The classification task of remote sensing images has been a challenge since the development of satellite technology. Due to its powerful learning strength, deep learning techniques, such as Convolutional Neural Network(CNN), outperforms traditional classification methods. The popularity of attention mechanisms gives birth to channel attention networks like SENet and spatial attention networks. Nevertheless, the integration of SE block and CNN is complicated, and it applies dimension reduction, which can have a negative influence on accuracy. This paper proposes a method that combines the Efficient Channel Attention(ECA) module with a ResNet model, which improves the classification performance by avoiding reducing dimensions and employing deep residual structure. Validation experiments on the AID dataset demonstrated that the ECA module achieves high accuracy.
Remote Sensing Image Scene Classification Based on ECA Attention Mechanism Convolutional Neural Network
12.10.2022
1446249 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch