像素构成分析的覆冰类型决策机制，实现类型判定；最后，利用实际输电线路覆冰图像进行实例验证。试验结果表明，所提方法对于雨凇、雾凇、雪凇等典型输电线路覆冰类型具有较高的辨识精度，对覆冰重量计算及严重程度评估具有一定的理论支撑及工程应用价值。更多还原

【Abstract】 Based on an analysis of the visual morphological characteristics of different ice coating types,a fine identification method using bilateral feature-fused semantic segmentation is proposed. First,a multi-scale semantic segmentation network is employed to extract features from ice-covered images,and a bilateral feature fusion module is used to capture both semantic and spatial features simultaneously. Subsequently,a deep aggregated pooling module based on a cascaded fusion strategy is introduced to enhance the network’s ability to perceive global features. A decision mechanism for ice type identification,based on the analysis of predefined color pixel composition,is then established to achieve type determination. Finally,actual ice-covered transmission line images are used for experimental validation. Test results demonstrate that the proposed method achieves high identification accuracy for typical transmission line ice types such as glaze,rime,and snow accretion. It provides theoretical support and practical engineering value for ice weight calculation and severity assessment.更多还原