【摘要】 针对目前极化合成孔径雷达(PolSAR)数据处理方法难以有效挖掘极化数据中潜在信息的问题，该文提出了一种新的基于双分支编码器与复合注意力特征融合的极化SAR分类网络模型。联合使用极化SAR幅值数据和极化分解数据避免了极化信息捕捉不充分的问题，使用高效通道注意力模块在通道维度重新校准特征信息权重，并引入坐标空间注意力模块增强了对局部空间特征的捕捉，实现多维特征的深度融合；为了减少下采样过程中的特征损失，使用跳跃连接并重新对齐了不同尺度的融合特征。基于广东肇庆地区的高分三号全极化数据集的实验结果表明，该文提出的算法相较其他主流方法在样本不平衡下的总体表现更好，平均精度达到92.5%,平均交并比为79.4%。更多还原

【Abstract】 In response to the issue that existing methods for processing polarized SAR data lack effective approaches to deeply extract polarization information within the dataset, a novel PolSAR image classification network based on dual-branch encoder and composite attention feature fusion was proposed in this paper. The joint utilization of polarimetric SAR amplitude data and polarimetric decomposition data addressed the issue of insufficient extraction of polarization information. By employing efficient channel attention module, the feature information weights were recalibrated along the channel dimension. Additionally, the coordinate attention module was introduced to enhance the capture of local spatial features, enabling the deep fusion of multidimensional features. To reduce feature loss during the downsampling process, skip connections were employed and the fused features at different scales were realigned. Experimental results based on the fully polarized SAR dataset from Gaofen-3,covering the Guangdong Zhaoqing region, demonstrated that the proposed method outperforms other conventional methods, achieving an overall accuracy of 92.5% and an average intersection ratio of 79.4%.更多还原