Hyperspectral and Multi-source Heterogeneous Data Fusion Classification Based on Multiscale Multi-source Interaction Attention Network

With the development of remote sensing technology, the performance varies among different sensors, and multi-source data fusion is widely concerned. However, existing methods often fall short in extracting rich features from multi-source data and tend to overlook the mutual complementarity between different data sources. To address this issue, this paper proposes a novel multiscale multi-source attention fusion network. The proposed approach introduces a squeeze-and-excitation convolution module to extract multiscale information and enrich the feature space. An attention mechanism is employed to enhance the importance of relevant spectral features. Additionally, a multi-source interaction attention module is designed to explore the semantic interactions between multiple sources of remote sensing data, thereby improving classification performance. Numerous experiments on two multi-source datasets demonstrate the effectiveness of the suggested approach.