摘要详情
91 / 2024-08-15 13:54:36
基于多时相和地表覆盖信息的MODIS地表反射率数据重建方法
缺失重建,MODIS地表反射率,多时相,地表覆盖
全文待审
摘 要：MODIS地表反射率数据被广泛应用于大区域和全球尺度的变化检测中，但由于云层覆盖等干扰导致数据存在多时相、大面积的缺失，限制了其进一步应用。因此，针对多时相缺失条件下的MODIS地表反射率数据重建问题，本文提出了一种基于多时相和地表覆盖信息的重建方法。该方法首先通过多尺度特征提取融合模块提取特征信息，再通过残差模块和空间-通道注意力模块重建缺失信息。在自制数据集MOD1和MOD2上进行重建精度评价，实验结果表明本文提出的MTL-RN方法能够有效地重建缺失区域，与PSTCR、STS-CNN、TLC-ResNet方法相比定量重建精度最优；峰值信噪比PSNR和光谱角制图SAM分别为44.6981 dB和1.5049°，能够有效地重建地物的空间细节和光谱信息。
Abstract : MODIS surface reflectance data are widely used in change detection at large regional and global scales, but the multi-temporal and large-area missing data due to cloud cover and other disturbances limit its further application. Therefore, to address the problem of reconstructing MODIS surface reflectance data under multi-temporal missing conditions, this paper proposes a reconstruction method based on multi-temporal and landcover information.The method firstly extracts feature information through the multiscale feature extraction fusion module, and then reconstructs the missing information through the residual module and the spatial-channel attention module. The reconstruction accuracy is evaluated on the homemade datasets MOD1 and MOD2, and the experimental results show that the MTL-RN method proposed in this paper can effectively reconstruct the missing regions, and has the optimal quantitative reconstruction accuracy compared with PSTCR, STS-CNN, and TLC-ResNet methods; The PSNR and SAM are 44.6981 dB and 1.5049°, which can effectively reconstruct the spatial details and spectral information of features.
Abstract : MODIS surface reflectance data are widely used in change detection at large regional and global scales, but the multi-temporal and large-area missing data due to cloud cover and other disturbances limit its further application. Therefore, to address the problem of reconstructing MODIS surface reflectance data under multi-temporal missing conditions, this paper proposes a reconstruction method based on multi-temporal and landcover information.The method firstly extracts feature information through the multiscale feature extraction fusion module, and then reconstructs the missing information through the residual module and the spatial-channel attention module. The reconstruction accuracy is evaluated on the homemade datasets MOD1 and MOD2, and the experimental results show that the MTL-RN method proposed in this paper can effectively reconstruct the missing regions, and has the optimal quantitative reconstruction accuracy compared with PSTCR, STS-CNN, and TLC-ResNet methods; The PSNR and SAM are 44.6981 dB and 1.5049°, which can effectively reconstruct the spatial details and spectral information of features.
重要日期
-
会议日期
09月20日
2024
至09月22日
2024
-
08月30日 2024
初稿截稿日期
-
09月22日 2024
注册截止日期
主办单位
山东省人民政府
中国电子学会
中国电子学会
承办单位
中国科学院学部
中国科学院空天信创新研究所息
复旦大学
中国科学院空天信创新研究所息
复旦大学
联系方式
- 李若明
- li******@aircas.ac.cn
- 177********
- 赵丽云
- zh******@aircas.ac.cn
- 186********
- 高华
- ra******@aircas.ac.cn
- 010*********
