Transactions-submission v3.docx (11.86 MB)

Multi-Scale Feature Reconstruction and Inter-Class Attention Weighting for Land Cover Classification

Download (11.86 MB)
posted on 2023-07-25, 02:31 authored by Xin WangXin Wang

Land cover classification has the goal to attribute each pixel of high-resolution remote sensing image with planimetric category labels (such as vegetation, building, water etc.). In recent years, many serial learning architectures (features are delivered through a single path such as in ResNet, MobileNet, Segformer etc.) based on Convolutional Neural Networks (CNN) and attention mechanisms have been widely explored in land cover classification. However, high-resolution remote sensing images typically have abundant textual details, variable scales in objects, large intra-class differences and similar inter-class distances, which brings challenges to land cover classification. In this work, we presented two pluggable modules to further boost serial learning architecture: first, to cope with ambiguous boundaries caused by lost details and fragmented segmentation stemmed from scale variances, a combination of channel attention and spatial attention is proposed to reconstruct multi-scale feature; Second, to mitigate the classification error caused by intra-class variance and inter-class correlation, we build feature vectors for each category, and apply a multi-head attention model to capture self-attention dependence among different categories. The experimental results demonstrate that the proposed modules are feasible to existing serial learning architectures and can improve OA by 5.64% on the ISPRS Vaihingen 2D dataset (using ResNet50 as backbone). In addition, compared to other state-of-art models, our method can achieve similar or even better classification results, yet offer superior inference performance 


National Natural Science Foundation of China (Grant No. 61871295); Natural Science Foundation of Hubei Province, China (No. 2022CFB727)


Email Address of Submitting Author

ORCID of Submitting Author

Submitting Author's Institution

School of Geodesy and Geomatics, Wuhan University

Submitting Author's Country

  • China